buzzert/Sybil-2
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

Compare commits

base: buzzert:master
Branches Tags
buzzert:master buzzert:codex/system-prompt-and-settings buzzert:wip/spacer-scroll buzzert:ios-pull-to-refresh buzzert:codex/ios-right-align-user-bubbles buzzert:codex/chat-with-search-results buzzert:codex/ios-software-keyboard-send buzzert:codex/ios-no-autofocus-after-send
buzzert:v1.11 buzzert:v1.2.0 buzzert:v1.10
..
compare: buzzert:codex/system-prompt-and-settings
Branches Tags
buzzert:master buzzert:codex/system-prompt-and-settings buzzert:wip/spacer-scroll buzzert:ios-pull-to-refresh buzzert:codex/ios-right-align-user-bubbles buzzert:codex/chat-with-search-results buzzert:codex/ios-software-keyboard-send buzzert:codex/ios-no-autofocus-after-send
buzzert:v1.11 buzzert:v1.2.0 buzzert:v1.10

2 Commits
master ... codex/syst

Author SHA1 Message Date
Agent
f3bb8503aa Add per-chat settings UI in web app for additional system prompt and tool checkboxes 2026-05-24 22:04:05 +00:00
Agent
93e34d086f Augment system prompt with date and user location (default SF) 2026-05-24 21:59:38 +00:00
61 changed files with 1554 additions and 5304 deletions
Expand all files Collapse all files

5
dist/default.conf vendored
View File

@@ -17,11 +17,6 @@ server {
proxy_set_header X-Forwarded-Proto $scheme;
}
location = /manifest.webmanifest {
default_type application/manifest+json;
try_files $uri =404;
}
location / {
try_files $uri $uri/ /index.html;
}

95
docs/api/rest.md
View File

@@ -42,23 +42,6 @@ Chat upload limits:
- `hermes-agent` is included only when `HERMES_AGENT_API_KEY` is configured. Set it to Hermes `API_SERVER_KEY`, or any non-empty value if that local server does not require auth. `HERMES_AGENT_API_BASE_URL` defaults to `http://127.0.0.1:8642/v1`; set `HERMES_AGENT_MODEL` only when you need an additional fallback/override model id.
- The backend loads provider model lists at startup and refreshes them about once every 24 hours. If a later provider refresh fails, the response keeps the last loaded model list for that provider and sets `error` to the latest failure message.
## Chat Tools
### `GET /v1/chat-tools`
- Response:
```json
{
"tools": [
{ "name": "web_search", "description": "..." },
{ "name": "fetch_url", "description": "..." }
]
}
```
Behavior notes:
- Lists Sybil-managed chat tools that can be enabled for `openai`, `anthropic`, and `xai` chat completions.
- Optional tools such as `codex_exec` and `shell_exec` appear only when enabled by server environment configuration.
## Active Runs
### `GET /v1/active-runs`
@@ -89,14 +72,10 @@ Behavior notes:
"title": "optional title",
"createdAt": "2026-02-14T00:00:00.000Z",
"updatedAt": "2026-02-14T00:00:00.000Z",
"starred": true,
"starredAt": "2026-02-14T01:00:00.000Z",
"initiatedProvider": "openai",
"initiatedModel": "gpt-4.1-mini",
"lastUsedProvider": "openai",
"lastUsedModel": "gpt-4.1-mini",
"additionalSystemPrompt": null,
"enabledTools": ["web_search", "fetch_url"]
"lastUsedModel": "gpt-4.1-mini"
},
{
"type": "search",
@@ -104,9 +83,7 @@ Behavior notes:
"title": "optional title",
"query": "search query",
"createdAt": "2026-02-14T00:00:00.000Z",
"updatedAt": "2026-02-14T00:00:00.000Z",
"starred": false,
"starredAt": null
"updatedAt": "2026-02-14T00:00:00.000Z"
}
]
}
@@ -116,7 +93,6 @@ Behavior notes:
- This endpoint is intended for combined conversation/search lists such as sidebars.
- The legacy `GET /v1/chats` and `GET /v1/searches` endpoints remain available for clients that need separate collections.
- The response currently combines up to 100 chats and up to 100 searches.
- `starred`/`starredAt` are backed by membership in a reserved `Project` with id `starred`; future project folders can reuse the same project item model.
## Chats
@@ -130,8 +106,6 @@ Behavior notes:
"title": "optional title",
"provider": "optional openai|anthropic|xai|hermes-agent",
"model": "optional model id",
"additionalSystemPrompt": "optional stored system prompt",
"enabledTools": ["web_search", "fetch_url"],
"messages": [
{
"role": "system|user|assistant|tool",
@@ -147,29 +121,13 @@ Behavior notes:
Behavior notes:
- `provider` and `model` must be supplied together when present.
- When `provider`/`model` are supplied, the new chat initializes `initiatedProvider`/`initiatedModel` and `lastUsedProvider`/`lastUsedModel`.
- `additionalSystemPrompt` is trimmed and stored on the chat; blank values are stored as `null`.
- `enabledTools` stores the enabled Sybil-managed tool names for future chat completions. Unknown tool names are ignored; omitted values default to all currently available tools.
- Optional `messages` are inserted as the initial transcript. Attachment metadata uses the same schema and limits as chat completion messages.
### `PATCH /v1/chats/:chatId`
- Body: any subset of `{ "title": string, "additionalSystemPrompt": string|null, "enabledTools": string[] }`
- Response: `{ "chat": ChatSummary }`
- Blank titles are rejected. The server trims surrounding whitespace before storing the title.
- `additionalSystemPrompt: null` clears the stored prompt. Blank string values are also stored as `null`.
- `enabledTools: []` disables Sybil-managed tools for this chat. Omitted settings are left unchanged.
- Updating chat fields changes the returned chat's `updatedAt`.
- Not found: `404 { "message": "chat not found" }`
### `PATCH /v1/chats/:chatId/star`
- Body: `{ "starred": boolean }`
- Body: `{ "title": string }`
- Response: `{ "chat": ChatSummary }`
- Not found: `404 { "message": "chat not found" }`
Behavior notes:
- Starring adds the chat to the reserved `starred` project and sets `starredAt` to the membership creation time.
- Unstarring removes that membership and returns `starred: false`, `starredAt: null`.
- This does not modify the chat transcript or chat `updatedAt`.
### `POST /v1/chats/title/suggest`
- Body:
```json
@@ -182,8 +140,7 @@ Behavior notes:
Behavior notes:
- If the chat already has a non-empty title, server returns the existing chat unchanged.
- If a title is set while suggestion generation is in flight, server returns the current chat instead of overwriting that title.
- When no title exists at write time, server uses OpenAI `gpt-4.1-mini` to generate a one-line title (up to ~4 words), updates the chat title, and returns the updated chat.
- Server always uses OpenAI `gpt-4.1-mini` to generate a one-line title (up to ~4 words), updates the chat title, and returns the updated chat.
### `DELETE /v1/chats/:chatId`
- Response: `{ "deleted": true }`
@@ -262,8 +219,6 @@ Notes:
]
}
],
"additionalSystemPrompt": "optional one-off system prompt",
"enabledTools": ["web_search", "fetch_url"],
"temperature": 0.2,
"maxTokens": 256
}
@@ -283,24 +238,21 @@ Notes:
Behavior notes:
- If `chatId` is present, server validates chat existence.
- For `chatId` calls, server stores only *new* non-assistant messages from provided history to avoid duplicates.
- `additionalSystemPrompt`, when present directly or loaded from stored chat settings, is prepended to the provider request as a `system` message and is not inserted into the persisted chat transcript by this endpoint.
- `enabledTools` limits Sybil-managed tools for this request. When omitted for a saved chat, the stored chat setting is used; otherwise all available tools are enabled by default. An empty array disables Sybil-managed tools.
- Server persists final assistant output and call metadata (`LlmCall`) in DB.
- Server updates chat-level model metadata on each call: `lastUsedProvider`/`lastUsedModel`; first successful/failed call also initializes `initiatedProvider`/`initiatedModel` if unset.
- Attachments are optional and currently apply to `user` messages. Persisted chat history stores them under `message.metadata.attachments`.
- Images are forwarded inline to providers as multimodal image parts. Use PNG or JPEG for cross-provider compatibility.
- Text files are forwarded as explicit text blocks rather than provider-managed file references. Large text attachments should already be truncated client-side before submission.
- For `openai`, backend calls OpenAI's Responses API and enables internal tool use with an internal system instruction.
- For `anthropic`, backend calls Anthropic's Messages API and enables internal tool use with Anthropic `tool_use`/`tool_result` content blocks.
- For `xai`, backend calls xAI's OpenAI-compatible Chat Completions API and enables internal tool use with the same internal system instruction.
- For `hermes-agent`, backend calls the configured Hermes Agent OpenAI-compatible Chat Completions API without adding Sybil-managed tool definitions; Hermes Agent handles its own tools server-side.
- For `openai`, image attachments are sent as Responses `input_image` items and text attachments are sent as `input_text` items.
- For `xai` and `hermes-agent`, image attachments are sent as Chat Completions content parts alongside text.
- For `openai`, Responses calls that can enter the server-managed tool loop use `store: true` so reasoning and function-call items can be passed between tool rounds.
- For `anthropic`, image attachments are sent as Messages API `image` blocks using base64 source data; text attachments are added as `text` blocks.
- Available Sybil-managed tool calls for `openai`, `anthropic`, and `xai`: `web_search` and `fetch_url`. When `CHAT_CODEX_TOOL_ENABLED=true`, `codex_exec` is also available. When `CHAT_SHELL_TOOL_ENABLED=true`, `shell_exec` is also available.
- Available Sybil-managed tool calls for `openai` and `xai`: `web_search` and `fetch_url`. When `CHAT_CODEX_TOOL_ENABLED=true`, `codex_exec` is also available. When `CHAT_SHELL_TOOL_ENABLED=true`, `shell_exec` is also available.
- `web_search` returns ranked results with per-result summaries/snippets. Its backend engine is selected by `CHAT_WEB_SEARCH_ENGINE` (`exa` default, or `searxng` with `SEARXNG_BASE_URL` set). SearXNG mode requires the instance to allow `format=json`.
- `fetch_url` fetches a URL with browser-like navigation headers and returns plaintext page content (HTML converted to text server-side).
- `fetch_url` fetches a URL and returns plaintext page content (HTML converted to text server-side).
- `codex_exec` delegates coding, shell, repository inspection, and other complex software tasks to a persistent remote Codex CLI workspace over SSH. The server runs `codex exec --dangerously-bypass-approvals-and-sandbox --skip-git-repo-check <non-interactive wrapped prompt>` on the configured devbox inside `CHAT_CODEX_REMOTE_WORKDIR`, with SSH stdin closed.
- `shell_exec` runs arbitrary non-interactive shell commands on the same configured devbox, starting in `CHAT_CODEX_REMOTE_WORKDIR`. It uses `bash -lc` when bash exists, otherwise `sh -lc`, closes SSH stdin, and does not run inside the Sybil server container.
- Devbox tool configuration:
@@ -315,7 +267,8 @@ Behavior notes:
- `CHAT_CODEX_SSH_PRIVATE_KEY_B64=<base64-private-key>` (optional fallback when a volume mount is not practical)
- `CHAT_CODEX_EXEC_TIMEOUT_MS=600000` (optional)
- `CHAT_SHELL_EXEC_TIMEOUT_MS=120000` (optional)
- When a tool call is executed, backend stores a chat `Message` with `role: "tool"` and tool metadata (`metadata.kind = "tool_call"`). Streaming requests emit an initiated SSE `tool_call` event before execution, then persist each completed or failed tool call as its terminal SSE `tool_call` event is emitted, then store the assistant output when the completion finishes.
- When a tool call is executed, backend stores a chat `Message` with `role: "tool"` and tool metadata (`metadata.kind = "tool_call"`). Streaming requests persist each completed tool call as its SSE `tool_call` event is emitted, then store the assistant output when the completion finishes.
- `anthropic` currently runs without server-managed tool calls.
## Searches
@@ -323,24 +276,8 @@ Behavior notes:
- Response: `{ "searches": SearchSummary[] }`
### `POST /v1/searches`
- Body: `{ "title"?: string, "query"?: string, "reuseByQuery"?: boolean }`
- Response: `{ "search": SearchSummary, "reused": boolean, "cacheHit": boolean }`
Behavior notes:
- `reuseByQuery` defaults to `false`, preserving the normal create-a-new-search behavior.
- When `reuseByQuery` is `true` and `query` is present, the backend normalizes the query with `trim().toLowerCase()` and returns the most recently updated existing search with that normalized query instead of creating a duplicate.
- `cacheHit` is `true` only when the reused search has persisted results or answer text, is not currently streaming, and was updated within the 24-hour search cache window. Clients can then fetch `GET /v1/searches/:searchId` and display it without running another search.
- If a matching search exists but `cacheHit` is `false`, clients may run the search again on the returned `search.id`; the run endpoints replace that search's persisted results and answer with the latest run.
### `PATCH /v1/searches/:searchId/star`
- Body: `{ "starred": boolean }`
- Body: `{ "title"?: string, "query"?: string }`
- Response: `{ "search": SearchSummary }`
- Not found: `404 { "message": "search not found" }`
Behavior notes:
- Starring adds the search to the reserved `starred` project and sets `starredAt` to the membership creation time.
- Unstarring removes that membership and returns `starred: false`, `starredAt: null`.
- This does not modify the search results or search `updatedAt`.
### `DELETE /v1/searches/:searchId`
- Response: `{ "deleted": true }`
@@ -414,14 +351,10 @@ Behavior notes:
"title": null,
"createdAt": "...",
"updatedAt": "...",
"starred": false,
"starredAt": null,
"initiatedProvider": "openai|anthropic|xai|hermes-agent|null",
"initiatedModel": "string|null",
"lastUsedProvider": "openai|anthropic|xai|hermes-agent|null",
"lastUsedModel": "string|null",
"additionalSystemPrompt": null,
"enabledTools": ["web_search", "fetch_url"]
"lastUsedModel": "string|null"
}
```
@@ -466,21 +399,17 @@ Behavior notes:
"title": null,
"createdAt": "...",
"updatedAt": "...",
"starred": false,
"starredAt": null,
"initiatedProvider": "openai|anthropic|xai|hermes-agent|null",
"initiatedModel": "string|null",
"lastUsedProvider": "openai|anthropic|xai|hermes-agent|null",
"lastUsedModel": "string|null",
"additionalSystemPrompt": null,
"enabledTools": ["web_search", "fetch_url"],
"messages": [Message]
}
```
`SearchSummary`
```json
{ "id": "...", "title": null, "query": null, "createdAt": "...", "updatedAt": "...", "starred": false, "starredAt": null }
{ "id": "...", "title": null, "query": null, "createdAt": "...", "updatedAt": "..." }
```
`SearchDetail`
@@ -491,8 +420,6 @@ Behavior notes:
"query": "...",
"createdAt": "...",
"updatedAt": "...",
"starred": false,
"starredAt": null,
"requestId": "...",
"latencyMs": 123,
"error": null,

31
docs/api/streaming-chat.md
View File

@@ -49,8 +49,6 @@ Authentication:
]
}
],
"additionalSystemPrompt": "optional one-off system prompt",
"enabledTools": ["web_search", "fetch_url"],
"temperature": 0.2,
"maxTokens": 256
}
@@ -62,8 +60,6 @@ Notes:
- If `chatId` is provided, backend validates it exists.
- If `persist` is `false`, `chatId` must be omitted. Backend does not create a chat and does not persist input messages, tool-call messages, assistant output, or `LlmCall` metadata.
- For persisted streams, backend stores only new non-assistant input history rows to avoid duplicates.
- `additionalSystemPrompt`, when present directly or loaded from stored chat settings, is prepended to the provider request as a `system` message and is not inserted into the persisted chat transcript by this endpoint.
- `enabledTools` limits Sybil-managed tools for this request. When omitted for a saved chat, the stored chat setting is used; otherwise all available tools are enabled by default. An empty array disables Sybil-managed tools.
- Attachments are optional and are persisted under `message.metadata.attachments` on stored user messages when `persist` is `true`.
Persisted chat streams with a `chatId` are backend-owned active runs:
@@ -91,8 +87,6 @@ Event order:
3. Zero or more `delta`
4. Exactly one terminal event: `done` or `error`
Each tool invocation can emit multiple `tool_call` events with the same `toolCallId`. The backend emits `status: "initiated"` before the tool starts executing, then emits `status: "completed"` or `status: "failed"` when execution finishes. Clients should upsert by `toolCallId` instead of appending each event.
### `meta`
```json
@@ -117,19 +111,6 @@ For `persist: false` streams, `chatId` and `callId` are `null`.
### `tool_call`
```json
{
"toolCallId": "call_123",
"name": "web_search",
"status": "initiated",
"summary": "Searching web for 'latest CPI release'.",
"args": { "query": "latest CPI release" },
"startedAt": "2026-03-02T10:00:00.000Z"
}
```
Terminal tool-call event:
```json
{
"toolCallId": "call_123",
@@ -140,12 +121,11 @@ Terminal tool-call event:
"startedAt": "2026-03-02T10:00:00.000Z",
"completedAt": "2026-03-02T10:00:00.820Z",
"durationMs": 820,
"error": null,
"resultPreview": "{\"ok\":true,...}"
}
```
`status` is one of `initiated`, `completed`, or `failed`. `completedAt` and `durationMs` are only present on terminal events. `error` is present on failed terminal events; `resultPreview` is present on terminal events when available.
### `done`
```json
@@ -171,20 +151,18 @@ Terminal tool-call event:
## Provider Streaming Behavior
- `openai`: backend uses OpenAI's Responses API and may execute internal function tool calls (`web_search`, `fetch_url`, optional `codex_exec`, and optional `shell_exec`) before producing final text.
- `anthropic`: backend uses Anthropic's Messages API and may execute the same internal tools with `tool_use`/`tool_result` content blocks before producing final text.
- `xai`: backend uses xAI's OpenAI-compatible Chat Completions API and may execute the same internal tool calls before producing final text.
- `fetch_url` sends browser-like navigation headers for outbound URL requests to reduce false 403s from sites that reject generic server clients.
- `hermes-agent`: backend uses the configured Hermes Agent OpenAI-compatible Chat Completions API. Sybil does not add its own tool definitions for this provider; Hermes Agent handles its own tools server-side. Custom Hermes stream events are normalized away unless they produce text deltas in this SSE contract.
- `openai`: image attachments are sent as Responses `input_image` items; text attachments are sent as `input_text` items.
- `xai` and `hermes-agent`: image attachments are sent as Chat Completions content parts; text attachments are inlined as text parts.
- `openai`: Responses calls that can enter the server-managed tool loop use `store: true` so reasoning and function-call items can be passed between tool rounds.
- `anthropic`: streamed via event stream; emits `delta` from `content_block_delta` with `text_delta`, and emits normalized `tool_call` SSE events when Anthropic `tool_use` blocks are executed. Image attachments are sent as base64 `image` blocks and text attachments are appended as `text` blocks.
- `anthropic`: streamed via event stream; emits `delta` from `content_block_delta` with `text_delta`. Image attachments are sent as base64 `image` blocks and text attachments are appended as `text` blocks.
- `web_search` uses `CHAT_WEB_SEARCH_ENGINE` (`exa` default, or `searxng` with `SEARXNG_BASE_URL` set). SearXNG mode requires the instance to allow `format=json`. This only affects chat-mode tool calls, not search-mode endpoints.
- `codex_exec` is available only when `CHAT_CODEX_TOOL_ENABLED=true`. It SSHes to `CHAT_CODEX_REMOTE_HOST`, creates/uses `CHAT_CODEX_REMOTE_WORKDIR`, and runs `codex exec --dangerously-bypass-approvals-and-sandbox --skip-git-repo-check <non-interactive wrapped prompt>` there with SSH stdin closed. Prefer `CHAT_CODEX_SSH_KEY_PATH` with a read-only mounted private key; `CHAT_CODEX_SSH_PRIVATE_KEY_B64` is also supported.
- `shell_exec` is available only when `CHAT_SHELL_TOOL_ENABLED=true`. It uses the same devbox SSH configuration, starts in `CHAT_CODEX_REMOTE_WORKDIR`, and runs non-interactive shell commands there with SSH stdin closed, not inside the Sybil server container.
- `CHAT_MAX_TOOL_ROUNDS` controls how many model/tool result cycles may occur before the backend returns a tool-call limit message; default is 100.
Tool-enabled streaming notes (`openai`/`anthropic`/`xai`):
Tool-enabled streaming notes (`openai`/`xai`):
- Stream still emits standard `meta`, `delta`, `done|error` events.
- Stream may emit `tool_call` events while tool calls are executed.
- `delta` events carry assistant text and are emitted incrementally for normal text rounds. The backend may buffer model-native text briefly while determining whether a provider round contains tool calls.
@@ -196,8 +174,7 @@ Backend database remains source of truth.
For persisted streams:
- Client may optimistically render accumulated `delta` text.
- Backend emits initiated tool-call events without persisting them.
- Backend persists each completed or failed tool call as a `tool` message before emitting its terminal `tool_call` SSE event, so chat detail refreshes can show completed tool calls while the assistant response is still running.
- Backend persists each completed tool call as a `tool` message before emitting its `tool_call` SSE event, so chat detail refreshes can show completed tool calls while the assistant response is still running.
On successful persisted completion:
- Backend persists assistant `Message` and updates `LlmCall` usage/latency in a transaction.

20
ios/.env.example
View File

@@ -1,20 +0,0 @@
FASTLANE_APP_IDENTIFIER=net.buzzert.sybil2
FASTLANE_TEAM_ID=DQQH5H6GBD
FASTLANE_USER=you@example.com
FASTLANE_APPLE_APPLICATION_SPECIFIC_PASSWORD=xxxx-xxxx-xxxx-xxxx
FASTLANE_SKIP_UPDATE_CHECK=1
FASTLANE_HIDE_CHANGELOG=1
SYBIL_APP_STORE_APPLE_ID=6759442828
SYBIL_PROVIDER_PUBLIC_ID=c043d167-ad88-4036-84ea-76c223f1b1b2
# Optional App Store Connect API key settings for non-interactive upload and
# TestFlight build-number lookup.
APP_STORE_CONNECT_API_KEY_ID=
APP_STORE_CONNECT_API_ISSUER_ID=
APP_STORE_CONNECT_API_KEY_PATH=
APP_STORE_CONNECT_API_KEY_CONTENT=
APP_STORE_CONNECT_API_KEY_CONTENT_BASE64=false
# Optional deployment overrides.
SYBIL_BUILD_NUMBER=
SYBIL_VERSION_TAG=

11
ios/.gitignore vendored
View File

@@ -1,11 +1,2 @@
*.xcodeproj
.env
.env.*
!.env.example
build/
*.ipa
*.dSYM.zip
fastlane/report.xml
fastlane/Preview.html
fastlane/screenshots/
fastlane/test_output/

4
ios/Apps/Sybil/project.yml
View File

@@ -24,8 +24,8 @@ targets:
GENERATE_INFOPLIST_FILE: YES
INFOPLIST_FILE: Apps/Sybil/Info.plist
ASSETCATALOG_COMPILER_APPICON_NAME: AppIcon
MARKETING_VERSION: "1.10"
CURRENT_PROJECT_VERSION: 11
MARKETING_VERSION: 1.9
CURRENT_PROJECT_VERSION: 10
INFOPLIST_KEY_CFBundleDisplayName: Sybil
INFOPLIST_KEY_ITSAppUsesNonExemptEncryption: NO
INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents: YES

3
ios/Gemfile
View File

@@ -1,3 +0,0 @@
source "https://rubygems.org"
gem "fastlane", "~> 2.227"

38
ios/Packages/Sybil/Sources/Sybil/SybilAPIClient.swift
View File

@@ -74,26 +74,6 @@ actor SybilAPIClient: SybilAPIClienting {
return response.chat
}
func updateChatTitle(chatID: String, title: String) async throws -> ChatSummary {
let response = try await request(
"/v1/chats/\(chatID)",
method: "PATCH",
body: AnyEncodable(ChatTitleUpdateBody(title: title)),
responseType: ChatCreateResponse.self
)
return response.chat
}
func updateChatStar(chatID: String, starred: Bool) async throws -> ChatSummary {
let response = try await request(
"/v1/chats/\(chatID)/star",
method: "PATCH",
body: AnyEncodable(StarUpdateBody(starred: starred)),
responseType: ChatCreateResponse.self
)
return response.chat
}
func deleteChat(chatID: String) async throws {
_ = try await request("/v1/chats/\(chatID)", method: "DELETE", responseType: DeleteResponse.self)
}
@@ -138,16 +118,6 @@ actor SybilAPIClient: SybilAPIClienting {
return response.chat
}
func updateSearchStar(searchID: String, starred: Bool) async throws -> SearchSummary {
let response = try await request(
"/v1/searches/\(searchID)/star",
method: "PATCH",
body: AnyEncodable(StarUpdateBody(starred: starred)),
responseType: SearchCreateResponse.self
)
return response.search
}
func deleteSearch(searchID: String) async throws {
_ = try await request("/v1/searches/\(searchID)", method: "DELETE", responseType: DeleteResponse.self)
}
@@ -671,14 +641,6 @@ private struct ChatCreateBody: Encodable {
var messages: [CompletionRequestMessage]?
}
private struct ChatTitleUpdateBody: Encodable {
var title: String
}
private struct StarUpdateBody: Encodable {
var starred: Bool
}
private struct SearchCreateBody: Encodable {
var title: String?
var query: String?

3
ios/Packages/Sybil/Sources/Sybil/SybilAPIClienting.swift
View File

@@ -11,15 +11,12 @@ protocol SybilAPIClienting: Sendable {
messages: [CompletionRequestMessage]?
) async throws -> ChatSummary
func getChat(chatID: String) async throws -> ChatDetail
func updateChatTitle(chatID: String, title: String) async throws -> ChatSummary
func updateChatStar(chatID: String, starred: Bool) async throws -> ChatSummary
func deleteChat(chatID: String) async throws
func suggestChatTitle(chatID: String, content: String) async throws -> ChatSummary
func listSearches() async throws -> [SearchSummary]
func createSearch(title: String?, query: String?) async throws -> SearchSummary
func getSearch(searchID: String) async throws -> SearchDetail
func createChatFromSearch(searchID: String, title: String?) async throws -> ChatSummary
func updateSearchStar(searchID: String, starred: Bool) async throws -> SearchSummary
func deleteSearch(searchID: String) async throws
func listModels() async throws -> ModelCatalogResponse
func getActiveRuns() async throws -> ActiveRunsResponse

400
ios/Packages/Sybil/Sources/Sybil/SybilChatTranscriptView.swift
View File

@@ -7,134 +7,39 @@ struct SybilChatTranscriptView: View {
var isSending: Bool
var topContentInset: CGFloat = 0
var bottomContentInset: CGFloat = 0
var bottomPinRequestID: Int = 0
@State private var hasTrackedToolCallMessages = false
@State private var knownToolCallMessageIDs: Set<String> = []
private let bottomAnchorID = "sybil-chat-transcript-bottom-anchor"
private var renderItems: [TranscriptRenderItem] {
buildTranscriptRenderItems(from: messages)
private var hasPendingAssistant: Bool {
messages.contains { message in
message.id.hasPrefix("temp-assistant-") && message.content.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty
}
private var toolCallMessageIDs: Set<String> {
Set(messages.compactMap { $0.toolCallMetadata == nil ? nil : $0.id })
}
private var enteringToolCallMessageIDs: Set<String> {
guard hasTrackedToolCallMessages else { return [] }
return toolCallMessageIDs.subtracting(knownToolCallMessageIDs)
}
private var toolCallMessageIDSignature: String {
toolCallMessageIDs.sorted().joined(separator: "|")
}
var body: some View {
ScrollViewReader { proxy in
ScrollView {
LazyVStack(alignment: .leading, spacing: 26) {
ForEach(messages.reversed()) { message in
MessageBubble(message: message, isSending: isSending)
.frame(maxWidth: .infinity)
.scaleEffect(x: 1, y: -1)
}
if isLoading && messages.isEmpty {
Text("Loading messages…")
.font(.sybil(.footnote))
.foregroundStyle(SybilTheme.textMuted)
.padding(.top, 24)
.scaleEffect(x: 1, y: -1)
}
ForEach(renderItems) { item in
switch item {
case let .message(message):
MessageBubble(message: message, isSending: isSending)
.frame(maxWidth: .infinity)
case let .toolGroup(id, messages):
ToolCallStackView(
groupID: id,
messages: messages,
entryAnimationIDs: enteringToolCallMessageIDs
)
.frame(maxWidth: .infinity)
.id(id)
}
}
Color.clear
.frame(height: 18 + bottomContentInset)
.id(bottomAnchorID)
}
.frame(maxWidth: .infinity, alignment: .leading)
.padding(.horizontal, 14)
.padding(.top, 18 + topContentInset)
.padding(.top, 18 + bottomContentInset)
.padding(.bottom, 18 + topContentInset)
}
.frame(maxWidth: .infinity, alignment: .leading)
.scrollDismissesKeyboard(.interactively)
.onAppear {
syncKnownToolCallMessageIDs()
scrollToBottom(with: proxy, animated: false)
.scaleEffect(x: 1, y: -1)
}
.onChange(of: toolCallMessageIDSignature) { _, _ in
syncKnownToolCallMessageIDs()
}
.onChange(of: bottomPinRequestID) { _, _ in
scrollToBottom(with: proxy, animated: true)
}
}
}
private func scrollToBottom(with proxy: ScrollViewProxy, animated: Bool) {
let action = {
proxy.scrollTo(bottomAnchorID, anchor: .bottom)
}
if animated {
withAnimation(.easeOut(duration: 0.18), action)
} else {
action()
}
}
private func syncKnownToolCallMessageIDs() {
guard !toolCallMessageIDs.isEmpty else { return }
knownToolCallMessageIDs.formUnion(toolCallMessageIDs)
hasTrackedToolCallMessages = true
}
}
enum TranscriptRenderItem: Identifiable {
case message(Message)
case toolGroup(id: String, messages: [Message])
var id: String {
switch self {
case let .message(message):
return message.id
case let .toolGroup(id, _):
return "tool-group-\(id)"
}
}
}
func buildTranscriptRenderItems(from messages: [Message]) -> [TranscriptRenderItem] {
var items: [TranscriptRenderItem] = []
var toolRun: [Message] = []
func flushToolRun() {
guard !toolRun.isEmpty else { return }
if toolRun.count == 1, let message = toolRun.first {
items.append(.message(message))
} else if let first = toolRun.first {
items.append(.toolGroup(id: first.id, messages: toolRun))
}
toolRun.removeAll(keepingCapacity: true)
}
for message in messages {
if message.toolCallMetadata != nil {
toolRun.append(message)
} else {
flushToolRun()
items.append(.message(message))
}
}
flushToolRun()
return items
}
private struct MessageBubble: View {
@@ -232,225 +137,10 @@ private struct MessageBubble: View {
}
}
private struct ToolCallStackView: View {
private struct CardLayout {
var x: CGFloat
var y: CGFloat
var scale: CGFloat
var opacity: Double
var zIndex: Double
}
var groupID: String
var messages: [Message]
var entryAnimationIDs: Set<String>
@Environment(\.accessibilityReduceMotion) private var reduceMotion
@State private var isExpanded = false
private let visibleCollapsedLimit = 4
private let cardHeight: CGFloat = 62
private let expandedGap: CGFloat = 10
private let collapsedStepX: CGFloat = 11
private let collapsedStepY: CGFloat = 10
private let toggleSize: CGFloat = 32
private let toggleGap: CGFloat = 12
private var animation: Animation? {
reduceMotion ? nil : .easeInOut(duration: 0.34)
}
private var visibleCollapsedCount: Int {
min(messages.count, visibleCollapsedLimit)
}
private var hiddenCount: Int {
max(0, messages.count - visibleCollapsedLimit)
}
private var containerHeight: CGFloat {
if isExpanded {
return cardHeight + CGFloat(max(0, messages.count - 1)) * (cardHeight + expandedGap)
}
return cardHeight + CGFloat(max(0, visibleCollapsedCount - 1)) * collapsedStepY
}
private var accessibilityLabel: String {
"\(messages.count) tool \(messages.count == 1 ? "call" : "calls")"
}
var body: some View {
HStack(alignment: .top, spacing: 0) {
GeometryReader { geometry in
let cardWidth = max(220, min(520, geometry.size.width - toggleSize - toggleGap))
let toggleX = cardWidth + toggleGap
ZStack(alignment: .topLeading) {
ForEach(Array(messages.enumerated()), id: \.element.id) { index, message in
let layout = layout(for: index)
let depth = messages.count - index - 1
let isHidden = !isExpanded && depth >= visibleCollapsedLimit
let shouldAnimateEntry = entryAnimationIDs.contains(message.id) && !isHidden
ToolCallStackCard(
message: message,
cardHeight: cardHeight,
compactLayout: true,
animateEntry: shouldAnimateEntry
)
.frame(width: cardWidth, height: cardHeight, alignment: .topLeading)
.scaleEffect(layout.scale, anchor: .topLeading)
.opacity(layout.opacity)
.offset(x: layout.x, y: layout.y)
.zIndex(layout.zIndex)
.allowsHitTesting(!isHidden)
.accessibilityHidden(isHidden)
}
if !isExpanded && hiddenCount > 0 {
Text("+\(hiddenCount)")
.font(.sybil(.caption2, weight: .semibold))
.foregroundStyle(SybilTheme.accent.opacity(0.95))
.padding(.horizontal, 7)
.padding(.vertical, 3)
.background(
Capsule()
.fill(Color.black.opacity(0.58))
.overlay(
Capsule()
.stroke(SybilTheme.accent.opacity(0.34), lineWidth: 1)
)
)
.offset(x: max(0, cardWidth - 56), y: containerHeight - 13)
.transition(.opacity)
}
Button {
withAnimation(animation) {
isExpanded.toggle()
}
} label: {
Image(systemName: isExpanded ? "chevron.up" : "chevron.down")
.font(.system(size: 14, weight: .bold))
.foregroundStyle(SybilTheme.accent.opacity(0.95))
.frame(width: toggleSize, height: toggleSize)
.background(
Circle()
.fill(
LinearGradient(
colors: [
Color(red: 0.06, green: 0.08, blue: 0.15).opacity(0.96),
Color(red: 0.03, green: 0.04, blue: 0.10).opacity(0.96)
],
startPoint: .top,
endPoint: .bottom
)
)
.overlay(
Circle()
.stroke(SybilTheme.accent.opacity(0.38), lineWidth: 1)
)
.shadow(color: Color.black.opacity(0.30), radius: 10, x: 0, y: 6)
)
}
.buttonStyle(.plain)
.accessibilityLabel("\(isExpanded ? "Collapse" : "Expand") \(accessibilityLabel)")
.offset(x: toggleX, y: 8)
.zIndex(Double(messages.count + 2))
}
.frame(width: cardWidth + toggleSize + toggleGap, height: containerHeight, alignment: .topLeading)
.animation(animation, value: isExpanded)
}
.frame(height: containerHeight)
Spacer(minLength: 0)
}
.frame(maxWidth: .infinity, alignment: .leading)
}
private func layout(for index: Int) -> CardLayout {
if isExpanded {
return CardLayout(
x: 0,
y: CGFloat(index) * (cardHeight + expandedGap),
scale: 1,
opacity: 1,
zIndex: Double(messages.count - index)
)
}
let depth = messages.count - index - 1
let visibleDepth = min(depth, visibleCollapsedLimit - 1)
let isHidden = depth >= visibleCollapsedLimit
return CardLayout(
x: CGFloat(visibleDepth) * collapsedStepX,
y: CGFloat(visibleDepth) * collapsedStepY,
scale: max(0.88, 1 - CGFloat(visibleDepth) * 0.035),
opacity: isHidden ? 0 : max(0.34, 1 - Double(visibleDepth) * 0.22),
zIndex: isHidden ? 0 : Double(visibleCollapsedCount - visibleDepth)
)
}
}
private struct ToolCallStackCard: View {
var message: Message
var cardHeight: CGFloat
var compactLayout: Bool
var animateEntry: Bool
@Environment(\.accessibilityReduceMotion) private var reduceMotion
@State private var entryAnimationArmed = false
@State private var didEnter = false
private var isPreparingEntry: Bool {
(animateEntry || entryAnimationArmed) && !didEnter
}
var body: some View {
Group {
if let metadata = message.toolCallMetadata {
ToolCallActivityChip(
metadata: metadata,
fallbackContent: message.content,
createdAt: message.createdAt,
compactLayout: compactLayout
)
}
}
.frame(height: cardHeight, alignment: .top)
.scaleEffect(isPreparingEntry ? 1.025 : 1, anchor: .topLeading)
.offset(y: isPreparingEntry ? -8 : 0)
.rotation3DEffect(.degrees(isPreparingEntry ? 3 : 0), axis: (x: 1, y: 0, z: 0), anchor: .top)
.opacity(isPreparingEntry ? 0.72 : 1)
.onAppear {
guard !didEnter, !entryAnimationArmed else { return }
guard animateEntry else {
didEnter = true
return
}
entryAnimationArmed = true
if reduceMotion {
didEnter = true
} else {
withAnimation(.easeOut(duration: 0.32).delay(0.03)) {
didEnter = true
}
}
}
}
}
private struct ToolCallActivityChip: View {
enum VisualState {
case initiated
case completed
case failed
}
var metadata: ToolCallMetadata
var fallbackContent: String
var createdAt: Date
var compactLayout: Bool = false
private var summary: String {
if let text = metadata.summary?.trimmingCharacters(in: .whitespacesAndNewlines), !text.isEmpty {
@@ -494,22 +184,11 @@ private struct ToolCallActivityChip: View {
}
private var isFailed: Bool {
visualState == .failed
}
private var visualState: VisualState {
switch (metadata.status ?? "").lowercased() {
case "failed":
return .failed
case "initiated":
return .initiated
default:
return .completed
}
(metadata.status ?? "").lowercased() == "failed"
}
private var detailLabel: String {
var pieces: [String] = [stateLabel]
var pieces: [String] = [isFailed ? "Failed" : "Completed"]
if let durationMs = metadata.durationMs, durationMs > 0 {
pieces.append("\(durationMs) ms")
}
@@ -521,14 +200,14 @@ private struct ToolCallActivityChip: View {
HStack(alignment: .top, spacing: 11) {
ZStack {
RoundedRectangle(cornerRadius: 9)
.fill(iconColor.opacity(0.13))
.fill((isFailed ? SybilTheme.danger : SybilTheme.accent).opacity(0.13))
.overlay(
RoundedRectangle(cornerRadius: 9)
.stroke(iconColor.opacity(0.34), lineWidth: 1)
.stroke((isFailed ? SybilTheme.danger : SybilTheme.accent).opacity(0.34), lineWidth: 1)
)
Image(systemName: iconName)
.font(.system(size: 14, weight: .semibold))
.foregroundStyle(iconColor)
.foregroundStyle(isFailed ? SybilTheme.danger : SybilTheme.accent)
}
.frame(width: 30, height: 30)
@@ -537,14 +216,12 @@ private struct ToolCallActivityChip: View {
.font(.sybil(.subheadline))
.foregroundStyle(isFailed ? SybilTheme.danger.opacity(0.96) : SybilTheme.text.opacity(0.94))
.lineSpacing(3)
.lineLimit(compactLayout ? 1 : nil)
.truncationMode(.tail)
.fixedSize(horizontal: false, vertical: !compactLayout)
.fixedSize(horizontal: false, vertical: true)
HStack(spacing: 6) {
Text(toolLabel)
.font(.sybil(.caption2, weight: .semibold))
.foregroundStyle(iconColor.opacity(0.90))
.foregroundStyle(isFailed ? SybilTheme.danger.opacity(0.84) : SybilTheme.accent.opacity(0.90))
.lineLimit(1)
Text(detailLabel)
@@ -559,45 +236,12 @@ private struct ToolCallActivityChip: View {
.padding(.vertical, 10)
.background(
RoundedRectangle(cornerRadius: 12)
.fill(backgroundGradient)
.fill(isFailed ? SybilTheme.failedToolCallGradient : SybilTheme.toolCallGradient)
.overlay(
RoundedRectangle(cornerRadius: 12)
.stroke(iconColor.opacity(0.34), lineWidth: 1)
.stroke((isFailed ? SybilTheme.danger : SybilTheme.accent).opacity(0.34), lineWidth: 1)
)
)
.frame(maxWidth: 520, alignment: .leading)
}
private var stateLabel: String {
switch visualState {
case .failed:
return "Failed"
case .initiated:
return "Running"
case .completed:
return "Completed"
}
}
private var iconColor: Color {
switch visualState {
case .failed:
return SybilTheme.danger
case .initiated:
return SybilTheme.warning
case .completed:
return SybilTheme.accent
}
}
private var backgroundGradient: LinearGradient {
switch visualState {
case .failed:
return SybilTheme.failedToolCallGradient
case .initiated:
return SybilTheme.runningToolCallGradient
case .completed:
return SybilTheme.toolCallGradient
}
}
}

24
ios/Packages/Sybil/Sources/Sybil/SybilModels.swift
View File

@@ -154,8 +154,6 @@ public struct ChatSummary: Codable, Identifiable, Hashable, Sendable {
public var title: String?
public var createdAt: Date
public var updatedAt: Date
public var starred = false
public var starredAt: Date?
public var initiatedProvider: Provider?
public var initiatedModel: String?
public var lastUsedProvider: Provider?
@@ -168,8 +166,6 @@ public struct SearchSummary: Codable, Identifiable, Hashable, Sendable {
public var query: String?
public var createdAt: Date
public var updatedAt: Date
public var starred = false
public var starredAt: Date?
}
public enum WorkspaceItemType: String, Codable, Hashable, Sendable {
@@ -184,8 +180,6 @@ public struct WorkspaceItem: Codable, Identifiable, Hashable, Sendable {
public var query: String?
public var createdAt: Date
public var updatedAt: Date
public var starred = false
public var starredAt: Date?
public var initiatedProvider: Provider?
public var initiatedModel: String?
public var lastUsedProvider: Provider?
@@ -198,8 +192,6 @@ public struct WorkspaceItem: Codable, Identifiable, Hashable, Sendable {
self.query = nil
self.createdAt = chat.createdAt
self.updatedAt = chat.updatedAt
self.starred = chat.starred
self.starredAt = chat.starredAt
self.initiatedProvider = chat.initiatedProvider
self.initiatedModel = chat.initiatedModel
self.lastUsedProvider = chat.lastUsedProvider
@@ -213,8 +205,6 @@ public struct WorkspaceItem: Codable, Identifiable, Hashable, Sendable {
self.query = search.query
self.createdAt = search.createdAt
self.updatedAt = search.updatedAt
self.starred = search.starred
self.starredAt = search.starredAt
self.initiatedProvider = nil
self.initiatedModel = nil
self.lastUsedProvider = nil
@@ -228,8 +218,6 @@ public struct WorkspaceItem: Codable, Identifiable, Hashable, Sendable {
title: title,
createdAt: createdAt,
updatedAt: updatedAt,
starred: starred,
starredAt: starredAt,
initiatedProvider: initiatedProvider,
initiatedModel: initiatedModel,
lastUsedProvider: lastUsedProvider,
@@ -244,9 +232,7 @@ public struct WorkspaceItem: Codable, Identifiable, Hashable, Sendable {
title: title,
query: query,
createdAt: createdAt,
updatedAt: updatedAt,
starred: starred,
starredAt: starredAt
updatedAt: updatedAt
)
}
}
@@ -391,8 +377,6 @@ public struct ChatDetail: Codable, Identifiable, Hashable, Sendable {
public var title: String?
public var createdAt: Date
public var updatedAt: Date
public var starred = false
public var starredAt: Date?
public var initiatedProvider: Provider?
public var initiatedModel: String?
public var lastUsedProvider: Provider?
@@ -431,8 +415,6 @@ public struct SearchDetail: Codable, Identifiable, Hashable, Sendable {
public var query: String?
public var createdAt: Date
public var updatedAt: Date
public var starred = false
public var starredAt: Date?
public var requestId: String?
public var latencyMs: Int?
public var error: String?
@@ -514,8 +496,8 @@ public struct CompletionStreamToolCall: Codable, Sendable {
public var summary: String
public var args: [String: JSONValue]
public var startedAt: String
public var completedAt: String?
public var durationMs: Int?
public var completedAt: String
public var durationMs: Int
public var error: String?
public var resultPreview: String?
}

58
ios/Packages/Sybil/Sources/Sybil/SybilSidebarView.swift
View File

@@ -111,22 +111,8 @@ struct SybilSidebarItemList: View {
@Bindable var viewModel: SybilViewModel
var isSelected: (SidebarItem) -> Bool
var onSelect: (SidebarItem) -> Void
@State private var renameTarget: SidebarItem?
@State private var renameTitle = ""
private var isRenameAlertPresented: Binding<Bool> {
Binding {
renameTarget != nil
} set: { isPresented in
if !isPresented {
renameTarget = nil
renameTitle = ""
}
}
}
var body: some View {
Group {
if viewModel.isLoadingCollections && viewModel.sidebarItems.isEmpty {
VStack(alignment: .leading, spacing: 8) {
ProgressView()
@@ -160,23 +146,6 @@ struct SybilSidebarItemList: View {
}
.buttonStyle(.plain)
.contextMenu {
Button {
Task {
await viewModel.setItemStarred(item.selection, starred: !item.starred)
}
} label: {
Label(item.starred ? "Unstar" : "Star", systemImage: item.starred ? "star.slash" : "star")
}
if item.kind == .chat {
Button {
renameTarget = item
renameTitle = item.title
} label: {
Label("Rename", systemImage: "pencil")
}
}
Button(role: .destructive) {
Task {
await viewModel.deleteItem(item.selection)
@@ -194,27 +163,6 @@ struct SybilSidebarItemList: View {
}
}
}
.alert("Rename Chat", isPresented: isRenameAlertPresented) {
TextField("Title", text: $renameTitle)
Button("Cancel", role: .cancel) {
renameTarget = nil
renameTitle = ""
}
Button("Save") {
let target = renameTarget
let title = renameTitle
renameTarget = nil
renameTitle = ""
if let target, case let .chat(chatID) = target.selection {
Task {
await viewModel.renameChat(chatID: chatID, title: title)
}
}
}
.disabled(renameTitle.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty)
}
}
}
struct SybilSidebarRow: View {
@@ -253,12 +201,6 @@ struct SybilSidebarRow: View {
.lineLimit(1)
.layoutPriority(1)
if item.starred {
Image(systemName: "star.fill")
.font(.system(size: 10, weight: .semibold))
.foregroundStyle(.yellow)
}
Spacer(minLength: 8)
if item.isRunning {

20
ios/Packages/Sybil/Sources/Sybil/SybilTheme.swift
View File

@@ -78,7 +78,6 @@ enum SybilTheme {
static let searchCard = Color(red: 0.07, green: 0.06, blue: 0.14)
static let userBubble = Color(red: 0.29, green: 0.13, blue: 0.65)
static let danger = Color(red: 0.96, green: 0.32, blue: 0.40)
static let warning = Color(red: 0.95, green: 0.69, blue: 0.25)
@MainActor static func applySystemAppearance() {
let navAppearance = UINavigationBarAppearance()
@@ -179,19 +178,8 @@ enum SybilTheme {
static var toolCallGradient: LinearGradient {
LinearGradient(
colors: [
Color(red: 0.01, green: 0.15, blue: 0.17),
Color(red: 0.03, green: 0.09, blue: 0.15)
],
startPoint: .leading,
endPoint: .trailing
)
}
static var runningToolCallGradient: LinearGradient {
LinearGradient(
colors: [
Color(red: 0.30, green: 0.19, blue: 0.04),
Color(red: 0.09, green: 0.05, blue: 0.17)
Color(red: 0.01, green: 0.15, blue: 0.17).opacity(0.70),
Color(red: 0.03, green: 0.09, blue: 0.15).opacity(0.78)
],
startPoint: .leading,
endPoint: .trailing
@@ -201,8 +189,8 @@ enum SybilTheme {
static var failedToolCallGradient: LinearGradient {
LinearGradient(
colors: [
Color(red: 0.27, green: 0.04, blue: 0.10),
Color(red: 0.15, green: 0.03, blue: 0.07)
danger.opacity(0.18),
Color(red: 0.15, green: 0.03, blue: 0.07).opacity(0.72)
],
startPoint: .leading,
endPoint: .trailing

174
ios/Packages/Sybil/Sources/Sybil/SybilViewModel.swift
View File

@@ -34,8 +34,6 @@ struct SidebarItem: Identifiable, Hashable {
var kind: Kind
var title: String
var updatedAt: Date
var starred: Bool
var starredAt: Date?
var initiatedLabel: String?
var isRunning: Bool
}
@@ -107,7 +105,6 @@ final class SybilViewModel {
var isLoadingCollections = false
var isLoadingSelection = false
var isCreatingSearchChat = false
var chatBottomPinRequestID = 0
var errorMessage: String?
var composer = ""
@@ -411,8 +408,6 @@ final class SybilViewModel {
kind: .chat,
title: chatTitle(title: item.title, messages: nil),
updatedAt: item.updatedAt,
starred: item.starred,
starredAt: item.starredAt,
initiatedLabel: initiatedLabel,
isRunning: isChatRowRunning(item.id)
)
@@ -423,8 +418,6 @@ final class SybilViewModel {
kind: .search,
title: searchTitle(title: item.title, query: item.query),
updatedAt: item.updatedAt,
starred: item.starred,
starredAt: item.starredAt,
initiatedLabel: "exa",
isRunning: isSearchRowRunning(item.id)
)
@@ -688,8 +681,6 @@ final class SybilViewModel {
title: chat.title,
createdAt: chat.createdAt,
updatedAt: chat.updatedAt,
starred: chat.starred,
starredAt: chat.starredAt,
initiatedProvider: chat.initiatedProvider,
initiatedModel: chat.initiatedModel,
lastUsedProvider: chat.lastUsedProvider,
@@ -860,57 +851,6 @@ final class SybilViewModel {
}
}
func renameChat(chatID: String, title: String) async {
guard isAuthenticated else {
return
}
let trimmedTitle = title.trimmingCharacters(in: .whitespacesAndNewlines)
guard !trimmedTitle.isEmpty else {
errorMessage = "Enter a chat title."
return
}
SybilLog.info(SybilLog.ui, "Renaming chat \(chatID)")
errorMessage = nil
do {
let updated = try await client().updateChatTitle(chatID: chatID, title: trimmedTitle)
applyChatSummary(updated, moveToFront: true)
} catch {
errorMessage = normalizeAPIError(error)
SybilLog.error(SybilLog.ui, "Rename failed", error: error)
}
}
func setItemStarred(_ selection: SidebarSelection, starred: Bool) async {
guard isAuthenticated else {
return
}
guard case .settings = selection else {
errorMessage = nil
do {
let client = try client()
switch selection {
case let .chat(chatID):
let updated = try await client.updateChatStar(chatID: chatID, starred: starred)
applyChatSummary(updated, moveToFront: false)
case let .search(searchID):
let updated = try await client.updateSearchStar(searchID: searchID, starred: starred)
applySearchSummary(updated, moveToFront: false)
case .settings:
break
}
} catch {
errorMessage = normalizeAPIError(error)
SybilLog.error(SybilLog.ui, "Star update failed", error: error)
}
return
}
}
func refreshAfterSettingsChange() async {
SybilLog.info(SybilLog.ui, "Settings changed, reconnecting")
settings.persist()
@@ -1187,7 +1127,7 @@ final class SybilViewModel {
break
case let .toolCall(payload):
upsertQuickQuestionToolCallMessage(payload)
insertQuickQuestionToolCallMessage(payload)
case let .delta(payload):
guard !payload.text.isEmpty else { return }
@@ -1441,47 +1381,6 @@ final class SybilViewModel {
searches = items.compactMap(\.searchSummary)
}
private func applyChatSummary(_ chat: ChatSummary, moveToFront: Bool) {
if let existingIndex = chats.firstIndex(where: { $0.id == chat.id }) {
chats.remove(at: existingIndex)
chats.insert(chat, at: moveToFront ? 0 : existingIndex)
} else {
chats.insert(chat, at: 0)
}
upsertWorkspaceChat(chat, moveToFront: moveToFront)
if selectedChat?.id == chat.id {
selectedChat?.title = chat.title
selectedChat?.updatedAt = chat.updatedAt
selectedChat?.starred = chat.starred
selectedChat?.starredAt = chat.starredAt
selectedChat?.initiatedProvider = chat.initiatedProvider
selectedChat?.initiatedModel = chat.initiatedModel
selectedChat?.lastUsedProvider = chat.lastUsedProvider
selectedChat?.lastUsedModel = chat.lastUsedModel
}
}
private func applySearchSummary(_ search: SearchSummary, moveToFront: Bool) {
if let existingIndex = searches.firstIndex(where: { $0.id == search.id }) {
searches.remove(at: existingIndex)
searches.insert(search, at: moveToFront ? 0 : existingIndex)
} else {
searches.insert(search, at: 0)
}
upsertWorkspaceSearch(search, moveToFront: moveToFront)
if selectedSearch?.id == search.id {
selectedSearch?.title = search.title
selectedSearch?.query = search.query
selectedSearch?.updatedAt = search.updatedAt
selectedSearch?.starred = search.starred
selectedSearch?.starredAt = search.starredAt
}
}
private func upsertWorkspaceChat(_ chat: ChatSummary, moveToFront: Bool = true) {
upsertWorkspaceItem(WorkspaceItem(chat: chat), moveToFront: moveToFront)
}
@@ -1700,10 +1599,6 @@ final class SybilViewModel {
isLoadingSelection = false
}
private func requestChatBottomPin() {
chatBottomPinRequestID += 1
}
private func startSelectionRefreshTask() -> Task<Void, Never> {
isLoadingSelection = true
let task = Task { [weak self] in
@@ -1757,7 +1652,6 @@ final class SybilViewModel {
}
selectedChat = chat
selectedSearch = nil
requestChatBottomPin()
if let provider = chat.lastUsedProvider,
let model = chat.lastUsedModel,
@@ -1830,7 +1724,6 @@ final class SybilViewModel {
} else {
pendingDraftChatState = PendingChatState(chatID: nil, messages: optimisticMessages)
}
requestChatBottomPin()
if chatID == nil {
let created = try await client.createChat(title: nil)
@@ -1852,8 +1745,6 @@ final class SybilViewModel {
title: created.title,
createdAt: created.createdAt,
updatedAt: created.updatedAt,
starred: created.starred,
starredAt: created.starredAt,
initiatedProvider: created.initiatedProvider,
initiatedModel: created.initiatedModel,
lastUsedProvider: created.lastUsedProvider,
@@ -1878,7 +1769,6 @@ final class SybilViewModel {
if let draftPending = pendingDraftChatState {
pendingDraftChatState = nil
pendingChatStates[chatID] = PendingChatState(chatID: chatID, messages: draftPending.messages)
requestChatBottomPin()
} else if pendingChatStates[chatID] == nil {
pendingChatStates[chatID] = PendingChatState(chatID: chatID, messages: optimisticMessages)
} else {
@@ -1915,7 +1805,18 @@ final class SybilViewModel {
let titleSeed = !content.isEmpty ? content : SybilChatAttachmentSupport.attachmentSummary(attachments)
let updated = try await client.suggestChatTitle(chatID: chatID, content: titleSeed.isEmpty ? "Uploaded files" : titleSeed)
await MainActor.run {
self.applyChatSummary(updated, moveToFront: false)
self.chats = self.chats.map { existing in
if existing.id == updated.id {
return updated
}
return existing
}
self.upsertWorkspaceChat(updated, moveToFront: false)
if self.selectedChat?.id == updated.id {
self.selectedChat?.title = updated.title
self.selectedChat?.updatedAt = updated.updatedAt
}
}
} catch {
SybilLog.warning(SybilLog.app, "Chat title suggestion failed: \(SybilLog.describe(error))")
@@ -2014,7 +1915,7 @@ final class SybilViewModel {
}
case let .toolCall(payload):
upsertPendingToolCallMessage(payload, chatID: chatID)
insertPendingToolCallMessage(payload, chatID: chatID)
case let .delta(payload):
guard !payload.text.isEmpty else { return }
@@ -2074,8 +1975,6 @@ final class SybilViewModel {
query: query,
createdAt: currentSelectedSearch?.createdAt ?? now,
updatedAt: now,
starred: currentSelectedSearch?.starred ?? false,
starredAt: currentSelectedSearch?.starredAt,
requestId: nil,
latencyMs: nil,
error: nil,
@@ -2230,14 +2129,12 @@ final class SybilViewModel {
quickQuestionMessages[index].content = transform(quickQuestionMessages[index].content)
}
private func upsertPendingToolCallMessage(_ payload: CompletionStreamToolCall, chatID: String) {
private func insertPendingToolCallMessage(_ payload: CompletionStreamToolCall, chatID: String) {
guard var pending = pendingChatStates[chatID] else {
return
}
if let existingIndex = pending.messages.firstIndex(where: { $0.toolCallMetadata?.toolCallId == payload.toolCallId || $0.id == "temp-tool-\(payload.toolCallId)" }) {
pending.messages[existingIndex] = toolCallMessage(for: payload, id: pending.messages[existingIndex].id)
pendingChatStates[chatID] = pending
if pending.messages.contains(where: { $0.toolCallMetadata?.toolCallId == payload.toolCallId }) {
return
}
@@ -2252,9 +2149,8 @@ final class SybilViewModel {
pendingChatStates[chatID] = pending
}
private func upsertQuickQuestionToolCallMessage(_ payload: CompletionStreamToolCall) {
if let existingIndex = quickQuestionMessages.firstIndex(where: { $0.toolCallMetadata?.toolCallId == payload.toolCallId || $0.id == "temp-tool-\(payload.toolCallId)" }) {
quickQuestionMessages[existingIndex] = toolCallMessage(for: payload, id: quickQuestionMessages[existingIndex].id)
private func insertQuickQuestionToolCallMessage(_ payload: CompletionStreamToolCall) {
if quickQuestionMessages.contains(where: { $0.toolCallMetadata?.toolCallId == payload.toolCallId }) {
return
}
@@ -2266,8 +2162,8 @@ final class SybilViewModel {
}
}
private func toolCallMessage(for payload: CompletionStreamToolCall, id: String? = nil) -> Message {
var metadataObject: [String: JSONValue] = [
private func toolCallMessage(for payload: CompletionStreamToolCall) -> Message {
let metadata: JSONValue = .object([
"kind": .string("tool_call"),
"toolCallId": .string(payload.toolCallId),
"toolName": .string(payload.name),
@@ -2275,26 +2171,19 @@ final class SybilViewModel {
"summary": .string(payload.summary),
"args": .object(payload.args),
"startedAt": .string(payload.startedAt),
"completedAt": .string(payload.completedAt),
"durationMs": .number(Double(payload.durationMs)),
"error": payload.error.map { .string($0) } ?? .null,
"resultPreview": payload.resultPreview.map { .string($0) } ?? .null
]
if let completedAt = payload.completedAt {
metadataObject["completedAt"] = .string(completedAt)
}
if let durationMs = payload.durationMs {
metadataObject["durationMs"] = .number(Double(durationMs))
}
let metadata: JSONValue = .object(metadataObject)
])
let summary = payload.summary.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty
? "Ran tool '\(payload.name)'."
: payload.summary
return Message(
id: id ?? "temp-tool-\(payload.toolCallId)",
createdAt: toolCallDate(from: payload.completedAt) ?? toolCallDate(from: payload.startedAt) ?? Date(),
id: "temp-tool-\(payload.toolCallId)",
createdAt: Date(),
role: .tool,
content: summary,
name: payload.name,
@@ -2302,19 +2191,6 @@ final class SybilViewModel {
)
}
private func toolCallDate(from value: String?) -> Date? {
guard let value else { return nil }
let fractionalFormatter = ISO8601DateFormatter()
fractionalFormatter.formatOptions = [.withInternetDateTime, .withFractionalSeconds]
if let date = fractionalFormatter.date(from: value) {
return date
}
let formatter = ISO8601DateFormatter()
formatter.formatOptions = [.withInternetDateTime]
return formatter.date(from: value)
}
private var currentChatID: String? {
if draftKind == .chat {
return nil

3
ios/Packages/Sybil/Sources/Sybil/SybilWorkspaceView.swift
View File

@@ -194,8 +194,7 @@ struct SybilWorkspaceView: View {
isLoading: viewModel.isLoadingSelection,
isSending: viewModel.isSendingVisibleChat,
topContentInset: showsCustomWorkspaceNavigation ? customWorkspaceNavigationContentInset : 0,
bottomContentInset: viewModel.showsComposer ? composerOverlayContentInset : 0,
bottomPinRequestID: viewModel.chatBottomPinRequestID
bottomContentInset: viewModel.showsComposer ? composerOverlayContentInset : 0
)
.id(transcriptScrollContextID)
}

203
ios/Packages/Sybil/Tests/SybilTests/SybilTests.swift
View File

@@ -9,9 +9,6 @@ private struct MockClientCallSnapshot: Sendable {
var listSearches = 0
var createChat = 0
var getChat = 0
var updateChatTitle = 0
var updateChatStar = 0
var updateSearchStar = 0
var getSearch = 0
var getActiveRuns = 0
var runCompletionStream = 0
@@ -35,9 +32,6 @@ private actor MockSybilClient: SybilAPIClienting {
private let chatDetails: [String: ChatDetail]
private let searchDetails: [String: SearchDetail]
private let createChatResponse: ChatSummary?
private let updateChatTitleResponses: [String: ChatSummary]
private let updateChatStarResponses: [String: ChatSummary]
private let updateSearchStarResponses: [String: SearchSummary]
private let activeRunsResponse: ActiveRunsResponse
private var snapshot = MockClientCallSnapshot()
@@ -63,9 +57,6 @@ private actor MockSybilClient: SybilAPIClienting {
chatDetails: [String: ChatDetail] = [:],
searchDetails: [String: SearchDetail] = [:],
createChatResponse: ChatSummary? = nil,
updateChatTitleResponses: [String: ChatSummary] = [:],
updateChatStarResponses: [String: ChatSummary] = [:],
updateSearchStarResponses: [String: SearchSummary] = [:],
activeRunsResponse: ActiveRunsResponse = ActiveRunsResponse(),
workspaceItemsResponse: [WorkspaceItem]? = nil
) {
@@ -75,9 +66,6 @@ private actor MockSybilClient: SybilAPIClienting {
self.chatDetails = chatDetails
self.searchDetails = searchDetails
self.createChatResponse = createChatResponse
self.updateChatTitleResponses = updateChatTitleResponses
self.updateChatStarResponses = updateChatStarResponses
self.updateSearchStarResponses = updateSearchStarResponses
self.activeRunsResponse = activeRunsResponse
}
@@ -194,22 +182,6 @@ private actor MockSybilClient: SybilAPIClienting {
return detail
}
func updateChatTitle(chatID: String, title: String) async throws -> ChatSummary {
snapshot.updateChatTitle += 1
guard let summary = updateChatTitleResponses[chatID] else {
throw UnexpectedClientCall()
}
return summary
}
func updateChatStar(chatID: String, starred: Bool) async throws -> ChatSummary {
snapshot.updateChatStar += 1
guard let summary = updateChatStarResponses[chatID] else {
throw UnexpectedClientCall()
}
return summary
}
func deleteChat(chatID: String) async throws {
throw UnexpectedClientCall()
}
@@ -245,14 +217,6 @@ private actor MockSybilClient: SybilAPIClienting {
throw UnexpectedClientCall()
}
func updateSearchStar(searchID: String, starred: Bool) async throws -> SearchSummary {
snapshot.updateSearchStar += 1
guard let summary = updateSearchStarResponses[searchID] else {
throw UnexpectedClientCall()
}
return summary
}
func deleteSearch(searchID: String) async throws {
throw UnexpectedClientCall()
}
@@ -402,70 +366,6 @@ private func makeSearchDetail(id: String, date: Date, answer: String) -> SearchD
)
}
private func makeToolCallMessage(id: String, date: Date, summary: String = "Ran a tool") -> Message {
Message(
id: id,
createdAt: date,
role: .tool,
content: summary,
name: "web_search",
metadata: .object([
"kind": .string("tool_call"),
"toolCallId": .string("call-\(id)"),
"toolName": .string("web_search"),
"status": .string("completed"),
"summary": .string(summary),
"durationMs": .number(120)
])
)
}
@Test func transcriptRenderItemsGroupAdjacentToolCalls() async throws {
let date = Date(timeIntervalSince1970: 1_700_000_000)
let user = Message(id: "user-1", createdAt: date, role: .user, content: "Search this", name: nil)
let toolA = makeToolCallMessage(id: "tool-a", date: date, summary: "Search A")
let toolB = makeToolCallMessage(id: "tool-b", date: date, summary: "Search B")
let assistant = Message(id: "assistant-1", createdAt: date, role: .assistant, content: "Answer", name: nil)
let items = buildTranscriptRenderItems(from: [user, toolA, toolB, assistant])
#expect(items.count == 3)
guard case let .message(firstMessage) = items[0] else {
Issue.record("Expected the first item to remain a normal message")
return
}
#expect(firstMessage.id == "user-1")
guard case let .toolGroup(groupID, groupedMessages) = items[1] else {
Issue.record("Expected adjacent tool calls to be grouped")
return
}
#expect(groupID == "tool-a")
#expect(groupedMessages.map(\.id) == ["tool-a", "tool-b"])
guard case let .message(lastMessage) = items[2] else {
Issue.record("Expected the assistant response to remain a normal message")
return
}
#expect(lastMessage.id == "assistant-1")
}
@Test func transcriptRenderItemsKeepSingleToolCallsInline() async throws {
let date = Date(timeIntervalSince1970: 1_700_000_000)
let user = Message(id: "user-1", createdAt: date, role: .user, content: "Search this", name: nil)
let tool = makeToolCallMessage(id: "tool-a", date: date)
let assistant = Message(id: "assistant-1", createdAt: date, role: .assistant, content: "Answer", name: nil)
let items = buildTranscriptRenderItems(from: [user, tool, assistant])
#expect(items.count == 3)
guard case let .message(toolMessage) = items[1] else {
Issue.record("Expected a single tool call to use the existing inline chip")
return
}
#expect(toolMessage.id == "tool-a")
}
@MainActor
@Test func normalizedAPIBaseURLPreservesExplicitAPIPath() async throws {
let defaults = UserDefaults(suiteName: #function)!
@@ -559,78 +459,6 @@ private func makeToolCallMessage(id: String, date: Date, summary: String = "Ran
#expect(snapshot.listSearches == 0)
#expect(snapshot.getChat == 1)
#expect(viewModel.selectedChat?.messages.first?.content == "refreshed transcript")
#expect(viewModel.chatBottomPinRequestID == 1)
}
@MainActor
@Test func renameChatUpdatesSidebarAndSelectedTranscriptTitle() async throws {
let date = Date(timeIntervalSince1970: 1_700_000_150)
let original = makeChatSummary(id: "chat-rename", date: date)
let renamed = ChatSummary(
id: "chat-rename",
title: "Renamed chat",
createdAt: date,
updatedAt: date.addingTimeInterval(60),
initiatedProvider: .openai,
initiatedModel: "gpt-4.1-mini",
lastUsedProvider: .openai,
lastUsedModel: "gpt-4.1-mini"
)
let detail = makeChatDetail(id: "chat-rename", date: date, body: "existing transcript")
let client = MockSybilClient(
chatsResponse: [original],
updateChatTitleResponses: ["chat-rename": renamed]
)
let viewModel = SybilViewModel(settings: testSettings(named: #function)) { _ in client }
viewModel.isAuthenticated = true
viewModel.isCheckingSession = false
viewModel.chats = [original]
viewModel.workspaceItems = [WorkspaceItem(chat: original)]
viewModel.selectedItem = .chat("chat-rename")
viewModel.selectedChat = detail
await viewModel.renameChat(chatID: "chat-rename", title: " Renamed chat ")
let snapshot = await client.currentSnapshot()
#expect(snapshot.updateChatTitle == 1)
#expect(viewModel.sidebarItems.first?.title == "Renamed chat")
#expect(viewModel.selectedChat?.title == "Renamed chat")
#expect(viewModel.errorMessage == nil)
}
@MainActor
@Test func starringItemsUpdatesSidebarState() async throws {
let date = Date(timeIntervalSince1970: 1_700_000_175)
let chat = makeChatSummary(id: "chat-star", date: date)
let search = makeSearchSummary(id: "search-star", date: date)
var starredChat = chat
starredChat.starred = true
starredChat.starredAt = date.addingTimeInterval(5)
var starredSearch = search
starredSearch.starred = true
starredSearch.starredAt = date.addingTimeInterval(10)
let client = MockSybilClient(
chatsResponse: [chat],
searchesResponse: [search],
updateChatStarResponses: ["chat-star": starredChat],
updateSearchStarResponses: ["search-star": starredSearch]
)
let viewModel = SybilViewModel(settings: testSettings(named: #function)) { _ in client }
viewModel.isAuthenticated = true
viewModel.isCheckingSession = false
viewModel.chats = [chat]
viewModel.searches = [search]
viewModel.workspaceItems = [WorkspaceItem(chat: chat), WorkspaceItem(search: search)]
await viewModel.setItemStarred(.chat("chat-star"), starred: true)
await viewModel.setItemStarred(.search("search-star"), starred: true)
let snapshot = await client.currentSnapshot()
#expect(snapshot.updateChatStar == 1)
#expect(snapshot.updateSearchStar == 1)
#expect(viewModel.sidebarItems.first(where: { $0.selection == .chat("chat-star") })?.starred == true)
#expect(viewModel.sidebarItems.first(where: { $0.selection == .search("search-star") })?.starred == true)
}
@MainActor
@@ -747,37 +575,6 @@ private func makeToolCallMessage(id: String, date: Date, summary: String = "Ran
await sendTask.value
}
@MainActor
@Test func chatBottomPinRequestDoesNotFollowAssistantStreaming() async throws {
let date = Date(timeIntervalSince1970: 1_700_000_245)
let chat = makeChatSummary(id: "chat-pin", date: date)
let detail = makeChatDetail(id: "chat-pin", date: date, body: "existing transcript")
let client = MockSybilClient(
chatsResponse: [chat],
chatDetails: ["chat-pin": detail]
)
await client.setCompletionStreamEvents([
.delta(CompletionStreamDelta(text: "partial ")),
.delta(CompletionStreamDelta(text: "response")),
.done(CompletionStreamDone(text: "partial response"))
])
let viewModel = SybilViewModel(settings: testSettings(named: #function)) { _ in client }
viewModel.isAuthenticated = true
viewModel.isCheckingSession = false
viewModel.chats = [chat]
viewModel.workspaceItems = [WorkspaceItem(chat: chat)]
viewModel.selectedItem = .chat("chat-pin")
viewModel.selectedChat = detail
viewModel.composer = "continue"
let initialPinRequestID = viewModel.chatBottomPinRequestID
await viewModel.sendComposer()
let snapshot = await client.currentSnapshot()
#expect(snapshot.runCompletionStream == 1)
#expect(viewModel.chatBottomPinRequestID == initialPinRequestID + 1)
}
@MainActor
@Test func quickQuestionRunsNonPersistentCompletionStream() async throws {
let client = MockSybilClient()

9
ios/fastlane/Appfile
View File

@@ -1,9 +0,0 @@
require "dotenv"
Dotenv.load(File.expand_path("../.env", __dir__))
app_identifier(ENV.fetch("FASTLANE_APP_IDENTIFIER", "net.buzzert.sybil2"))
team_id(ENV.fetch("FASTLANE_TEAM_ID", "DQQH5H6GBD"))
apple_id(ENV["FASTLANE_USER"]) if ENV["FASTLANE_USER"].to_s.strip.length.positive?
itc_team_id(ENV["FASTLANE_ITC_TEAM_ID"]) if ENV["FASTLANE_ITC_TEAM_ID"].to_s.strip.length.positive?

177
ios/fastlane/Fastfile
View File

@@ -1,177 +0,0 @@
require "dotenv"
require "open3"
require "shellwords"
require "yaml"
Dotenv.load(File.expand_path("../.env", __dir__))
default_platform(:ios)
APP_IDENTIFIER = ENV.fetch("FASTLANE_APP_IDENTIFIER", "net.buzzert.sybil2")
TEAM_ID = ENV.fetch("FASTLANE_TEAM_ID", "DQQH5H6GBD")
APP_STORE_APPLE_ID = ENV.fetch("SYBIL_APP_STORE_APPLE_ID", "6759442828")
PROVIDER_PUBLIC_ID = ENV.fetch("SYBIL_PROVIDER_PUBLIC_ID", "c043d167-ad88-4036-84ea-76c223f1b1b2")
IOS_ROOT = File.expand_path("..", __dir__)
PROJECT_FILE = File.join(IOS_ROOT, "Sybil.xcodeproj")
PROJECT_SPEC = File.join(IOS_ROOT, "project.yml")
APP_SPEC = File.join(IOS_ROOT, "Apps/Sybil/project.yml")
SCHEME = "Sybil"
TARGET = "SybilApp"
def present?(value)
!value.to_s.strip.empty?
end
def capture(command)
stdout, stderr, status = Open3.capture3(command)
return stdout.strip if status.success?
UI.user_error!("Command failed: #{command}\n#{stderr.strip}")
end
def app_project_settings
YAML.safe_load(File.read(APP_SPEC)).fetch("targets").fetch(TARGET).fetch("settings").fetch("base")
end
def local_marketing_version
app_project_settings.fetch("MARKETING_VERSION").to_s
end
def local_build_number
app_project_settings.fetch("CURRENT_PROJECT_VERSION").to_i
end
def normalize_version_tag(tag)
version = tag.to_s.strip.sub(/\Av/, "")
unless version.match?(/\A\d+\.\d+(\.\d+)?\z/)
UI.user_error!("Release tag #{tag.inspect} must look like v1.10 or v1.10.0")
end
version
end
def release_version
tag = ENV["SYBIL_VERSION_TAG"]
tag = capture("git describe --tags --abbrev=0") unless present?(tag)
normalize_version_tag(tag)
end
def xcode_build_setting(key, value)
"#{key}=#{value.to_s.shellescape}"
end
def app_store_connect_key_options
key_id = ENV["APP_STORE_CONNECT_API_KEY_ID"]
issuer_id = ENV["APP_STORE_CONNECT_API_ISSUER_ID"]
return nil unless present?(key_id) && present?(issuer_id)
key_path = ENV["APP_STORE_CONNECT_API_KEY_PATH"]
key_content = ENV["APP_STORE_CONNECT_API_KEY_CONTENT"]
if present?(key_path)
{
key_id: key_id,
issuer_id: issuer_id,
key_filepath: key_path
}
elsif present?(key_content)
{
key_id: key_id,
issuer_id: issuer_id,
key_content: key_content,
is_key_content_base64: ENV["APP_STORE_CONNECT_API_KEY_CONTENT_BASE64"].to_s == "true"
}
end
end
platform :ios do
desc "Show the version Fastlane will stamp into the next TestFlight archive"
lane :version do
UI.message("Git tag version: #{release_version}")
UI.message("Checked-in app version: #{local_marketing_version}")
UI.message("Checked-in build number: #{local_build_number}")
end
desc "Build Sybil and upload it to TestFlight"
lane :beta do
version = release_version
build_number = ENV["SYBIL_BUILD_NUMBER"].to_s
api_key = nil
if app_store_connect_key_options
api_key = app_store_connect_api_key(app_store_connect_key_options)
end
unless present?(build_number)
build_number = (local_build_number + 1).to_s
if api_key
begin
latest = latest_testflight_build_number(
app_identifier: APP_IDENTIFIER,
version: version,
api_key: api_key,
initial_build_number: local_build_number
).to_i
build_number = [latest + 1, local_build_number + 1].max.to_s
rescue StandardError => e
UI.important("Could not look up TestFlight build number: #{e.message}")
UI.important("Using checked-in build number + 1: #{build_number}")
end
end
end
UI.user_error!("Build number must be a positive integer") unless build_number.match?(/\A[1-9]\d*\z/)
sh("xcodegen --spec #{PROJECT_SPEC.shellescape}")
xcode_args = [
"-allowProvisioningUpdates",
xcode_build_setting("MARKETING_VERSION", version),
xcode_build_setting("CURRENT_PROJECT_VERSION", build_number)
].join(" ")
ipa_path = build_app(
project: PROJECT_FILE,
scheme: SCHEME,
clean: true,
sdk: "iphoneos",
export_method: "app-store",
output_directory: File.join(IOS_ROOT, "build/fastlane"),
output_name: "Sybil-#{version}-#{build_number}.ipa",
xcargs: xcode_args,
export_xcargs: "-allowProvisioningUpdates",
export_options: {
method: "app-store-connect",
destination: "export",
signingStyle: "automatic",
teamID: TEAM_ID,
manageAppVersionAndBuildNumber: false,
uploadSymbols: true,
stripSwiftSymbols: true
}
)
ipa_path ||= lane_context[SharedValues::IPA_OUTPUT_PATH]
UI.user_error!("IPA export failed; no IPA path was returned") unless present?(ipa_path) && File.exist?(ipa_path)
password = ENV["FASTLANE_APPLE_APPLICATION_SPECIFIC_PASSWORD"]
UI.user_error!("FASTLANE_USER is required for altool upload") unless present?(ENV["FASTLANE_USER"])
UI.user_error!("FASTLANE_APPLE_APPLICATION_SPECIFIC_PASSWORD is required for altool upload") unless present?(password)
UI.user_error!("SYBIL_APP_STORE_APPLE_ID is required for altool upload") unless present?(APP_STORE_APPLE_ID)
UI.user_error!("SYBIL_PROVIDER_PUBLIC_ID is required for altool upload") unless present?(PROVIDER_PUBLIC_ID)
ENV["ITMS_TRANSPORTER_PASSWORD"] = password
sh([
"xcrun altool",
"--upload-package #{ipa_path.shellescape}",
"--platform ios",
"--apple-id #{APP_STORE_APPLE_ID.shellescape}",
"--bundle-id #{APP_IDENTIFIER.shellescape}",
"--bundle-version #{build_number.shellescape}",
"--bundle-short-version-string #{version.shellescape}",
"--provider-public-id #{PROVIDER_PUBLIC_ID.shellescape}",
"--username #{ENV.fetch("FASTLANE_USER").shellescape}",
"--password @env:ITMS_TRANSPORTER_PASSWORD",
"--show-progress"
].join(" "))
end
end

40
ios/fastlane/README.md
View File

@@ -1,40 +0,0 @@
fastlane documentation
----
# Installation
Make sure you have the latest version of the Xcode command line tools installed:
```sh
xcode-select --install
```
For _fastlane_ installation instructions, see [Installing _fastlane_](https://docs.fastlane.tools/#installing-fastlane)
# Available Actions
## iOS
### ios version
```sh
[bundle exec] fastlane ios version
```
Show the version Fastlane will stamp into the next TestFlight archive
### ios beta
```sh
[bundle exec] fastlane ios beta
```
Build Sybil and upload it to TestFlight
----
This README.md is auto-generated and will be re-generated every time [_fastlane_](https://fastlane.tools) is run.
More information about _fastlane_ can be found on [fastlane.tools](https://fastlane.tools).
The documentation of _fastlane_ can be found on [docs.fastlane.tools](https://docs.fastlane.tools).

12
ios/justfile
View File

@@ -5,10 +5,8 @@ derived_data := "build/DerivedData"
default:
@just build
generate:
xcodegen --spec project.yml
build: generate
build:
if [ ! -d "Sybil.xcodeproj" ]; then xcodegen --spec project.yml; fi
if command -v xcbeautify >/dev/null 2>&1; then \
xcodebuild -scheme Sybil -destination '{{simulator}}' | xcbeautify; \
else \
@@ -18,15 +16,13 @@ build: generate
test:
cd Packages/Sybil && xcodebuild test -scheme Sybil -destination '{{simulator}}' -parallel-testing-enabled NO
run: generate
run:
if [ ! -d "Sybil.xcodeproj" ]; then xcodegen --spec project.yml; fi
xcrun simctl boot '{{simulator_name}}' 2>/dev/null || true
xcodebuild -scheme Sybil -destination '{{simulator}}' -derivedDataPath '{{derived_data}}'
xcrun simctl install booted '{{derived_data}}/Build/Products/Debug-iphonesimulator/Sybil.app'
xcrun simctl launch booted net.buzzert.sybil2
beta:
fastlane ios beta
screenshot path="build/sybil-screenshot.png":
mkdir -p "$(dirname '{{path}}')"
xcrun simctl io booted screenshot '{{path}}'

44
server/prisma/migrations/20260529090000_add_project_starred_items/migration.sql
View File

@@ -1,44 +0,0 @@
-- CreateTable
CREATE TABLE "Project" (
"id" TEXT NOT NULL PRIMARY KEY,
"createdAt" DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
"updatedAt" DATETIME NOT NULL,
"kind" TEXT NOT NULL DEFAULT 'folder',
"title" TEXT NOT NULL,
"userId" TEXT,
CONSTRAINT "Project_userId_fkey" FOREIGN KEY ("userId") REFERENCES "User" ("id") ON DELETE CASCADE ON UPDATE CASCADE
);
-- CreateTable
CREATE TABLE "ProjectItem" (
"id" TEXT NOT NULL PRIMARY KEY,
"createdAt" DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
"projectId" TEXT NOT NULL,
"chatId" TEXT,
"searchId" TEXT,
CONSTRAINT "ProjectItem_projectId_fkey" FOREIGN KEY ("projectId") REFERENCES "Project" ("id") ON DELETE CASCADE ON UPDATE CASCADE,
CONSTRAINT "ProjectItem_chatId_fkey" FOREIGN KEY ("chatId") REFERENCES "Chat" ("id") ON DELETE CASCADE ON UPDATE CASCADE,
CONSTRAINT "ProjectItem_searchId_fkey" FOREIGN KEY ("searchId") REFERENCES "Search" ("id") ON DELETE CASCADE ON UPDATE CASCADE,
CONSTRAINT "ProjectItem_one_target_check" CHECK (("chatId" IS NOT NULL AND "searchId" IS NULL) OR ("chatId" IS NULL AND "searchId" IS NOT NULL))
);
-- CreateIndex
CREATE INDEX "Project_kind_idx" ON "Project"("kind");
-- CreateIndex
CREATE INDEX "Project_userId_idx" ON "Project"("userId");
-- CreateIndex
CREATE UNIQUE INDEX "ProjectItem_projectId_chatId_key" ON "ProjectItem"("projectId", "chatId");
-- CreateIndex
CREATE UNIQUE INDEX "ProjectItem_projectId_searchId_key" ON "ProjectItem"("projectId", "searchId");
-- CreateIndex
CREATE INDEX "ProjectItem_projectId_createdAt_idx" ON "ProjectItem"("projectId", "createdAt");
-- CreateIndex
CREATE INDEX "ProjectItem_chatId_idx" ON "ProjectItem"("chatId");
-- CreateIndex
CREATE INDEX "ProjectItem_searchId_idx" ON "ProjectItem"("searchId");

8
server/prisma/migrations/20260531000000_add_search_query_cache_key/migration.sql
View File

@@ -1,8 +0,0 @@
-- Add normalized search query lookup key for cache/reuse behavior.
ALTER TABLE "Search" ADD COLUMN "queryNormalized" TEXT;
UPDATE "Search"
SET "queryNormalized" = lower(trim("query"))
WHERE "query" IS NOT NULL AND trim("query") != '';
CREATE INDEX "Search_queryNormalized_updatedAt_idx" ON "Search"("queryNormalized", "updatedAt");

47
server/prisma/schema.prisma
View File

@@ -27,11 +27,6 @@ enum SearchSource {
exa
}
enum ProjectKind {
starred
folder
}
model User {
id String @id @default(cuid())
createdAt DateTime @default(now())
@@ -42,7 +37,6 @@ model User {
chats Chat[]
searches Search[]
projects Project[]
}
model Chat {
@@ -65,7 +59,6 @@ model Chat {
messages Message[]
calls LlmCall[]
projectItems ProjectItem[]
@@index([userId])
}
@@ -121,7 +114,6 @@ model Search {
title String?
query String?
queryNormalized String?
source SearchSource @default(exa)
@@ -140,10 +132,8 @@ model Search {
userId String?
results SearchResult[]
projectItems ProjectItem[]
@@index([updatedAt])
@@index([queryNormalized, updatedAt])
@@index([userId])
}
@@ -169,40 +159,3 @@ model SearchResult {
@@index([searchId, rank])
}
model Project {
id String @id @default(cuid())
createdAt DateTime @default(now())
updatedAt DateTime @updatedAt
kind ProjectKind @default(folder)
title String
user User? @relation(fields: [userId], references: [id], onDelete: Cascade)
userId String?
items ProjectItem[]
@@index([kind])
@@index([userId])
}
model ProjectItem {
id String @id @default(cuid())
createdAt DateTime @default(now())
project Project @relation(fields: [projectId], references: [id], onDelete: Cascade)
projectId String
chat Chat? @relation(fields: [chatId], references: [id], onDelete: Cascade)
chatId String?
search Search? @relation(fields: [searchId], references: [id], onDelete: Cascade)
searchId String?
@@unique([projectId, chatId])
@@unique([projectId, searchId])
@@index([projectId, createdAt])
@@index([chatId])
@@index([searchId])
}

26
server/src/browser-fetch-headers.ts
View File

@@ -1,26 +0,0 @@
export const CHROMIUM_USER_AGENT =
"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/130.0.0.0 Safari/537.36";
export const BROWSER_ACCEPT_LANGUAGE = "en-US,en;q=0.9";
export const FETCH_URL_ACCEPT =
"text/html,application/xhtml+xml,application/xml;q=0.9,application/pdf;q=0.9,*/*;q=0.8";
export function buildBrowserLikeRequestHeaders(accept: string): Record<string, string> {
return {
"User-Agent": CHROMIUM_USER_AGENT,
Accept: accept,
"Accept-Language": BROWSER_ACCEPT_LANGUAGE,
};
}
export function buildBrowserLikeNavigationHeaders(accept = FETCH_URL_ACCEPT): Record<string, string> {
return {
...buildBrowserLikeRequestHeaders(accept),
"Upgrade-Insecure-Requests": "1",
"Sec-Fetch-Dest": "document",
"Sec-Fetch-Mode": "navigate",
"Sec-Fetch-Site": "none",
"Sec-Fetch-User": "?1",
};
}

696
server/src/llm/chat-tools.ts
View File

@@ -4,14 +4,19 @@ import os from "node:os";
import path from "node:path";
import { promisify } from "node:util";
import { convert as htmlToText } from "html-to-text";
import type OpenAI from "openai";
import { z } from "zod";
import { buildBrowserLikeNavigationHeaders } from "../browser-fetch-headers.js";
import { env } from "../env.js";
import { exaClient } from "../search/exa.js";
import { searchSearxng } from "../search/searxng.js";
import {
buildOpenAIConversationMessage,
buildOpenAIResponsesInputMessage,
buildSystemPromptAugmentationMessage,
} from "./message-content.js";
import type { ChatMessage } from "./types.js";
export const MAX_TOOL_ROUNDS = env.CHAT_MAX_TOOL_ROUNDS;
const MAX_TOOL_ROUNDS = env.CHAT_MAX_TOOL_ROUNDS;
const DEFAULT_WEB_RESULTS = 5;
const MAX_WEB_RESULTS = 10;
const DEFAULT_FETCH_MAX_CHARACTERS = 12_000;
@@ -24,7 +29,7 @@ const MAX_SHELL_COMMAND_CHARACTERS = 20_000;
const DEFAULT_SHELL_MAX_OUTPUT_CHARACTERS = 24_000;
const MAX_SHELL_MAX_OUTPUT_CHARACTERS = 80_000;
const REMOTE_EXEC_MAX_BUFFER_BYTES = 1_000_000;
export const MAX_DANGLING_TOOL_INTENT_RETRIES = 1;
const MAX_DANGLING_TOOL_INTENT_RETRIES = 1;
const execFileAsync = promisify(execFile);
@@ -214,7 +219,7 @@ function getEnabledToolSet(params: Pick<ToolAwareCompletionParams, "enabledTools
return new Set(normalizeEnabledChatTools(params.enabledTools));
}
export function getEnabledChatTools(params: Pick<ToolAwareCompletionParams, "enabledTools">) {
function getEnabledChatTools(params: Pick<ToolAwareCompletionParams, "enabledTools">) {
const enabled = getEnabledToolSet(params);
return CHAT_TOOLS.filter((tool) => {
const name = getToolName(tool);
@@ -222,6 +227,19 @@ export function getEnabledChatTools(params: Pick<ToolAwareCompletionParams, "ena
});
}
function toResponsesChatTools(tools: any[]) {
return tools.map((tool) => {
if (tool?.type !== "function") return tool;
return {
type: "function",
name: tool.function.name,
description: tool.function.description,
parameters: tool.function.parameters,
strict: false,
};
});
}
export const CHAT_TOOL_SYSTEM_PROMPT =
"You can use tools to gather up-to-date web information when needed. " +
"Use web_search for discovery and recent facts, and fetch_url to read the full content of a specific page. " +
@@ -235,18 +253,18 @@ export const CHAT_TOOL_SYSTEM_PROMPT =
: "") +
"Do not fabricate tool outputs; reason only from provided tool results.";
export type ToolRunOutcome = {
type ToolRunOutcome = {
ok: boolean;
[key: string]: unknown;
};
export type ToolAwareUsage = {
type ToolAwareUsage = {
inputTokens?: number;
outputTokens?: number;
totalTokens?: number;
};
export type ToolAwareCompletionResult = {
type ToolAwareCompletionResult = {
text: string;
usage?: ToolAwareUsage;
raw: unknown;
@@ -258,8 +276,8 @@ export type ToolAwareStreamingEvent =
| { type: "tool_call"; event: ToolExecutionEvent }
| { type: "done"; result: ToolAwareCompletionResult };
export type ToolAwareCompletionParams = {
client: any;
type ToolAwareCompletionParams = {
client: OpenAI;
model: string;
messages: ChatMessage[];
enabledTools?: string[];
@@ -274,17 +292,15 @@ export type ToolAwareCompletionParams = {
};
};
export type ToolExecutionStatus = "initiated" | "completed" | "failed";
export type ToolExecutionEvent = {
toolCallId: string;
name: string;
status: ToolExecutionStatus;
status: "completed" | "failed";
summary: string;
args: Record<string, unknown>;
startedAt: string;
completedAt?: string;
durationMs?: number;
completedAt: string;
durationMs: number;
error?: string;
resultPreview?: string;
};
@@ -312,13 +328,10 @@ function toSingleLine(value: string, maxLength = 220) {
);
}
function buildToolSummary(name: string, args: Record<string, unknown>, status: ToolExecutionStatus, error?: string) {
function buildToolSummary(name: string, args: Record<string, unknown>, status: "completed" | "failed", error?: string) {
const errSuffix = status === "failed" && error ? ` Error: ${toSingleLine(error, 140)}` : "";
if (name === "web_search") {
const query = typeof args.query === "string" ? args.query.trim() : "";
if (status === "initiated") {
return query ? `Searching web for '${toSingleLine(query, 100)}'.` : "Searching web.";
}
if (status === "completed") {
return query ? `Performed web search for '${toSingleLine(query, 100)}'.` : "Performed web search.";
}
@@ -327,9 +340,6 @@ function buildToolSummary(name: string, args: Record<string, unknown>, status: T
if (name === "fetch_url") {
const url = typeof args.url === "string" ? args.url.trim() : "";
if (status === "initiated") {
return url ? `Fetching URL ${toSingleLine(url, 140)}.` : "Fetching URL.";
}
if (status === "completed") {
return url ? `Fetched URL ${toSingleLine(url, 140)}.` : "Fetched URL.";
}
@@ -338,9 +348,6 @@ function buildToolSummary(name: string, args: Record<string, unknown>, status: T
if (name === "codex_exec") {
const prompt = typeof args.prompt === "string" ? args.prompt.trim() : "";
if (status === "initiated") {
return prompt ? `Running Codex task: '${toSingleLine(prompt, 120)}'.` : "Running Codex task.";
}
if (status === "completed") {
return prompt ? `Ran Codex task: '${toSingleLine(prompt, 120)}'.` : "Ran Codex task.";
}
@@ -349,9 +356,6 @@ function buildToolSummary(name: string, args: Record<string, unknown>, status: T
if (name === "shell_exec") {
const command = typeof args.command === "string" ? args.command.trim() : "";
if (status === "initiated") {
return command ? `Running devbox shell command: '${toSingleLine(command, 120)}'.` : "Running devbox shell command.";
}
if (status === "completed") {
return command ? `Ran devbox shell command: '${toSingleLine(command, 120)}'.` : "Ran devbox shell command.";
}
@@ -360,9 +364,6 @@ function buildToolSummary(name: string, args: Record<string, unknown>, status: T
: `Devbox shell command failed.${errSuffix}`;
}
if (status === "initiated") {
return `Running tool '${name}'.`;
}
if (status === "completed") {
return `Ran tool '${name}'.`;
}
@@ -421,7 +422,7 @@ function extractHtmlTitle(html: string) {
);
}
export function buildChatToolSystemPrompt(params: Pick<ToolAwareCompletionParams, "enabledTools">) {
function buildChatToolSystemPrompt(params: Pick<ToolAwareCompletionParams, "enabledTools">) {
const enabled = getEnabledToolSet(params);
return (
"You can use tools to gather up-to-date web information when needed. " +
@@ -439,6 +440,22 @@ export function buildChatToolSystemPrompt(params: Pick<ToolAwareCompletionParams
);
}
function normalizeIncomingMessages(messages: ChatMessage[], userLocation?: string, params: Pick<ToolAwareCompletionParams, "enabledTools"> = {}) {
const normalized = messages.map((message) => buildOpenAIConversationMessage(message));
return [{ role: "system", content: buildChatToolSystemPrompt(params) }, buildSystemPromptAugmentationMessage(userLocation), ...normalized];
}
function normalizePlainIncomingMessages(messages: ChatMessage[], userLocation?: string) {
return [buildSystemPromptAugmentationMessage(userLocation), ...messages.map((message) => buildOpenAIConversationMessage(message))];
}
function normalizeIncomingResponsesInput(messages: ChatMessage[], userLocation?: string, params: Pick<ToolAwareCompletionParams, "enabledTools"> = {}) {
const normalized = messages.map((message) => buildOpenAIResponsesInputMessage(message));
return [{ role: "system", content: buildChatToolSystemPrompt(params) }, buildSystemPromptAugmentationMessage(userLocation), ...normalized];
}
async function runExaWebSearchTool(args: WebSearchArgs): Promise<ToolRunOutcome> {
const exa = exaClient();
const response = await exa.search(args.query, {
@@ -536,7 +553,10 @@ async function runFetchUrlTool(input: unknown): Promise<ToolRunOutcome> {
response = await fetch(parsed.toString(), {
redirect: "follow",
signal: controller.signal,
headers: buildBrowserLikeNavigationHeaders(),
headers: {
"User-Agent": "SybilBot/1.0 (+https://sybil.local)",
Accept: "text/html, text/plain, application/json;q=0.9, */*;q=0.5",
},
});
} finally {
clearTimeout(timeout);
@@ -807,7 +827,7 @@ async function executeTool(name: string, args: unknown): Promise<ToolRunOutcome>
return { ok: false, error: `Unknown tool: ${name}` };
}
export function parseToolArgs(raw: unknown) {
function parseToolArgs(raw: unknown) {
if (typeof raw !== "string") return {};
const trimmed = raw.trim();
if (!trimmed) return {};
@@ -836,7 +856,7 @@ function buildEventArgs(name: string, args: Record<string, unknown>) {
return args;
}
export function looksLikeDanglingToolIntent(text: string) {
function looksLikeDanglingToolIntent(text: string) {
const normalized = text
.toLowerCase()
.replace(/[`*_>#-]/g, " ")
@@ -852,7 +872,7 @@ export function looksLikeDanglingToolIntent(text: string) {
);
}
export function appendDanglingToolIntentCorrection(conversation: any[], text: string) {
function appendDanglingToolIntentCorrection(conversation: any[], text: string) {
conversation.push({ role: "assistant", content: text });
conversation.push({
role: "system",
@@ -861,7 +881,7 @@ export function appendDanglingToolIntentCorrection(conversation: any[], text: st
});
}
export function mergeUsage(acc: Required<ToolAwareUsage>, usage: any) {
function mergeUsage(acc: Required<ToolAwareUsage>, usage: any) {
if (!usage) return false;
acc.inputTokens += usage.prompt_tokens ?? 0;
acc.outputTokens += usage.completion_tokens ?? 0;
@@ -869,19 +889,79 @@ export function mergeUsage(acc: Required<ToolAwareUsage>, usage: any) {
return true;
}
export function getUnstreamedText(finalText: string, streamedText: string) {
function mergeResponsesUsage(acc: Required<ToolAwareUsage>, usage: any) {
if (!usage) return false;
acc.inputTokens += usage.input_tokens ?? 0;
acc.outputTokens += usage.output_tokens ?? 0;
acc.totalTokens += usage.total_tokens ?? 0;
return true;
}
function getResponseOutputItems(response: any) {
return Array.isArray(response?.output) ? response.output : [];
}
function extractResponsesText(response: any, fallback = "") {
if (typeof response?.output_text === "string") return response.output_text;
const parts: string[] = [];
for (const item of getResponseOutputItems(response)) {
if (item?.type !== "message" || !Array.isArray(item.content)) continue;
for (const content of item.content) {
if (content?.type === "output_text" && typeof content.text === "string") {
parts.push(content.text);
} else if (content?.type === "refusal" && typeof content.refusal === "string") {
parts.push(content.refusal);
}
}
}
return parts.join("") || fallback;
}
function extractChatCompletionContent(message: any) {
if (typeof message?.content === "string") return message.content;
if (!Array.isArray(message?.content)) return "";
return message.content
.map((part: any) => {
if (typeof part === "string") return part;
if (typeof part?.text === "string") return part.text;
if (typeof part?.content === "string") return part.content;
return "";
})
.join("");
}
function getUnstreamedText(finalText: string, streamedText: string) {
if (!finalText) return "";
if (!streamedText) return finalText;
return finalText.startsWith(streamedText) ? finalText.slice(streamedText.length) : "";
}
export type NormalizedToolCall = {
function getResponseFailureMessage(response: any) {
if (response?.status !== "failed" && response?.status !== "incomplete") return null;
const errorMessage = typeof response?.error?.message === "string" ? response.error.message : null;
const incompleteReason = typeof response?.incomplete_details?.reason === "string" ? response.incomplete_details.reason : null;
return errorMessage ?? (incompleteReason ? `Response incomplete: ${incompleteReason}` : `Response ${response.status}.`);
}
function normalizeResponsesToolCalls(outputItems: any[], round: number): NormalizedToolCall[] {
return outputItems
.filter((item) => item?.type === "function_call")
.map((call: any, index: number) => ({
id: call.call_id ?? call.id ?? `tool_call_${round}_${index}`,
name: call.name ?? "unknown_tool",
arguments: call.arguments ?? "{}",
}));
}
type NormalizedToolCall = {
id: string;
name: string;
arguments: string;
};
export function normalizeModelToolCalls(toolCalls: any[], round: number): NormalizedToolCall[] {
function normalizeModelToolCalls(toolCalls: any[], round: number): NormalizedToolCall[] {
return toolCalls.map((call: any, index: number) => ({
id: call?.id ?? `tool_call_${round}_${index}`,
name: call?.function?.name ?? "unknown_tool",
@@ -889,55 +969,17 @@ export function normalizeModelToolCalls(toolCalls: any[], round: number): Normal
}));
}
export type PreparedToolCallExecution = {
startedAtMs: number;
startedAt: string;
parsedArgs: Record<string, unknown>;
eventArgs: Record<string, unknown>;
parseError?: unknown;
};
export function prepareToolCallExecution(call: NormalizedToolCall): { event: ToolExecutionEvent; execution: PreparedToolCallExecution } {
const startedAtMs = Date.now();
const startedAt = new Date(startedAtMs).toISOString();
let parsedArgs: Record<string, unknown> = {};
let parseError: unknown;
try {
parsedArgs = toRecord(parseToolArgs(call.arguments));
} catch (err) {
parseError = err;
}
const eventArgs = buildEventArgs(call.name, parsedArgs);
return {
event: {
toolCallId: call.id,
name: call.name,
status: "initiated",
summary: buildToolSummary(call.name, eventArgs, "initiated"),
args: eventArgs,
startedAt,
},
execution: {
startedAtMs,
startedAt,
parsedArgs,
eventArgs,
parseError,
},
};
}
export async function executeToolCallAndBuildEvent(
async function executeToolCallAndBuildEvent(
call: NormalizedToolCall,
execution: PreparedToolCallExecution,
params: ToolAwareCompletionParams
): Promise<{ event: ToolExecutionEvent; toolResult: ToolRunOutcome }> {
const startedAtMs = Date.now();
const startedAt = new Date(startedAtMs).toISOString();
let toolResult: ToolRunOutcome;
let parsedArgs: Record<string, unknown> = {};
try {
if (execution.parseError) throw execution.parseError;
toolResult = await executeTool(call.name, execution.parsedArgs);
parsedArgs = toRecord(parseToolArgs(call.arguments));
toolResult = await executeTool(call.name, parsedArgs);
} catch (err: any) {
toolResult = {
ok: false,
@@ -954,15 +996,16 @@ export async function executeToolCallAndBuildEvent(
: undefined;
const completedAtMs = Date.now();
const eventArgs = buildEventArgs(call.name, parsedArgs);
const event: ToolExecutionEvent = {
toolCallId: call.id,
name: call.name,
status,
summary: buildToolSummary(call.name, execution.eventArgs, status, error),
args: execution.eventArgs,
startedAt: execution.startedAt,
summary: buildToolSummary(call.name, eventArgs, status, error),
args: eventArgs,
startedAt,
completedAt: new Date(completedAtMs).toISOString(),
durationMs: completedAtMs - execution.startedAtMs,
durationMs: completedAtMs - startedAtMs,
error,
resultPreview: buildResultPreview(toolResult),
};
@@ -973,3 +1016,482 @@ export async function executeToolCallAndBuildEvent(
return { event, toolResult };
}
export async function runToolAwareOpenAIChat(params: ToolAwareCompletionParams): Promise<ToolAwareCompletionResult> {
const enabledTools = getEnabledChatTools(params);
const input: any[] = normalizeIncomingResponsesInput(params.messages, params.userLocation, params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const response = await params.client.responses.create({
model: params.model,
input,
temperature: params.temperature,
max_output_tokens: params.maxTokens,
tools: toResponsesChatTools(enabledTools),
tool_choice: "auto",
parallel_tool_calls: true,
// Tool loops pass response output items back as input; reasoning items need persistence.
store: true,
} as any);
rawResponses.push(response);
sawUsage = mergeResponsesUsage(usageAcc, response?.usage) || sawUsage;
const failureMessage = getResponseFailureMessage(response);
if (failureMessage) {
throw new Error(failureMessage);
}
const outputItems = getResponseOutputItems(response);
const normalizedToolCalls = normalizeResponsesToolCalls(outputItems, round);
if (!normalizedToolCalls.length) {
const text = extractResponsesText(response);
if (danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES && looksLikeDanglingToolIntent(text)) {
danglingToolIntentRetries += 1;
appendDanglingToolIntentCorrection(input, text);
continue;
}
return {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, api: "responses" },
toolEvents,
};
}
totalToolCalls += normalizedToolCalls.length;
input.push(...outputItems);
for (const call of normalizedToolCalls) {
const { event, toolResult } = await executeToolCallAndBuildEvent(call, params);
toolEvents.push(event);
input.push({
type: "function_call_output",
call_id: call.id,
output: JSON.stringify(toolResult),
});
}
}
return {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true, api: "responses" },
toolEvents,
};
}
export async function runToolAwareChatCompletions(params: ToolAwareCompletionParams): Promise<ToolAwareCompletionResult> {
const enabledTools = getEnabledChatTools(params);
const conversation: any[] = normalizeIncomingMessages(params.messages, params.userLocation, params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const completion = await params.client.chat.completions.create({
model: params.model,
messages: conversation,
temperature: params.temperature,
max_tokens: params.maxTokens,
tools: enabledTools,
tool_choice: "auto",
} as any);
rawResponses.push(completion);
sawUsage = mergeUsage(usageAcc, completion?.usage) || sawUsage;
const message = completion?.choices?.[0]?.message;
if (!message) {
return {
text: "",
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, missingMessage: true },
toolEvents,
};
}
const toolCalls = Array.isArray(message.tool_calls) ? message.tool_calls : [];
if (!toolCalls.length) {
const text = typeof message.content === "string" ? message.content : "";
if (danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES && looksLikeDanglingToolIntent(text)) {
danglingToolIntentRetries += 1;
appendDanglingToolIntentCorrection(conversation, text);
continue;
}
return {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls },
toolEvents,
};
}
const normalizedToolCalls = normalizeModelToolCalls(toolCalls, round);
totalToolCalls += normalizedToolCalls.length;
const assistantToolCallMessage: any = {
role: "assistant",
tool_calls: normalizedToolCalls.map((call) => ({
id: call.id,
type: "function",
function: {
name: call.name,
arguments: call.arguments,
},
})),
};
if (typeof message.content === "string" && message.content.length) {
assistantToolCallMessage.content = message.content;
}
conversation.push(assistantToolCallMessage);
for (const call of normalizedToolCalls) {
const { event, toolResult } = await executeToolCallAndBuildEvent(call, params);
toolEvents.push(event);
conversation.push({
role: "tool",
tool_call_id: call.id,
content: JSON.stringify(toolResult),
});
}
}
return {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true },
toolEvents,
};
}
export async function runPlainChatCompletions(params: ToolAwareCompletionParams): Promise<ToolAwareCompletionResult> {
const completion = await params.client.chat.completions.create({
model: params.model,
messages: normalizePlainIncomingMessages(params.messages, params.userLocation),
temperature: params.temperature,
max_tokens: params.maxTokens,
} as any);
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
const sawUsage = mergeUsage(usageAcc, completion?.usage);
const message = completion?.choices?.[0]?.message;
return {
text: extractChatCompletionContent(message),
usage: sawUsage ? usageAcc : undefined,
raw: { response: completion, api: "chat.completions" },
toolEvents: [],
};
}
export async function* runToolAwareOpenAIChatStream(
params: ToolAwareCompletionParams
): AsyncGenerator<ToolAwareStreamingEvent> {
const enabledTools = getEnabledChatTools(params);
const input: any[] = normalizeIncomingResponsesInput(params.messages, params.userLocation, params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const stream = await params.client.responses.create({
model: params.model,
input,
temperature: params.temperature,
max_output_tokens: params.maxTokens,
tools: toResponsesChatTools(enabledTools),
tool_choice: "auto",
parallel_tool_calls: true,
// Tool loops pass response output items back as input; reasoning items need persistence.
store: true,
stream: true,
} as any);
let roundText = "";
let streamedRoundText = "";
let roundHasToolCalls = false;
let canStreamRoundText = false;
let completedResponse: any | null = null;
const completedOutputItems: any[] = [];
for await (const event of stream as any as AsyncIterable<any>) {
rawResponses.push(event);
if (event?.type === "response.output_text.delta" && typeof event.delta === "string") {
roundText += event.delta;
if (canStreamRoundText && !roundHasToolCalls && event.delta.length) {
streamedRoundText += event.delta;
yield { type: "delta", text: event.delta };
}
} else if (event?.type === "response.output_item.added" && event.item) {
if (event.item.type === "function_call") {
roundHasToolCalls = true;
canStreamRoundText = false;
} else if (event.item.type === "message" && !roundHasToolCalls) {
canStreamRoundText = true;
}
} else if (event?.type === "response.output_item.done" && event.item) {
completedOutputItems[event.output_index ?? completedOutputItems.length] = event.item;
if (event.item.type === "function_call") {
roundHasToolCalls = true;
canStreamRoundText = false;
}
} else if (event?.type === "response.completed") {
completedResponse = event.response;
sawUsage = mergeResponsesUsage(usageAcc, event.response?.usage) || sawUsage;
} else if (event?.type === "response.failed" || event?.type === "response.incomplete") {
completedResponse = event.response;
sawUsage = mergeResponsesUsage(usageAcc, event.response?.usage) || sawUsage;
} else if (event?.type === "error") {
throw new Error(event.message ?? "OpenAI Responses stream failed.");
}
}
const failureMessage = getResponseFailureMessage(completedResponse);
if (failureMessage) {
throw new Error(failureMessage);
}
const outputItems = getResponseOutputItems(completedResponse);
const responseOutputItems = outputItems.length ? outputItems : completedOutputItems.filter(Boolean);
const normalizedToolCalls = normalizeResponsesToolCalls(responseOutputItems, round);
if (!normalizedToolCalls.length) {
const text = extractResponsesText(completedResponse, roundText);
if (
!streamedRoundText &&
danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES &&
looksLikeDanglingToolIntent(text)
) {
danglingToolIntentRetries += 1;
appendDanglingToolIntentCorrection(input, text);
continue;
}
const unstreamedText = getUnstreamedText(text, streamedRoundText);
if (unstreamedText) {
yield { type: "delta", text: unstreamedText };
}
yield {
type: "done",
result: {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls, api: "responses" },
toolEvents,
},
};
return;
}
totalToolCalls += normalizedToolCalls.length;
input.push(...responseOutputItems);
for (const call of normalizedToolCalls) {
const { event, toolResult } = await executeToolCallAndBuildEvent(call, params);
toolEvents.push(event);
yield { type: "tool_call", event };
input.push({
type: "function_call_output",
call_id: call.id,
output: JSON.stringify(toolResult),
});
}
}
yield {
type: "done",
result: {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true, api: "responses" },
toolEvents,
},
};
}
export async function* runToolAwareChatCompletionsStream(
params: ToolAwareCompletionParams
): AsyncGenerator<ToolAwareStreamingEvent> {
const enabledTools = getEnabledChatTools(params);
const conversation: any[] = normalizeIncomingMessages(params.messages, params.userLocation, params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const stream = await params.client.chat.completions.create({
model: params.model,
messages: conversation,
temperature: params.temperature,
max_tokens: params.maxTokens,
tools: enabledTools,
tool_choice: "auto",
stream: true,
stream_options: { include_usage: true },
} as any);
let roundText = "";
let streamedRoundText = "";
let roundHasToolCalls = false;
const roundToolCalls = new Map<number, { id?: string; name?: string; arguments: string }>();
for await (const chunk of stream as any as AsyncIterable<any>) {
rawResponses.push(chunk);
sawUsage = mergeUsage(usageAcc, chunk?.usage) || sawUsage;
const choice = chunk?.choices?.[0];
const deltaText = choice?.delta?.content ?? "";
if (typeof deltaText === "string" && deltaText.length) {
roundText += deltaText;
if (!roundHasToolCalls) {
streamedRoundText += deltaText;
yield { type: "delta", text: deltaText };
}
}
const deltaToolCalls = Array.isArray(choice?.delta?.tool_calls) ? choice.delta.tool_calls : [];
if (deltaToolCalls.length) {
roundHasToolCalls = true;
}
for (const toolCall of deltaToolCalls) {
const idx = typeof toolCall?.index === "number" ? toolCall.index : 0;
const entry = roundToolCalls.get(idx) ?? { arguments: "" };
if (typeof toolCall?.id === "string" && toolCall.id.length) {
entry.id = toolCall.id;
}
if (typeof toolCall?.function?.name === "string" && toolCall.function.name.length) {
entry.name = toolCall.function.name;
}
if (typeof toolCall?.function?.arguments === "string" && toolCall.function.arguments.length) {
entry.arguments += toolCall.function.arguments;
}
roundToolCalls.set(idx, entry);
}
}
const normalizedToolCalls: NormalizedToolCall[] = [...roundToolCalls.entries()]
.sort((a, b) => a[0] - b[0])
.map(([_, call], index) => ({
id: call.id ?? `tool_call_${round}_${index}`,
name: call.name ?? "unknown_tool",
arguments: call.arguments || "{}",
}));
if (!normalizedToolCalls.length) {
if (
!streamedRoundText &&
danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES &&
looksLikeDanglingToolIntent(roundText)
) {
danglingToolIntentRetries += 1;
appendDanglingToolIntentCorrection(conversation, roundText);
continue;
}
const unstreamedText = getUnstreamedText(roundText, streamedRoundText);
if (unstreamedText) {
yield { type: "delta", text: unstreamedText };
}
yield {
type: "done",
result: {
text: roundText,
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls },
toolEvents,
},
};
return;
}
totalToolCalls += normalizedToolCalls.length;
const assistantToolCallMessage: any = {
role: "assistant",
tool_calls: normalizedToolCalls.map((call) => ({
id: call.id,
type: "function",
function: {
name: call.name,
arguments: call.arguments,
},
})),
};
if (roundText) {
assistantToolCallMessage.content = roundText;
}
conversation.push(assistantToolCallMessage);
for (const call of normalizedToolCalls) {
const { event, toolResult } = await executeToolCallAndBuildEvent(call, params);
toolEvents.push(event);
yield { type: "tool_call", event };
conversation.push({
role: "tool",
tool_call_id: call.id,
content: JSON.stringify(toolResult),
});
}
}
yield {
type: "done",
result: {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true },
toolEvents,
},
};
}
export async function* runPlainChatCompletionsStream(
params: ToolAwareCompletionParams
): AsyncGenerator<ToolAwareStreamingEvent> {
const rawResponses: unknown[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let text = "";
const stream = await params.client.chat.completions.create({
model: params.model,
messages: normalizePlainIncomingMessages(params.messages, params.userLocation),
temperature: params.temperature,
max_tokens: params.maxTokens,
stream: true,
} as any);
for await (const chunk of stream as any as AsyncIterable<any>) {
rawResponses.push(chunk);
sawUsage = mergeUsage(usageAcc, chunk?.usage) || sawUsage;
const deltaText = chunk?.choices?.[0]?.delta?.content ?? "";
if (typeof deltaText === "string" && deltaText.length) {
text += deltaText;
yield { type: "delta", text: deltaText };
}
}
yield {
type: "done",
result: {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, api: "chat.completions" },
toolEvents: [],
},
};
}

189
server/src/llm/message-content.ts
View File

@@ -18,21 +18,21 @@ function escapeAttribute(value: string) {
return value.replace(/"/g, "&quot;");
}
export function getImageAttachments(message: ChatMessage) {
function getImageAttachments(message: ChatMessage) {
return (message.attachments ?? []).filter((attachment): attachment is ChatImageAttachment => attachment.kind === "image");
}
export function getTextAttachments(message: ChatMessage) {
function getTextAttachments(message: ChatMessage) {
return (message.attachments ?? []).filter((attachment): attachment is ChatTextAttachment => attachment.kind === "text");
}
export function buildImageSummaryText(attachments: ChatImageAttachment[]) {
function buildImageSummaryText(attachments: ChatImageAttachment[]) {
if (!attachments.length) return null;
const label = attachments.length === 1 ? "Attached image" : "Attached images";
return `${label}: ${attachments.map((attachment) => attachment.filename).join(", ")}.`;
}
export function buildTextAttachmentPrompt(attachment: ChatTextAttachment) {
function buildTextAttachmentPrompt(attachment: ChatTextAttachment) {
const truncationNote = attachment.truncated ? ' truncated="true"' : "";
return [
`Attached text file: ${attachment.filename}${attachment.truncated ? " (content truncated)" : ""}`,
@@ -42,7 +42,83 @@ export function buildTextAttachmentPrompt(attachment: ChatTextAttachment) {
].join("\n");
}
export function parseImageDataUrl(attachment: ChatImageAttachment) {
function toOpenAIContent(message: ChatMessage) {
const imageAttachments = getImageAttachments(message);
const textAttachments = getTextAttachments(message);
if (!imageAttachments.length && !textAttachments.length) {
return message.content;
}
const parts: Array<Record<string, unknown>> = [];
for (const attachment of imageAttachments) {
parts.push({
type: "image_url",
image_url: {
url: attachment.dataUrl,
detail: "auto",
},
});
}
const imageSummary = buildImageSummaryText(imageAttachments);
if (imageSummary) {
parts.push({ type: "text", text: imageSummary });
}
for (const attachment of textAttachments) {
parts.push({ type: "text", text: buildTextAttachmentPrompt(attachment) });
}
if (message.content.trim()) {
parts.push({ type: "text", text: message.content });
}
if (parts.length === 1 && parts[0]?.type === "text" && typeof parts[0].text === "string") {
return parts[0].text;
}
return parts;
}
function toOpenAIResponsesContent(message: ChatMessage) {
const imageAttachments = getImageAttachments(message);
const textAttachments = getTextAttachments(message);
if (!imageAttachments.length && !textAttachments.length) {
return message.content;
}
const parts: Array<Record<string, unknown>> = [];
for (const attachment of imageAttachments) {
parts.push({
type: "input_image",
image_url: attachment.dataUrl,
detail: "auto",
});
}
const imageSummary = buildImageSummaryText(imageAttachments);
if (imageSummary) {
parts.push({ type: "input_text", text: imageSummary });
}
for (const attachment of textAttachments) {
parts.push({ type: "input_text", text: buildTextAttachmentPrompt(attachment) });
}
if (message.content.trim()) {
parts.push({ type: "input_text", text: message.content });
}
if (parts.length === 1 && parts[0]?.type === "input_text" && typeof parts[0].text === "string") {
return parts[0].text;
}
return parts;
}
function parseImageDataUrl(attachment: ChatImageAttachment) {
const match = attachment.dataUrl.match(/^data:(image\/(?:png|jpeg));base64,([a-z0-9+/=\s]+)$/i);
if (!match) {
throw new Error(`Invalid image attachment data URL for '${attachment.filename}'.`);
@@ -59,6 +135,83 @@ export function parseImageDataUrl(attachment: ChatImageAttachment) {
};
}
function toAnthropicContent(message: ChatMessage) {
const imageAttachments = getImageAttachments(message);
const textAttachments = getTextAttachments(message);
if (!imageAttachments.length && !textAttachments.length) {
return message.content;
}
const blocks: Array<Record<string, unknown>> = [];
for (const attachment of imageAttachments) {
const source = parseImageDataUrl(attachment);
blocks.push({
type: "image",
source: {
type: "base64",
media_type: source.mediaType,
data: source.data,
},
});
}
const imageSummary = buildImageSummaryText(imageAttachments);
if (imageSummary) {
blocks.push({ type: "text", text: imageSummary });
}
for (const attachment of textAttachments) {
blocks.push({ type: "text", text: buildTextAttachmentPrompt(attachment) });
}
if (message.content.trim()) {
blocks.push({ type: "text", text: message.content });
}
if (blocks.length === 1 && blocks[0]?.type === "text" && typeof blocks[0].text === "string") {
return blocks[0].text;
}
return blocks;
}
export function buildOpenAIConversationMessage(message: ChatMessage) {
if (message.role === "tool") {
const name = message.name?.trim() || "tool";
return {
role: "user",
content: `Tool output (${name}):\n${message.content}`,
};
}
const out: Record<string, unknown> = {
role: message.role,
content: toOpenAIContent(message),
};
if (message.name && (message.role === "assistant" || message.role === "user")) {
out.name = message.name;
}
return out;
}
export function buildOpenAIResponsesInputMessage(message: ChatMessage) {
if (message.role === "tool") {
const name = message.name?.trim() || "tool";
return {
role: "user",
content: `Tool output (${name}):\n${message.content}`,
};
}
return {
role: message.role,
content: toOpenAIResponsesContent(message),
};
}
export function buildSystemPromptAugmentationMessage(userLocation?: string) {
return {
role: "system",
@@ -66,12 +219,34 @@ export function buildSystemPromptAugmentationMessage(userLocation?: string) {
};
}
export function buildTopLevelSystemPrompt(messages: ChatMessage[], userLocation?: string, toolSystemPrompt?: string) {
return [toolSystemPrompt, buildSystemPromptAugmentation(userLocation), messages.find((message) => message.role === "system")?.content]
const ANTHROPIC_NO_SERVER_TOOLS_PROMPT =
"This Anthropic backend path does not have server-managed tool calls. Do not claim to run shell commands, Codex tasks, web searches, or fetch URLs. If the user asks for tool execution, explain that they should switch to OpenAI or xAI in this app for tool-enabled chat.";
export function getAnthropicSystemPrompt(messages: ChatMessage[], userLocation?: string) {
return [ANTHROPIC_NO_SERVER_TOOLS_PROMPT, buildSystemPromptAugmentation(userLocation), messages.find((message) => message.role === "system")?.content]
.filter(Boolean)
.join("\n\n");
}
export function buildAnthropicConversationMessage(message: ChatMessage) {
if (message.role === "system") {
throw new Error("System messages must be handled separately for Anthropic.");
}
if (message.role === "tool") {
const name = message.name?.trim() || "tool";
return {
role: "user",
content: `Tool output (${name}):\n${message.content}`,
};
}
return {
role: message.role === "assistant" ? "assistant" : "user",
content: toAnthropicContent(message),
};
}
export function buildComparableAttachments(input: unknown): ChatAttachment[] {
if (!Array.isArray(input)) return [];

61
server/src/llm/model-catalog.ts
View File

@@ -1,9 +1,6 @@
import type { FastifyBaseLogger } from "fastify";
import {
fetchProviderCatalogModels,
getProviderCatalogFallbackModels,
listModelCatalogProviders,
} from "./provider-adapters.js";
import { env } from "../env.js";
import { anthropicClient, hermesAgentClient, isHermesAgentConfigured, openaiClient, xaiClient } from "./providers.js";
import type { Provider } from "./types.js";
export type ProviderModelSnapshot = {
@@ -14,13 +11,35 @@ export type ProviderModelSnapshot = {
export type ModelCatalogSnapshot = Partial<Record<Provider, ProviderModelSnapshot>>;
const baseProviders: Provider[] = ["openai", "anthropic", "xai"];
const MODEL_FETCH_TIMEOUT_MS = 15000;
const MODEL_CATALOG_REFRESH_INTERVAL_MS = 24 * 60 * 60 * 1000;
const modelCatalog: ModelCatalogSnapshot = {};
const modelCatalog: ModelCatalogSnapshot = {
openai: { models: [], loadedAt: null, error: null },
anthropic: { models: [], loadedAt: null, error: null },
xai: { models: [], loadedAt: null, error: null },
};
let catalogRefreshPromise: Promise<void> | null = null;
function getCatalogProviders(): Provider[] {
return isHermesAgentConfigured() ? [...baseProviders, "hermes-agent"] : baseProviders;
}
function uniqSorted(models: string[]) {
return [...new Set(models.map((value) => value.trim()).filter(Boolean))].sort((a, b) => a.localeCompare(b));
}
function isLikelyOpenAIResponsesModel(model: string) {
const id = model.toLowerCase();
if (id.includes("embedding") || id.includes("moderation")) return false;
if (id.includes("audio") || id.includes("realtime") || id.includes("transcribe") || id.includes("tts")) return false;
if (id.includes("image") || id.includes("dall-e") || id.includes("sora")) return false;
if (id.includes("search") || id.includes("computer-use")) return false;
return /^(gpt-|o\d|chatgpt-)/.test(id);
}
async function withTimeout<T>(promise: Promise<T>, timeoutMs: number, label: string) {
let timeoutId: NodeJS.Timeout | null = null;
try {
@@ -37,9 +56,31 @@ async function withTimeout<T>(promise: Promise<T>, timeoutMs: number, label: str
}
}
async function fetchProviderModels(provider: Provider) {
if (provider === "openai") {
const page = await openaiClient().models.list();
return uniqSorted(page.data.map((model) => model.id).filter(isLikelyOpenAIResponsesModel));
}
if (provider === "anthropic") {
const page = await anthropicClient().models.list({ limit: 200 });
return uniqSorted(page.data.map((model) => model.id));
}
if (provider === "xai") {
const page = await xaiClient().models.list();
return uniqSorted(page.data.map((model) => model.id));
}
const page = await hermesAgentClient().models.list();
const models = page.data.map((model) => model.id);
if (env.HERMES_AGENT_MODEL) models.push(env.HERMES_AGENT_MODEL);
return uniqSorted(models);
}
async function refreshProviderModels(provider: Provider, logger?: FastifyBaseLogger) {
try {
const models = await withTimeout(fetchProviderCatalogModels(provider), MODEL_FETCH_TIMEOUT_MS, `${provider} model fetch`);
const models = await withTimeout(fetchProviderModels(provider), MODEL_FETCH_TIMEOUT_MS, `${provider} model fetch`);
modelCatalog[provider] = {
models,
loadedAt: new Date().toISOString(),
@@ -49,7 +90,7 @@ async function refreshProviderModels(provider: Provider, logger?: FastifyBaseLog
} catch (err: any) {
const message = err?.message ?? String(err);
const previous = modelCatalog[provider];
const fallbackModels = getProviderCatalogFallbackModels(provider);
const fallbackModels = provider === "hermes-agent" && env.HERMES_AGENT_MODEL ? [env.HERMES_AGENT_MODEL] : [];
modelCatalog[provider] = {
models: previous?.models.length ? previous.models : fallbackModels,
loadedAt: previous?.loadedAt ?? null,
@@ -62,7 +103,7 @@ async function refreshProviderModels(provider: Provider, logger?: FastifyBaseLog
export async function refreshModelCatalog(logger?: FastifyBaseLogger) {
if (catalogRefreshPromise) return catalogRefreshPromise;
catalogRefreshPromise = Promise.all(listModelCatalogProviders().map((provider) => refreshProviderModels(provider, logger)))
catalogRefreshPromise = Promise.all(getCatalogProviders().map((provider) => refreshProviderModels(provider, logger)))
.then(() => undefined)
.finally(() => {
catalogRefreshPromise = null;
@@ -88,7 +129,7 @@ export function startModelCatalogRefreshLoop(logger?: FastifyBaseLogger) {
export function getModelCatalogSnapshot(): ModelCatalogSnapshot {
const snapshot: ModelCatalogSnapshot = {};
for (const provider of listModelCatalogProviders()) {
for (const provider of getCatalogProviders()) {
const entry = modelCatalog[provider] ?? { models: [], loadedAt: null, error: null };
snapshot[provider] = {
models: [...entry.models],

84
server/src/llm/multiplexer.ts
View File

@@ -1,7 +1,8 @@
import { performance } from "node:perf_hooks";
import { prisma } from "../db.js";
import { buildToolLogMessageData } from "./chat-tools.js";
import { getProviderChatAdapter } from "./provider-adapters.js";
import { anthropicClient, hermesAgentClient, openaiClient, xaiClient } from "./providers.js";
import { buildToolLogMessageData, normalizeEnabledChatTools, runPlainChatCompletions, runToolAwareChatCompletions, runToolAwareOpenAIChat } from "./chat-tools.js";
import { buildAnthropicConversationMessage, getAnthropicSystemPrompt } from "./message-content.js";
import { toPrismaProvider } from "./provider-ids.js";
import type { MultiplexRequest, MultiplexResponse, Provider } from "./types.js";
@@ -46,11 +47,15 @@ export async function runMultiplex(req: MultiplexRequest): Promise<MultiplexResp
let usage: MultiplexResponse["usage"] | undefined;
let raw: unknown;
let toolMessages: ReturnType<typeof buildToolLogMessageData>[] = [];
const adapter = getProviderChatAdapter(req.provider);
const r = await adapter.complete({
const enabledTools = normalizeEnabledChatTools(req.enabledTools);
if (req.provider === "openai" && enabledTools.length > 0) {
const client = openaiClient();
const r = await runToolAwareOpenAIChat({
client,
model: req.model,
messages: req.messages,
enabledTools: req.enabledTools,
enabledTools,
userLocation: req.userLocation,
temperature: req.temperature,
maxTokens: req.maxTokens,
@@ -64,6 +69,75 @@ export async function runMultiplex(req: MultiplexRequest): Promise<MultiplexResp
outText = r.text;
usage = r.usage;
toolMessages = r.toolEvents.map((event) => buildToolLogMessageData(call.chatId, event));
} else if (req.provider === "xai" && enabledTools.length > 0) {
const client = xaiClient();
const r = await runToolAwareChatCompletions({
client,
model: req.model,
messages: req.messages,
enabledTools,
userLocation: req.userLocation,
temperature: req.temperature,
maxTokens: req.maxTokens,
logContext: {
provider: req.provider,
model: req.model,
chatId,
},
});
raw = r.raw;
outText = r.text;
usage = r.usage;
toolMessages = r.toolEvents.map((event) => buildToolLogMessageData(call.chatId, event));
} else if (req.provider === "openai" || req.provider === "xai" || req.provider === "hermes-agent") {
const client = req.provider === "openai" ? openaiClient() : req.provider === "xai" ? xaiClient() : hermesAgentClient();
const r = await runPlainChatCompletions({
client,
model: req.model,
messages: req.messages,
userLocation: req.userLocation,
temperature: req.temperature,
maxTokens: req.maxTokens,
logContext: {
provider: req.provider,
model: req.model,
chatId,
},
});
raw = r.raw;
outText = r.text;
usage = r.usage;
} else if (req.provider === "anthropic") {
const client = anthropicClient();
const system = getAnthropicSystemPrompt(req.messages, req.userLocation);
const msgs = req.messages.filter((message) => message.role !== "system").map((message) => buildAnthropicConversationMessage(message));
const r = await client.messages.create({
model: req.model,
system,
max_tokens: req.maxTokens ?? 1024,
temperature: req.temperature,
messages: msgs as any,
});
raw = r;
outText = r.content
.map((c: any) => (c.type === "text" ? c.text : ""))
.join("")
.trim();
// Anthropic usage (SDK typing varies by version)
const ru: any = (r as any).usage;
if (ru) {
usage = {
inputTokens: ru.input_tokens,
outputTokens: ru.output_tokens,
totalTokens: (ru.input_tokens ?? 0) + (ru.output_tokens ?? 0),
};
}
} else {
throw new Error(`unknown provider: ${req.provider}`);
}
const latencyMs = Math.round(performance.now() - t0);

386
server/src/llm/protocols/chat-completions-api.ts
View File

@@ -1,386 +0,0 @@
import {
appendDanglingToolIntentCorrection,
buildChatToolSystemPrompt,
executeToolCallAndBuildEvent,
getEnabledChatTools,
getUnstreamedText,
looksLikeDanglingToolIntent,
MAX_DANGLING_TOOL_INTENT_RETRIES,
MAX_TOOL_ROUNDS,
mergeUsage,
normalizeModelToolCalls,
prepareToolCallExecution,
type NormalizedToolCall,
type ToolAwareCompletionParams,
type ToolAwareCompletionResult,
type ToolAwareStreamingEvent,
type ToolExecutionEvent,
} from "../chat-tools.js";
import {
buildImageSummaryText,
buildSystemPromptAugmentationMessage,
buildTextAttachmentPrompt,
getImageAttachments,
getTextAttachments,
} from "../message-content.js";
import type { ChatMessage } from "../types.js";
function toContentParts(message: ChatMessage) {
const imageAttachments = getImageAttachments(message);
const textAttachments = getTextAttachments(message);
if (!imageAttachments.length && !textAttachments.length) {
return message.content;
}
const parts: Array<Record<string, unknown>> = [];
for (const attachment of imageAttachments) {
parts.push({
type: "image_url",
image_url: {
url: attachment.dataUrl,
detail: "auto",
},
});
}
const imageSummary = buildImageSummaryText(imageAttachments);
if (imageSummary) {
parts.push({ type: "text", text: imageSummary });
}
for (const attachment of textAttachments) {
parts.push({ type: "text", text: buildTextAttachmentPrompt(attachment) });
}
if (message.content.trim()) {
parts.push({ type: "text", text: message.content });
}
if (parts.length === 1 && parts[0]?.type === "text" && typeof parts[0].text === "string") {
return parts[0].text;
}
return parts;
}
function buildConversationMessage(message: ChatMessage) {
if (message.role === "tool") {
const name = message.name?.trim() || "tool";
return {
role: "user",
content: `Tool output (${name}):\n${message.content}`,
};
}
const out: Record<string, unknown> = {
role: message.role,
content: toContentParts(message),
};
if (message.name && (message.role === "assistant" || message.role === "user")) {
out.name = message.name;
}
return out;
}
function normalizeMessages(messages: ChatMessage[], userLocation?: string, params: Pick<ToolAwareCompletionParams, "enabledTools"> = {}) {
const normalized = messages.map((message) => buildConversationMessage(message));
return [{ role: "system", content: buildChatToolSystemPrompt(params) }, buildSystemPromptAugmentationMessage(userLocation), ...normalized];
}
function normalizePlainMessages(messages: ChatMessage[], userLocation?: string) {
return [buildSystemPromptAugmentationMessage(userLocation), ...messages.map((message) => buildConversationMessage(message))];
}
function extractContent(message: any) {
if (typeof message?.content === "string") return message.content;
if (!Array.isArray(message?.content)) return "";
return message.content
.map((part: any) => {
if (typeof part === "string") return part;
if (typeof part?.text === "string") return part.text;
if (typeof part?.content === "string") return part.content;
return "";
})
.join("");
}
export async function completeWithChatCompletionsApi(params: ToolAwareCompletionParams): Promise<ToolAwareCompletionResult> {
const enabledTools = getEnabledChatTools(params);
if (!enabledTools.length) {
const completion = await params.client.chat.completions.create({
model: params.model,
messages: normalizePlainMessages(params.messages, params.userLocation),
temperature: params.temperature,
max_tokens: params.maxTokens,
} as any);
const usageAcc: Required<NonNullable<ToolAwareCompletionResult["usage"]>> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
const sawUsage = mergeUsage(usageAcc, completion?.usage);
const message = completion?.choices?.[0]?.message;
return {
text: extractContent(message),
usage: sawUsage ? usageAcc : undefined,
raw: { response: completion, api: "chat.completions" },
toolEvents: [],
};
}
const conversation: any[] = normalizeMessages(params.messages, params.userLocation, params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<NonNullable<ToolAwareCompletionResult["usage"]>> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const completion = await params.client.chat.completions.create({
model: params.model,
messages: conversation,
temperature: params.temperature,
max_tokens: params.maxTokens,
tools: enabledTools,
tool_choice: "auto",
} as any);
rawResponses.push(completion);
sawUsage = mergeUsage(usageAcc, completion?.usage) || sawUsage;
const message = completion?.choices?.[0]?.message;
if (!message) {
return {
text: "",
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, missingMessage: true },
toolEvents,
};
}
const toolCalls = Array.isArray(message.tool_calls) ? message.tool_calls : [];
if (!toolCalls.length) {
const text = typeof message.content === "string" ? message.content : "";
if (danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES && looksLikeDanglingToolIntent(text)) {
danglingToolIntentRetries += 1;
appendDanglingToolIntentCorrection(conversation, text);
continue;
}
return {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls },
toolEvents,
};
}
const normalizedToolCalls = normalizeModelToolCalls(toolCalls, round);
totalToolCalls += normalizedToolCalls.length;
const assistantToolCallMessage: any = {
role: "assistant",
tool_calls: normalizedToolCalls.map((call) => ({
id: call.id,
type: "function",
function: {
name: call.name,
arguments: call.arguments,
},
})),
};
if (typeof message.content === "string" && message.content.length) {
assistantToolCallMessage.content = message.content;
}
conversation.push(assistantToolCallMessage);
for (const call of normalizedToolCalls) {
const { execution } = prepareToolCallExecution(call);
const { event, toolResult } = await executeToolCallAndBuildEvent(call, execution, params);
toolEvents.push(event);
conversation.push({
role: "tool",
tool_call_id: call.id,
content: JSON.stringify(toolResult),
});
}
}
return {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true },
toolEvents,
};
}
export async function* streamWithChatCompletionsApi(params: ToolAwareCompletionParams): AsyncGenerator<ToolAwareStreamingEvent> {
const enabledTools = getEnabledChatTools(params);
if (!enabledTools.length) {
const rawResponses: unknown[] = [];
const usageAcc: Required<NonNullable<ToolAwareCompletionResult["usage"]>> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let text = "";
const stream = await params.client.chat.completions.create({
model: params.model,
messages: normalizePlainMessages(params.messages, params.userLocation),
temperature: params.temperature,
max_tokens: params.maxTokens,
stream: true,
} as any);
for await (const chunk of stream as any as AsyncIterable<any>) {
rawResponses.push(chunk);
sawUsage = mergeUsage(usageAcc, chunk?.usage) || sawUsage;
const deltaText = chunk?.choices?.[0]?.delta?.content ?? "";
if (typeof deltaText === "string" && deltaText.length) {
text += deltaText;
yield { type: "delta", text: deltaText };
}
}
yield {
type: "done",
result: {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, api: "chat.completions" },
toolEvents: [],
},
};
return;
}
const conversation: any[] = normalizeMessages(params.messages, params.userLocation, params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<NonNullable<ToolAwareCompletionResult["usage"]>> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const stream = await params.client.chat.completions.create({
model: params.model,
messages: conversation,
temperature: params.temperature,
max_tokens: params.maxTokens,
tools: enabledTools,
tool_choice: "auto",
stream: true,
stream_options: { include_usage: true },
} as any);
let roundText = "";
let streamedRoundText = "";
let roundHasToolCalls = false;
const roundToolCalls = new Map<number, { id?: string; name?: string; arguments: string }>();
for await (const chunk of stream as any as AsyncIterable<any>) {
rawResponses.push(chunk);
sawUsage = mergeUsage(usageAcc, chunk?.usage) || sawUsage;
const choice = chunk?.choices?.[0];
const deltaText = choice?.delta?.content ?? "";
if (typeof deltaText === "string" && deltaText.length) {
roundText += deltaText;
if (!roundHasToolCalls) {
streamedRoundText += deltaText;
yield { type: "delta", text: deltaText };
}
}
const deltaToolCalls = Array.isArray(choice?.delta?.tool_calls) ? choice.delta.tool_calls : [];
if (deltaToolCalls.length) {
roundHasToolCalls = true;
}
for (const toolCall of deltaToolCalls) {
const idx = typeof toolCall?.index === "number" ? toolCall.index : 0;
const entry = roundToolCalls.get(idx) ?? { arguments: "" };
if (typeof toolCall?.id === "string" && toolCall.id.length) {
entry.id = toolCall.id;
}
if (typeof toolCall?.function?.name === "string" && toolCall.function.name.length) {
entry.name = toolCall.function.name;
}
if (typeof toolCall?.function?.arguments === "string" && toolCall.function.arguments.length) {
entry.arguments += toolCall.function.arguments;
}
roundToolCalls.set(idx, entry);
}
}
const normalizedToolCalls: NormalizedToolCall[] = [...roundToolCalls.entries()]
.sort((a, b) => a[0] - b[0])
.map(([_, call], index) => ({
id: call.id ?? `tool_call_${round}_${index}`,
name: call.name ?? "unknown_tool",
arguments: call.arguments || "{}",
}));
if (!normalizedToolCalls.length) {
if (!streamedRoundText && danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES && looksLikeDanglingToolIntent(roundText)) {
danglingToolIntentRetries += 1;
appendDanglingToolIntentCorrection(conversation, roundText);
continue;
}
const unstreamedText = getUnstreamedText(roundText, streamedRoundText);
if (unstreamedText) {
yield { type: "delta", text: unstreamedText };
}
yield {
type: "done",
result: {
text: roundText,
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls },
toolEvents,
},
};
return;
}
totalToolCalls += normalizedToolCalls.length;
const assistantToolCallMessage: any = {
role: "assistant",
tool_calls: normalizedToolCalls.map((call) => ({
id: call.id,
type: "function",
function: {
name: call.name,
arguments: call.arguments,
},
})),
};
if (roundText) {
assistantToolCallMessage.content = roundText;
}
conversation.push(assistantToolCallMessage);
for (const call of normalizedToolCalls) {
const { event: initiatedEvent, execution } = prepareToolCallExecution(call);
yield { type: "tool_call", event: initiatedEvent };
const { event, toolResult } = await executeToolCallAndBuildEvent(call, execution, params);
toolEvents.push(event);
yield { type: "tool_call", event };
conversation.push({
role: "tool",
tool_call_id: call.id,
content: JSON.stringify(toolResult),
});
}
}
yield {
type: "done",
result: {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true },
toolEvents,
},
};
}

470
server/src/llm/protocols/messages-api.ts
View File

@@ -1,470 +0,0 @@
import {
buildChatToolSystemPrompt,
executeToolCallAndBuildEvent,
getEnabledChatTools,
looksLikeDanglingToolIntent,
MAX_DANGLING_TOOL_INTENT_RETRIES,
MAX_TOOL_ROUNDS,
parseToolArgs,
prepareToolCallExecution,
type NormalizedToolCall,
type ToolAwareCompletionParams,
type ToolAwareCompletionResult,
type ToolAwareStreamingEvent,
type ToolAwareUsage,
type ToolExecutionEvent,
type ToolRunOutcome,
} from "../chat-tools.js";
import {
buildImageSummaryText,
buildTextAttachmentPrompt,
buildTopLevelSystemPrompt,
getImageAttachments,
getTextAttachments,
parseImageDataUrl,
} from "../message-content.js";
import type { ChatMessage } from "../types.js";
const INTERNAL_CORRECTION =
"Internal correction: the previous assistant message claimed it would run a tool, but no tool call was made. If the task needs an available tool, call it now. Otherwise provide the final answer directly without saying you will run a tool.";
function toTools(tools: any[]) {
return tools
.map((tool) => {
if (tool?.type !== "function") return null;
return {
name: tool.function.name,
description: tool.function.description,
input_schema: tool.function.parameters,
};
})
.filter(Boolean);
}
function toContentBlocks(message: ChatMessage) {
const imageAttachments = getImageAttachments(message);
const textAttachments = getTextAttachments(message);
if (!imageAttachments.length && !textAttachments.length) {
return message.content;
}
const blocks: Array<Record<string, unknown>> = [];
for (const attachment of imageAttachments) {
const source = parseImageDataUrl(attachment);
blocks.push({
type: "image",
source: {
type: "base64",
media_type: source.mediaType,
data: source.data,
},
});
}
const imageSummary = buildImageSummaryText(imageAttachments);
if (imageSummary) {
blocks.push({ type: "text", text: imageSummary });
}
for (const attachment of textAttachments) {
blocks.push({ type: "text", text: buildTextAttachmentPrompt(attachment) });
}
if (message.content.trim()) {
blocks.push({ type: "text", text: message.content });
}
if (blocks.length === 1 && blocks[0]?.type === "text" && typeof blocks[0].text === "string") {
return blocks[0].text;
}
return blocks;
}
function buildConversationMessage(message: ChatMessage) {
if (message.role === "system") {
throw new Error("System messages must be handled separately for top-level-system protocols.");
}
if (message.role === "tool") {
const name = message.name?.trim() || "tool";
return {
role: "user",
content: `Tool output (${name}):\n${message.content}`,
};
}
return {
role: message.role === "assistant" ? "assistant" : "user",
content: toContentBlocks(message),
};
}
function buildBaseMessages(params: ToolAwareCompletionParams) {
return params.messages.filter((message) => message.role !== "system").map((message) => buildConversationMessage(message));
}
function stringifyToolInput(input: unknown) {
if (typeof input === "string") return input;
try {
return JSON.stringify(input ?? {});
} catch {
return "{}";
}
}
function normalizeToolCalls(content: any[], round: number): NormalizedToolCall[] {
return content
.filter((item) => item?.type === "tool_use")
.map((call: any, index: number) => ({
id: call?.id ?? `tool_call_${round}_${index}`,
name: call?.name ?? "unknown_tool",
arguments: stringifyToolInput(call?.input),
}));
}
function extractText(response: any) {
if (!Array.isArray(response?.content)) return "";
return response.content
.map((content: any) => (content?.type === "text" && typeof content.text === "string" ? content.text : ""))
.join("")
.trim();
}
function buildToolResultBlock(call: NormalizedToolCall, toolResult: ToolRunOutcome) {
return {
type: "tool_result",
tool_use_id: call.id,
content: JSON.stringify(toolResult),
is_error: !toolResult.ok,
};
}
function appendCorrection(conversation: any[], text: string) {
conversation.push({ role: "assistant", content: text });
conversation.push({
role: "user",
content: INTERNAL_CORRECTION,
});
}
function mergeUsage(acc: Required<ToolAwareUsage>, usage: any) {
if (!usage) return false;
const inputTokens = usage.input_tokens ?? 0;
const outputTokens = usage.output_tokens ?? 0;
acc.inputTokens += inputTokens;
acc.outputTokens += outputTokens;
acc.totalTokens += inputTokens + outputTokens;
return true;
}
export async function completeWithMessagesApi(params: ToolAwareCompletionParams): Promise<ToolAwareCompletionResult> {
const enabledTools = getEnabledChatTools(params);
if (!enabledTools.length) {
const response = await params.client.messages.create({
model: params.model,
system: buildTopLevelSystemPrompt(params.messages, params.userLocation),
max_tokens: params.maxTokens ?? 1024,
temperature: params.temperature,
messages: buildBaseMessages(params),
} as any);
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
const sawUsage = mergeUsage(usageAcc, response?.usage);
return {
text: extractText(response),
usage: sawUsage ? usageAcc : undefined,
raw: { response, api: "messages" },
toolEvents: [],
};
}
const conversation: any[] = buildBaseMessages(params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const response = await params.client.messages.create({
model: params.model,
system: buildTopLevelSystemPrompt(params.messages, params.userLocation, buildChatToolSystemPrompt(params)),
max_tokens: params.maxTokens ?? 1024,
temperature: params.temperature,
messages: conversation,
tools: toTools(enabledTools),
tool_choice: { type: "auto" },
} as any);
rawResponses.push(response);
sawUsage = mergeUsage(usageAcc, response?.usage) || sawUsage;
const content = Array.isArray(response?.content) ? response.content : [];
const normalizedToolCalls = normalizeToolCalls(content, round);
if (!normalizedToolCalls.length) {
const text = extractText(response);
if (danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES && looksLikeDanglingToolIntent(text)) {
danglingToolIntentRetries += 1;
appendCorrection(conversation, text);
continue;
}
return {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, api: "messages" },
toolEvents,
};
}
totalToolCalls += normalizedToolCalls.length;
conversation.push({
role: "assistant",
content,
});
const toolResultBlocks: any[] = [];
for (const call of normalizedToolCalls) {
const { execution } = prepareToolCallExecution(call);
const { event, toolResult } = await executeToolCallAndBuildEvent(call, execution, params);
toolEvents.push(event);
toolResultBlocks.push(buildToolResultBlock(call, toolResult));
}
conversation.push({
role: "user",
content: toolResultBlocks,
});
}
return {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true, api: "messages" },
toolEvents,
};
}
export async function* streamWithMessagesApi(params: ToolAwareCompletionParams): AsyncGenerator<ToolAwareStreamingEvent> {
const enabledTools = getEnabledChatTools(params);
if (!enabledTools.length) {
const rawResponses: unknown[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let roundInputTokens = 0;
let roundOutputTokens = 0;
let text = "";
const stream = await params.client.messages.create({
model: params.model,
system: buildTopLevelSystemPrompt(params.messages, params.userLocation),
max_tokens: params.maxTokens ?? 1024,
temperature: params.temperature,
messages: buildBaseMessages(params),
stream: true,
} as any);
for await (const ev of stream as any as AsyncIterable<any>) {
rawResponses.push(ev);
if (ev?.type === "message_start" && ev?.message?.usage) {
roundInputTokens = ev.message.usage.input_tokens ?? roundInputTokens;
sawUsage = true;
}
if (ev?.type === "content_block_delta" && ev?.delta?.type === "text_delta") {
const delta = ev.delta.text ?? "";
if (delta) {
text += delta;
yield { type: "delta", text: delta };
}
}
if (ev?.type === "message_delta" && ev.usage) {
roundInputTokens = ev.usage.input_tokens ?? roundInputTokens;
roundOutputTokens = ev.usage.output_tokens ?? roundOutputTokens;
sawUsage = true;
}
}
if (sawUsage) {
usageAcc.inputTokens += roundInputTokens;
usageAcc.outputTokens += roundOutputTokens;
usageAcc.totalTokens += roundInputTokens + roundOutputTokens;
}
yield {
type: "done",
result: {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: 0, api: "messages" },
toolEvents: [],
},
};
return;
}
const conversation: any[] = buildBaseMessages(params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const stream = await params.client.messages.create({
model: params.model,
system: buildTopLevelSystemPrompt(params.messages, params.userLocation, buildChatToolSystemPrompt(params)),
max_tokens: params.maxTokens ?? 1024,
temperature: params.temperature,
messages: conversation,
tools: toTools(enabledTools),
tool_choice: { type: "auto" },
stream: true,
} as any);
const contentByIndex = new Map<number, any>();
const toolArgumentByIndex = new Map<number, string>();
let roundText = "";
let roundHasToolCalls = false;
let roundInputTokens = 0;
let roundOutputTokens = 0;
let sawRoundUsage = false;
for await (const ev of stream as any as AsyncIterable<any>) {
rawResponses.push(ev);
if (ev?.type === "message_start" && ev?.message?.usage) {
roundInputTokens = ev.message.usage.input_tokens ?? roundInputTokens;
sawRoundUsage = true;
}
if (ev?.type === "content_block_start" && typeof ev.index === "number") {
const block = ev.content_block ?? {};
if (block.type === "tool_use") {
roundHasToolCalls = true;
contentByIndex.set(ev.index, {
type: "tool_use",
id: block.id,
name: block.name,
input: block.input ?? {},
});
toolArgumentByIndex.set(ev.index, "");
} else if (block.type === "text") {
contentByIndex.set(ev.index, {
type: "text",
text: typeof block.text === "string" ? block.text : "",
});
} else if (block.type) {
contentByIndex.set(ev.index, block);
}
}
if (ev?.type === "content_block_delta" && typeof ev.index === "number") {
if (ev.delta?.type === "text_delta") {
const delta = typeof ev.delta.text === "string" ? ev.delta.text : "";
if (delta) {
const block = contentByIndex.get(ev.index) ?? { type: "text", text: "" };
if (block.type === "text") {
block.text = `${typeof block.text === "string" ? block.text : ""}${delta}`;
contentByIndex.set(ev.index, block);
}
roundText += delta;
}
} else if (ev.delta?.type === "input_json_delta") {
roundHasToolCalls = true;
const partialJson = typeof ev.delta.partial_json === "string" ? ev.delta.partial_json : "";
toolArgumentByIndex.set(ev.index, `${toolArgumentByIndex.get(ev.index) ?? ""}${partialJson}`);
}
}
if (ev?.type === "content_block_stop" && typeof ev.index === "number") {
const block = contentByIndex.get(ev.index);
if (block?.type === "tool_use") {
const rawArguments = toolArgumentByIndex.get(ev.index) || stringifyToolInput(block.input);
try {
block.input = parseToolArgs(rawArguments);
} catch {
block.input = {};
}
contentByIndex.set(ev.index, block);
}
}
if (ev?.type === "message_delta" && ev.usage) {
roundInputTokens = ev.usage.input_tokens ?? roundInputTokens;
roundOutputTokens = ev.usage.output_tokens ?? roundOutputTokens;
sawRoundUsage = true;
}
}
if (sawRoundUsage) {
usageAcc.inputTokens += roundInputTokens;
usageAcc.outputTokens += roundOutputTokens;
usageAcc.totalTokens += roundInputTokens + roundOutputTokens;
sawUsage = true;
}
const indexedContent = [...contentByIndex.entries()].sort((a, b) => a[0] - b[0]);
const assistantContent = indexedContent.map(([, block]) => block);
const normalizedToolCalls: NormalizedToolCall[] = indexedContent
.filter(([, block]) => block?.type === "tool_use")
.map(([index, block], callIndex) => ({
id: block.id ?? `tool_call_${round}_${callIndex}`,
name: block.name ?? "unknown_tool",
arguments: toolArgumentByIndex.get(index) || stringifyToolInput(block.input),
}));
if (!normalizedToolCalls.length) {
if (danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES && looksLikeDanglingToolIntent(roundText)) {
danglingToolIntentRetries += 1;
appendCorrection(conversation, roundText);
continue;
}
if (roundText) {
yield { type: "delta", text: roundText };
}
yield {
type: "done",
result: {
text: roundText,
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls, api: "messages" },
toolEvents,
},
};
return;
}
totalToolCalls += normalizedToolCalls.length;
conversation.push({
role: "assistant",
content: assistantContent,
});
const toolResultBlocks: any[] = [];
for (const call of normalizedToolCalls) {
const { event: initiatedEvent, execution } = prepareToolCallExecution(call);
yield { type: "tool_call", event: initiatedEvent };
const { event, toolResult } = await executeToolCallAndBuildEvent(call, execution, params);
toolEvents.push(event);
yield { type: "tool_call", event };
toolResultBlocks.push(buildToolResultBlock(call, toolResult));
}
conversation.push({
role: "user",
content: toolResultBlocks,
});
}
yield {
type: "done",
result: {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true, api: "messages" },
toolEvents,
},
};
}

332
server/src/llm/protocols/responses-api.ts
View File

@@ -1,332 +0,0 @@
import {
appendDanglingToolIntentCorrection,
buildChatToolSystemPrompt,
executeToolCallAndBuildEvent,
getEnabledChatTools,
getUnstreamedText,
looksLikeDanglingToolIntent,
MAX_DANGLING_TOOL_INTENT_RETRIES,
MAX_TOOL_ROUNDS,
prepareToolCallExecution,
type NormalizedToolCall,
type ToolAwareCompletionParams,
type ToolAwareCompletionResult,
type ToolAwareStreamingEvent,
type ToolAwareUsage,
type ToolExecutionEvent,
} from "../chat-tools.js";
import {
buildImageSummaryText,
buildSystemPromptAugmentationMessage,
buildTextAttachmentPrompt,
getImageAttachments,
getTextAttachments,
} from "../message-content.js";
import type { ChatMessage } from "../types.js";
function toResponsesTools(tools: any[]) {
return tools.map((tool) => {
if (tool?.type !== "function") return tool;
return {
type: "function",
name: tool.function.name,
description: tool.function.description,
parameters: tool.function.parameters,
strict: false,
};
});
}
function toContentParts(message: ChatMessage) {
const imageAttachments = getImageAttachments(message);
const textAttachments = getTextAttachments(message);
if (!imageAttachments.length && !textAttachments.length) {
return message.content;
}
const parts: Array<Record<string, unknown>> = [];
for (const attachment of imageAttachments) {
parts.push({
type: "input_image",
image_url: attachment.dataUrl,
detail: "auto",
});
}
const imageSummary = buildImageSummaryText(imageAttachments);
if (imageSummary) {
parts.push({ type: "input_text", text: imageSummary });
}
for (const attachment of textAttachments) {
parts.push({ type: "input_text", text: buildTextAttachmentPrompt(attachment) });
}
if (message.content.trim()) {
parts.push({ type: "input_text", text: message.content });
}
if (parts.length === 1 && parts[0]?.type === "input_text" && typeof parts[0].text === "string") {
return parts[0].text;
}
return parts;
}
function buildInputMessage(message: ChatMessage) {
if (message.role === "tool") {
const name = message.name?.trim() || "tool";
return {
role: "user",
content: `Tool output (${name}):\n${message.content}`,
};
}
return {
role: message.role,
content: toContentParts(message),
};
}
function normalizeInput(messages: ChatMessage[], userLocation?: string, params: Pick<ToolAwareCompletionParams, "enabledTools"> = {}) {
const normalized = messages.map((message) => buildInputMessage(message));
return [{ role: "system", content: buildChatToolSystemPrompt(params) }, buildSystemPromptAugmentationMessage(userLocation), ...normalized];
}
function mergeUsage(acc: Required<ToolAwareUsage>, usage: any) {
if (!usage) return false;
acc.inputTokens += usage.input_tokens ?? 0;
acc.outputTokens += usage.output_tokens ?? 0;
acc.totalTokens += usage.total_tokens ?? 0;
return true;
}
function getOutputItems(response: any) {
return Array.isArray(response?.output) ? response.output : [];
}
function extractText(response: any, fallback = "") {
if (typeof response?.output_text === "string") return response.output_text;
const parts: string[] = [];
for (const item of getOutputItems(response)) {
if (item?.type !== "message" || !Array.isArray(item.content)) continue;
for (const content of item.content) {
if (content?.type === "output_text" && typeof content.text === "string") {
parts.push(content.text);
} else if (content?.type === "refusal" && typeof content.refusal === "string") {
parts.push(content.refusal);
}
}
}
return parts.join("") || fallback;
}
function getFailureMessage(response: any) {
if (response?.status !== "failed" && response?.status !== "incomplete") return null;
const errorMessage = typeof response?.error?.message === "string" ? response.error.message : null;
const incompleteReason = typeof response?.incomplete_details?.reason === "string" ? response.incomplete_details.reason : null;
return errorMessage ?? (incompleteReason ? `Response incomplete: ${incompleteReason}` : `Response ${response.status}.`);
}
function normalizeToolCalls(outputItems: any[], round: number): NormalizedToolCall[] {
return outputItems
.filter((item) => item?.type === "function_call")
.map((call: any, index: number) => ({
id: call.call_id ?? call.id ?? `tool_call_${round}_${index}`,
name: call.name ?? "unknown_tool",
arguments: call.arguments ?? "{}",
}));
}
export async function completeWithResponsesApi(params: ToolAwareCompletionParams): Promise<ToolAwareCompletionResult> {
const enabledTools = getEnabledChatTools(params);
const input: any[] = normalizeInput(params.messages, params.userLocation, params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const response = await params.client.responses.create({
model: params.model,
input,
temperature: params.temperature,
max_output_tokens: params.maxTokens,
tools: toResponsesTools(enabledTools),
tool_choice: "auto",
parallel_tool_calls: true,
store: true,
} as any);
rawResponses.push(response);
sawUsage = mergeUsage(usageAcc, response?.usage) || sawUsage;
const failureMessage = getFailureMessage(response);
if (failureMessage) {
throw new Error(failureMessage);
}
const outputItems = getOutputItems(response);
const normalizedToolCalls = normalizeToolCalls(outputItems, round);
if (!normalizedToolCalls.length) {
const text = extractText(response);
if (danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES && looksLikeDanglingToolIntent(text)) {
danglingToolIntentRetries += 1;
appendDanglingToolIntentCorrection(input, text);
continue;
}
return {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, api: "responses" },
toolEvents,
};
}
totalToolCalls += normalizedToolCalls.length;
input.push(...outputItems);
for (const call of normalizedToolCalls) {
const { execution } = prepareToolCallExecution(call);
const { event, toolResult } = await executeToolCallAndBuildEvent(call, execution, params);
toolEvents.push(event);
input.push({
type: "function_call_output",
call_id: call.id,
output: JSON.stringify(toolResult),
});
}
}
return {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true, api: "responses" },
toolEvents,
};
}
export async function* streamWithResponsesApi(params: ToolAwareCompletionParams): AsyncGenerator<ToolAwareStreamingEvent> {
const enabledTools = getEnabledChatTools(params);
const input: any[] = normalizeInput(params.messages, params.userLocation, params);
const rawResponses: unknown[] = [];
const toolEvents: ToolExecutionEvent[] = [];
const usageAcc: Required<ToolAwareUsage> = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
let sawUsage = false;
let totalToolCalls = 0;
let danglingToolIntentRetries = 0;
for (let round = 0; round < MAX_TOOL_ROUNDS; round += 1) {
const stream = await params.client.responses.create({
model: params.model,
input,
temperature: params.temperature,
max_output_tokens: params.maxTokens,
tools: toResponsesTools(enabledTools),
tool_choice: "auto",
parallel_tool_calls: true,
store: true,
stream: true,
} as any);
let roundText = "";
let streamedRoundText = "";
let roundHasToolCalls = false;
let canStreamRoundText = false;
let completedResponse: any | null = null;
const completedOutputItems: any[] = [];
for await (const event of stream as any as AsyncIterable<any>) {
rawResponses.push(event);
if (event?.type === "response.output_text.delta" && typeof event.delta === "string") {
roundText += event.delta;
if (canStreamRoundText && !roundHasToolCalls && event.delta.length) {
streamedRoundText += event.delta;
yield { type: "delta", text: event.delta };
}
} else if (event?.type === "response.output_item.added" && event.item) {
if (event.item.type === "function_call") {
roundHasToolCalls = true;
canStreamRoundText = false;
} else if (event.item.type === "message" && !roundHasToolCalls) {
canStreamRoundText = true;
}
} else if (event?.type === "response.output_item.done" && event.item) {
completedOutputItems[event.output_index ?? completedOutputItems.length] = event.item;
if (event.item.type === "function_call") {
roundHasToolCalls = true;
canStreamRoundText = false;
}
} else if (event?.type === "response.completed") {
completedResponse = event.response;
sawUsage = mergeUsage(usageAcc, event.response?.usage) || sawUsage;
} else if (event?.type === "response.failed" || event?.type === "response.incomplete") {
completedResponse = event.response;
sawUsage = mergeUsage(usageAcc, event.response?.usage) || sawUsage;
} else if (event?.type === "error") {
throw new Error(event.message ?? "Responses stream failed.");
}
}
const failureMessage = getFailureMessage(completedResponse);
if (failureMessage) {
throw new Error(failureMessage);
}
const outputItems = getOutputItems(completedResponse);
const responseOutputItems = outputItems.length ? outputItems : completedOutputItems.filter(Boolean);
const normalizedToolCalls = normalizeToolCalls(responseOutputItems, round);
if (!normalizedToolCalls.length) {
const text = extractText(completedResponse, roundText);
if (!streamedRoundText && danglingToolIntentRetries < MAX_DANGLING_TOOL_INTENT_RETRIES && looksLikeDanglingToolIntent(text)) {
danglingToolIntentRetries += 1;
appendDanglingToolIntentCorrection(input, text);
continue;
}
const unstreamedText = getUnstreamedText(text, streamedRoundText);
if (unstreamedText) {
yield { type: "delta", text: unstreamedText };
}
yield {
type: "done",
result: {
text,
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls, api: "responses" },
toolEvents,
},
};
return;
}
totalToolCalls += normalizedToolCalls.length;
input.push(...responseOutputItems);
for (const call of normalizedToolCalls) {
const { event: initiatedEvent, execution } = prepareToolCallExecution(call);
yield { type: "tool_call", event: initiatedEvent };
const { event, toolResult } = await executeToolCallAndBuildEvent(call, execution, params);
toolEvents.push(event);
yield { type: "tool_call", event };
input.push({
type: "function_call_output",
call_id: call.id,
output: JSON.stringify(toolResult),
});
}
}
yield {
type: "done",
result: {
text: "I reached the tool-call limit while gathering information. Please narrow the request and try again.",
usage: sawUsage ? usageAcc : undefined,
raw: { streamed: true, responses: rawResponses, toolCallsUsed: totalToolCalls, toolCallLimitReached: true, api: "responses" },
toolEvents,
},
};
}

217
server/src/llm/provider-adapters.ts
View File

@@ -1,217 +0,0 @@
import {
normalizeEnabledChatTools,
type ToolAwareCompletionParams,
type ToolAwareCompletionResult,
type ToolAwareStreamingEvent,
} from "./chat-tools.js";
import { completeWithChatCompletionsApi, streamWithChatCompletionsApi } from "./protocols/chat-completions-api.js";
import { completeWithMessagesApi, streamWithMessagesApi } from "./protocols/messages-api.js";
import { completeWithResponsesApi, streamWithResponsesApi } from "./protocols/responses-api.js";
import { env } from "../env.js";
import { anthropicClient, hermesAgentClient, isHermesAgentConfigured, openaiClient, xaiClient } from "./providers.js";
import type { ChatMessage, Provider } from "./types.js";
type ProviderAdapterParams = {
model: string;
messages: ChatMessage[];
enabledTools?: string[];
userLocation?: string;
temperature?: number;
maxTokens?: number;
logContext?: ToolAwareCompletionParams["logContext"];
};
export type ProviderChatAdapter = {
provider: Provider;
complete(params: ProviderAdapterParams): Promise<ToolAwareCompletionResult>;
stream(params: ProviderAdapterParams): AsyncGenerator<ToolAwareStreamingEvent>;
};
type ChatProtocolId = "chat-completions" | "messages" | "responses";
type ChatProtocol = {
id: ChatProtocolId;
complete(params: ToolAwareCompletionParams): Promise<ToolAwareCompletionResult>;
stream(params: ToolAwareCompletionParams): AsyncGenerator<ToolAwareStreamingEvent>;
};
type ModelCatalogSpec = {
enabled?: () => boolean;
fetchModels(client: any): Promise<string[]>;
fallbackModels?: () => string[];
};
type ProviderBackendSpec = {
createClient: () => any;
plainProtocol: ChatProtocol;
toolProtocol?: ChatProtocol;
managedTools?: boolean;
modelCatalog?: ModelCatalogSpec;
};
const chatCompletionsProtocol: ChatProtocol = {
id: "chat-completions",
complete: completeWithChatCompletionsApi,
stream: streamWithChatCompletionsApi,
};
const messagesProtocol: ChatProtocol = {
id: "messages",
complete: completeWithMessagesApi,
stream: streamWithMessagesApi,
};
const responsesProtocol: ChatProtocol = {
id: "responses",
complete: completeWithResponsesApi,
stream: streamWithResponsesApi,
};
function uniqSorted(values: string[]) {
return [...new Set(values.map((value) => value.trim()).filter(Boolean))].sort((a, b) => a.localeCompare(b));
}
function modelIdsFromListResponse(page: any) {
return Array.isArray(page?.data)
? page.data.map((model: any) => model?.id).filter((id: unknown): id is string => typeof id === "string")
: [];
}
function isLikelyResponsesApiModel(model: string) {
const id = model.toLowerCase();
if (id.includes("embedding") || id.includes("moderation")) return false;
if (id.includes("audio") || id.includes("realtime") || id.includes("transcribe") || id.includes("tts")) return false;
if (id.includes("image") || id.includes("dall-e") || id.includes("sora")) return false;
if (id.includes("search") || id.includes("computer-use")) return false;
return /^(gpt-|o\d|chatgpt-)/.test(id);
}
function withClient(params: ProviderAdapterParams, client: any, enabledTools?: string[]): ToolAwareCompletionParams {
return {
client,
model: params.model,
messages: params.messages,
enabledTools,
userLocation: params.userLocation,
temperature: params.temperature,
maxTokens: params.maxTokens,
logContext: params.logContext,
};
}
function selectChatProtocol(spec: ProviderBackendSpec, params: Pick<ProviderAdapterParams, "enabledTools">) {
const enabledTools = normalizeEnabledChatTools(params.enabledTools);
const useManagedTools = spec.managedTools === true && spec.toolProtocol && enabledTools.length > 0;
return {
protocol: useManagedTools ? spec.toolProtocol! : spec.plainProtocol,
enabledTools: useManagedTools ? enabledTools : [],
managedTools: Boolean(useManagedTools),
};
}
function createProviderChatAdapter(provider: Provider, spec: ProviderBackendSpec): ProviderChatAdapter {
return {
provider,
complete(params) {
const selected = selectChatProtocol(spec, params);
return selected.protocol.complete(withClient(params, spec.createClient(), selected.enabledTools));
},
stream(params) {
const selected = selectChatProtocol(spec, params);
return selected.protocol.stream(withClient(params, spec.createClient(), selected.enabledTools));
},
};
}
const backendSpecs: Record<Provider, ProviderBackendSpec> = {
openai: {
createClient: openaiClient,
plainProtocol: chatCompletionsProtocol,
toolProtocol: responsesProtocol,
managedTools: true,
modelCatalog: {
async fetchModels(client) {
const page = await client.models.list();
return modelIdsFromListResponse(page).filter(isLikelyResponsesApiModel);
},
},
},
anthropic: {
createClient: anthropicClient,
plainProtocol: messagesProtocol,
toolProtocol: messagesProtocol,
managedTools: true,
modelCatalog: {
async fetchModels(client) {
const page = await client.models.list({ limit: 200 });
return modelIdsFromListResponse(page);
},
},
},
xai: {
createClient: xaiClient,
plainProtocol: chatCompletionsProtocol,
toolProtocol: chatCompletionsProtocol,
managedTools: true,
modelCatalog: {
async fetchModels(client) {
const page = await client.models.list();
return modelIdsFromListResponse(page);
},
},
},
"hermes-agent": {
createClient: hermesAgentClient,
plainProtocol: chatCompletionsProtocol,
managedTools: false,
modelCatalog: {
enabled: isHermesAgentConfigured,
async fetchModels(client) {
const page = await client.models.list();
const models = modelIdsFromListResponse(page);
if (env.HERMES_AGENT_MODEL) models.push(env.HERMES_AGENT_MODEL);
return models;
},
fallbackModels() {
return env.HERMES_AGENT_MODEL ? [env.HERMES_AGENT_MODEL] : [];
},
},
},
};
const providerChatAdapters: Record<Provider, ProviderChatAdapter> = Object.fromEntries(
Object.entries(backendSpecs).map(([provider, spec]) => [provider, createProviderChatAdapter(provider as Provider, spec)])
) as Record<Provider, ProviderChatAdapter>;
export function getProviderChatAdapter(provider: Provider) {
return providerChatAdapters[provider];
}
export function describeProviderChatBackend(provider: Provider, enabledTools?: string[]) {
const selected = selectChatProtocol(backendSpecs[provider], { enabledTools });
return {
provider,
protocol: selected.protocol.id,
managedTools: selected.managedTools,
enabledTools: selected.enabledTools,
};
}
export function listModelCatalogProviders(): Provider[] {
return (Object.entries(backendSpecs) as [Provider, ProviderBackendSpec][])
.filter(([, spec]) => {
const catalog = spec.modelCatalog;
return catalog !== undefined && catalog.enabled?.() !== false;
})
.map(([provider]) => provider);
}
export async function fetchProviderCatalogModels(provider: Provider) {
const spec = backendSpecs[provider].modelCatalog;
if (!spec) return [];
return uniqSorted(await spec.fetchModels(backendSpecs[provider].createClient()));
}
export function getProviderCatalogFallbackModels(provider: Provider) {
return uniqSorted(backendSpecs[provider].modelCatalog?.fallbackModels?.() ?? []);
}

21
server/src/llm/provider-ids.ts
View File

@@ -2,28 +2,15 @@ import type { Provider } from "./types.js";
type PrismaProvider = Exclude<Provider, "hermes-agent"> | "hermes_agent";
const apiToPrismaProvider = {
openai: "openai",
anthropic: "anthropic",
xai: "xai",
"hermes-agent": "hermes_agent",
} as const satisfies Record<Provider, PrismaProvider>;
const prismaToApiProvider = {
openai: "openai",
anthropic: "anthropic",
xai: "xai",
hermes_agent: "hermes-agent",
"hermes-agent": "hermes-agent",
} as const satisfies Record<PrismaProvider | "hermes-agent", Provider>;
export function toPrismaProvider(provider: Provider): PrismaProvider {
return apiToPrismaProvider[provider];
return provider === "hermes-agent" ? "hermes_agent" : provider;
}
export function fromPrismaProvider(provider: unknown): Provider | null {
if (provider === null || provider === undefined) return null;
return prismaToApiProvider[provider as keyof typeof prismaToApiProvider] ?? null;
if (provider === "hermes_agent" || provider === "hermes-agent") return "hermes-agent";
if (provider === "openai" || provider === "anthropic" || provider === "xai") return provider;
return null;
}
export function serializeProviderFields<T extends Record<string, any>>(value: T): T {

88
server/src/llm/streaming.ts
View File

@@ -1,10 +1,15 @@
import { performance } from "node:perf_hooks";
import { prisma } from "../db.js";
import { anthropicClient, hermesAgentClient, openaiClient, xaiClient } from "./providers.js";
import {
buildToolLogMessageData,
normalizeEnabledChatTools,
runPlainChatCompletionsStream,
runToolAwareChatCompletionsStream,
runToolAwareOpenAIChatStream,
type ToolExecutionEvent,
} from "./chat-tools.js";
import { getProviderChatAdapter } from "./provider-adapters.js";
import { buildAnthropicConversationMessage, getAnthropicSystemPrompt } from "./message-content.js";
import { toPrismaProvider } from "./provider-ids.js";
import type { MultiplexRequest, Provider } from "./types.js";
@@ -70,11 +75,44 @@ export async function* runMultiplexStream(req: MultiplexRequest): AsyncGenerator
let raw: unknown = { streamed: true };
try {
const adapter = getProviderChatAdapter(req.provider);
const streamEvents = adapter.stream({
if (req.provider === "openai" || req.provider === "xai" || req.provider === "hermes-agent") {
const client = req.provider === "openai" ? openaiClient() : req.provider === "xai" ? xaiClient() : hermesAgentClient();
const enabledTools = normalizeEnabledChatTools(req.enabledTools);
const streamEvents =
req.provider === "openai" && enabledTools.length > 0
? runToolAwareOpenAIChatStream({
client,
model: req.model,
messages: req.messages,
enabledTools: req.enabledTools,
enabledTools,
userLocation: req.userLocation,
temperature: req.temperature,
maxTokens: req.maxTokens,
logContext: {
provider: req.provider,
model: req.model,
chatId: chatId ?? undefined,
},
})
: req.provider === "hermes-agent" || enabledTools.length === 0
? runPlainChatCompletionsStream({
client,
model: req.model,
messages: req.messages,
userLocation: req.userLocation,
temperature: req.temperature,
maxTokens: req.maxTokens,
logContext: {
provider: req.provider,
model: req.model,
chatId: chatId ?? undefined,
},
})
: runToolAwareChatCompletionsStream({
client,
model: req.model,
messages: req.messages,
enabledTools,
userLocation: req.userLocation,
temperature: req.temperature,
maxTokens: req.maxTokens,
@@ -84,7 +122,6 @@ export async function* runMultiplexStream(req: MultiplexRequest): AsyncGenerator
chatId: chatId ?? undefined,
},
});
for await (const ev of streamEvents) {
if (ev.type === "delta") {
text += ev.text;
@@ -93,7 +130,7 @@ export async function* runMultiplexStream(req: MultiplexRequest): AsyncGenerator
}
if (ev.type === "tool_call") {
if (ev.event.status !== "initiated" && shouldPersist && chatId) {
if (shouldPersist && chatId) {
const toolMessage = buildToolLogMessageData(chatId, ev.event);
await prisma.message.create({
data: {
@@ -113,6 +150,45 @@ export async function* runMultiplexStream(req: MultiplexRequest): AsyncGenerator
usage = ev.result.usage;
text = ev.result.text;
}
} else if (req.provider === "anthropic") {
const client = anthropicClient();
const system = getAnthropicSystemPrompt(req.messages, req.userLocation);
const msgs = req.messages.filter((message) => message.role !== "system").map((message) => buildAnthropicConversationMessage(message));
const stream = await client.messages.create({
model: req.model,
system,
max_tokens: req.maxTokens ?? 1024,
temperature: req.temperature,
messages: msgs as any,
stream: true,
});
for await (const ev of stream as any as AsyncIterable<any>) {
// Anthropic streaming events include content_block_delta with text_delta
if (ev?.type === "content_block_delta" && ev?.delta?.type === "text_delta") {
const delta = ev.delta.text ?? "";
if (delta) {
text += delta;
yield { type: "delta", text: delta };
}
}
// capture usage if present on message_delta
if (ev?.type === "message_delta" && ev?.usage) {
usage = {
inputTokens: ev.usage.input_tokens,
outputTokens: ev.usage.output_tokens,
totalTokens:
(ev.usage.input_tokens ?? 0) + (ev.usage.output_tokens ?? 0),
};
}
// some streams end with message_stop
}
raw = { streamed: true, provider: "anthropic" };
} else {
throw new Error(`unknown provider: ${req.provider}`);
}
const latencyMs = Math.round(performance.now() - t0);

334
server/src/routes.ts
View File

@@ -13,7 +13,6 @@ import { getModelCatalogSnapshot } from "./llm/model-catalog.js";
import { openaiClient } from "./llm/providers.js";
import { serializeProviderFields, toPrismaProvider } from "./llm/provider-ids.js";
import { exaClient } from "./search/exa.js";
import { isFreshSearchCacheHit, normalizeSearchQuery } from "./search-cache.js";
import type { ChatAttachment } from "./llm/types.js";
const ProviderSchema = z.enum(["openai", "anthropic", "xai", "hermes-agent"]);
@@ -400,15 +399,21 @@ type SearchRunRequest = z.infer<typeof SearchRunBody>;
const activeChatStreams = new Map<string, ActiveSseStream>();
const activeSearchStreams = new Map<string, ActiveSseStream>();
const STARRED_PROJECT_ID = "starred";
const starredProjectItemsSelect = {
where: { projectId: STARRED_PROJECT_ID },
select: { createdAt: true },
take: 1,
} as const;
function getErrorMessage(err: unknown) {
return err instanceof Error ? err.message : String(err);
}
const chatSummarySelect = {
function compareUpdatedAtDesc(a: { updatedAt: Date | string }, b: { updatedAt: Date | string }) {
return new Date(b.updatedAt).getTime() - new Date(a.updatedAt).getTime();
}
async function listWorkspaceItems() {
const [chats, searches] = await Promise.all([
prisma.chat.findMany({
orderBy: { updatedAt: "desc" },
take: 100,
select: {
id: true,
title: true,
createdAt: true,
@@ -419,131 +424,18 @@ const chatSummarySelect = {
lastUsedModel: true,
additionalSystemPrompt: true,
enabledTools: true,
projectItems: starredProjectItemsSelect,
} as const;
const searchSummarySelect = {
id: true,
title: true,
query: true,
createdAt: true,
updatedAt: true,
projectItems: starredProjectItemsSelect,
} as const;
function getErrorMessage(err: unknown) {
return err instanceof Error ? err.message : String(err);
}
function compareUpdatedAtDesc(a: { updatedAt: Date | string }, b: { updatedAt: Date | string }) {
return new Date(b.updatedAt).getTime() - new Date(a.updatedAt).getTime();
}
function serializeStarFields(item: { projectItems?: Array<{ createdAt: Date }> }) {
const star = item.projectItems?.[0];
return {
starred: Boolean(star),
starredAt: star?.createdAt ?? null,
};
}
function serializeChatLike<T extends Record<string, any>>(chat: T) {
const { projectItems: _projectItems, ...rest } = chat;
return {
...serializeProviderFields(rest),
...serializeStarFields(chat),
};
}
function serializeSearchLike<T extends Record<string, any>>(search: T) {
const { projectItems: _projectItems, queryNormalized: _queryNormalized, ...rest } = search;
return {
...rest,
...serializeStarFields(search),
};
}
async function ensureStarredProject() {
await prisma.project.upsert({
where: { id: STARRED_PROJECT_ID },
update: {},
create: {
id: STARRED_PROJECT_ID,
kind: "starred" as any,
title: "Starred",
},
});
}
async function getChatSummary(chatId: string) {
const chat = await prisma.chat.findUnique({
where: { id: chatId },
select: chatSummarySelect,
});
return chat ? serializeChatLike(chat) : null;
}
async function getSearchSummary(searchId: string) {
const search = await prisma.search.findUnique({
where: { id: searchId },
select: searchSummarySelect,
});
return search ? serializeSearchLike(search) : null;
}
async function setChatStarred(chatId: string, starred: boolean) {
const exists = await prisma.chat.findUnique({ where: { id: chatId }, select: { id: true } });
if (!exists) return null;
if (starred) {
await ensureStarredProject();
await prisma.projectItem.upsert({
where: { projectId_chatId: { projectId: STARRED_PROJECT_ID, chatId } },
update: {},
create: { projectId: STARRED_PROJECT_ID, chatId },
});
} else {
await prisma.projectItem.deleteMany({ where: { projectId: STARRED_PROJECT_ID, chatId } });
}
return getChatSummary(chatId);
}
async function setSearchStarred(searchId: string, starred: boolean) {
const exists = await prisma.search.findUnique({ where: { id: searchId }, select: { id: true } });
if (!exists) return null;
if (starred) {
await ensureStarredProject();
await prisma.projectItem.upsert({
where: { projectId_searchId: { projectId: STARRED_PROJECT_ID, searchId } },
update: {},
create: { projectId: STARRED_PROJECT_ID, searchId },
});
} else {
await prisma.projectItem.deleteMany({ where: { projectId: STARRED_PROJECT_ID, searchId } });
}
return getSearchSummary(searchId);
}
async function listWorkspaceItems() {
const [chats, searches] = await Promise.all([
prisma.chat.findMany({
orderBy: { updatedAt: "desc" },
take: 100,
select: chatSummarySelect,
}),
prisma.search.findMany({
orderBy: { updatedAt: "desc" },
take: 100,
select: searchSummarySelect,
select: { id: true, title: true, query: true, createdAt: true, updatedAt: true },
}),
]);
return [
...chats.map((chat) => ({ type: "chat" as const, ...serializeChatLike(chat) })),
...searches.map((search) => ({ type: "search" as const, ...serializeSearchLike(search) })),
...chats.map((chat) => ({ type: "chat" as const, ...serializeProviderFields(chat) })),
...searches.map((search) => ({ type: "search" as const, ...search })),
].sort(compareUpdatedAtDesc);
}
@@ -730,7 +622,6 @@ async function executeSearchRunStream(searchId: string, body: SearchRunRequest,
where: { id: searchId },
data: {
query,
queryNormalized: normalizeSearchQuery(query),
title: normalizedTitle,
requestId: searchResponse?.requestId ?? null,
rawResponse: searchResponse as any,
@@ -751,15 +642,12 @@ async function executeSearchRunStream(searchId: string, body: SearchRunRequest,
const search = await prisma.search.findUnique({
where: { id: searchId },
include: {
results: { orderBy: { rank: "asc" } },
projectItems: starredProjectItemsSelect,
},
include: { results: { orderBy: { rank: "asc" } } },
});
if (!search) {
stream.complete({ event: "error", data: { message: "search not found" } });
} else {
stream.complete({ event: "done", data: { search: serializeSearchLike(search) } });
stream.complete({ event: "done", data: { search } });
}
} catch (err) {
const message = getErrorMessage(err);
@@ -768,7 +656,6 @@ async function executeSearchRunStream(searchId: string, body: SearchRunRequest,
where: { id: searchId },
data: {
query,
queryNormalized: normalizeSearchQuery(query),
title: normalizedTitle,
latencyMs: Math.round(performance.now() - startedAt),
error: message,
@@ -819,9 +706,20 @@ export async function registerRoutes(app: FastifyInstance) {
const chats = await prisma.chat.findMany({
orderBy: { updatedAt: "desc" },
take: 100,
select: chatSummarySelect,
select: {
id: true,
title: true,
createdAt: true,
updatedAt: true,
initiatedProvider: true,
initiatedModel: true,
lastUsedProvider: true,
lastUsedModel: true,
additionalSystemPrompt: true,
enabledTools: true,
},
});
return { chats: chats.map((chat) => serializeChatLike(chat)) };
return { chats: chats.map((chat) => serializeProviderFields(chat)) };
});
app.post("/v1/chats", async (req) => {
@@ -874,9 +772,20 @@ export async function registerRoutes(app: FastifyInstance) {
}
: undefined,
},
select: chatSummarySelect,
select: {
id: true,
title: true,
createdAt: true,
updatedAt: true,
initiatedProvider: true,
initiatedModel: true,
lastUsedProvider: true,
lastUsedModel: true,
additionalSystemPrompt: true,
enabledTools: true,
},
});
return { chat: serializeChatLike(chat) };
return { chat: serializeProviderFields(chat) };
});
app.patch("/v1/chats/:chatId", async (req) => {
@@ -902,21 +811,23 @@ export async function registerRoutes(app: FastifyInstance) {
if (updated.count === 0) return app.httpErrors.notFound("chat not found");
const chat = await getChatSummary(chatId);
if (!chat) return app.httpErrors.notFound("chat not found");
return { chat };
const chat = await prisma.chat.findUnique({
where: { id: chatId },
select: {
id: true,
title: true,
createdAt: true,
updatedAt: true,
initiatedProvider: true,
initiatedModel: true,
lastUsedProvider: true,
lastUsedModel: true,
additionalSystemPrompt: true,
enabledTools: true,
},
});
app.patch("/v1/chats/:chatId/star", async (req) => {
requireAdmin(req);
const Params = z.object({ chatId: z.string() });
const Body = z.object({ starred: z.boolean() });
const { chatId } = Params.parse(req.params);
const body = Body.parse(req.body ?? {});
const chat = await setChatStarred(chatId, body.starred);
if (!chat) return app.httpErrors.notFound("chat not found");
return { chat };
return { chat: serializeProviderFields(chat) };
});
app.post("/v1/chats/title/suggest", async (req) => {
@@ -929,24 +840,44 @@ export async function registerRoutes(app: FastifyInstance) {
const existing = await prisma.chat.findUnique({
where: { id: body.chatId },
select: chatSummarySelect,
select: {
id: true,
title: true,
createdAt: true,
updatedAt: true,
initiatedProvider: true,
initiatedModel: true,
lastUsedProvider: true,
lastUsedModel: true,
additionalSystemPrompt: true,
enabledTools: true,
},
});
if (!existing) return app.httpErrors.notFound("chat not found");
if (existing.title?.trim()) return { chat: serializeChatLike(existing) };
if (existing.title?.trim()) return { chat: serializeProviderFields(existing) };
const fallback = body.content.split(/\r?\n/)[0]?.trim().slice(0, 48) || "New chat";
const suggestedRaw = await generateChatTitle(body.content);
const title = normalizeSuggestedTitle(suggestedRaw, fallback);
await prisma.chat.updateMany({
where: { id: body.chatId, title: existing.title },
const chat = await prisma.chat.update({
where: { id: body.chatId },
data: { title },
select: {
id: true,
title: true,
createdAt: true,
updatedAt: true,
initiatedProvider: true,
initiatedModel: true,
lastUsedProvider: true,
lastUsedModel: true,
additionalSystemPrompt: true,
enabledTools: true,
},
});
const chat = await getChatSummary(body.chatId);
if (!chat) return app.httpErrors.notFound("chat not found");
return { chat };
return { chat: serializeProviderFields(chat) };
});
app.delete("/v1/chats/:chatId", async (req) => {
@@ -971,69 +902,24 @@ export async function registerRoutes(app: FastifyInstance) {
const searches = await prisma.search.findMany({
orderBy: { updatedAt: "desc" },
take: 100,
select: searchSummarySelect,
select: { id: true, title: true, query: true, createdAt: true, updatedAt: true },
});
return { searches: searches.map((search) => serializeSearchLike(search)) };
return { searches };
});
app.post("/v1/searches", async (req) => {
requireAdmin(req);
const Body = z.object({
title: z.string().optional(),
query: z.string().optional(),
reuseByQuery: z.boolean().optional(),
});
const Body = z.object({ title: z.string().optional(), query: z.string().optional() });
const body = Body.parse(req.body ?? {});
const title = body.title?.trim() || body.query?.trim()?.slice(0, 80);
const query = body.query?.trim() || null;
const queryNormalized = normalizeSearchQuery(query);
if (body.reuseByQuery && queryNormalized) {
const existing = await prisma.search.findFirst({
where: { queryNormalized },
orderBy: { updatedAt: "desc" },
select: {
...searchSummarySelect,
answerText: true,
_count: { select: { results: true } },
},
});
if (existing) {
const { _count, answerText: _answerText, ...search } = existing;
return {
search: serializeSearchLike(search),
reused: true,
cacheHit: isFreshSearchCacheHit({
updatedAt: existing.updatedAt,
resultCount: _count.results,
answerText: existing.answerText,
isActive: activeSearchStreams.has(existing.id),
}),
};
}
}
const search = await prisma.search.create({
data: {
title: title || null,
query,
queryNormalized,
},
select: searchSummarySelect,
select: { id: true, title: true, query: true, createdAt: true, updatedAt: true },
});
return { search: serializeSearchLike(search), reused: false, cacheHit: false };
});
app.patch("/v1/searches/:searchId/star", async (req) => {
requireAdmin(req);
const Params = z.object({ searchId: z.string() });
const Body = z.object({ starred: z.boolean() });
const { searchId } = Params.parse(req.params);
const body = Body.parse(req.body ?? {});
const search = await setSearchStarred(searchId, body.starred);
if (!search) return app.httpErrors.notFound("search not found");
return { search };
});
@@ -1060,13 +946,10 @@ export async function registerRoutes(app: FastifyInstance) {
const search = await prisma.search.findUnique({
where: { id: searchId },
include: {
results: { orderBy: { rank: "asc" } },
projectItems: starredProjectItemsSelect,
},
include: { results: { orderBy: { rank: "asc" } } },
});
if (!search) return app.httpErrors.notFound("search not found");
return { search: serializeSearchLike(search) };
return { search };
});
app.post("/v1/searches/:searchId/chat", async (req) => {
@@ -1102,10 +985,21 @@ export async function registerRoutes(app: FastifyInstance) {
},
},
},
select: chatSummarySelect,
select: {
id: true,
title: true,
createdAt: true,
updatedAt: true,
initiatedProvider: true,
initiatedModel: true,
lastUsedProvider: true,
lastUsedModel: true,
additionalSystemPrompt: true,
enabledTools: true,
},
});
return { chat: serializeChatLike(chat) };
return { chat: serializeProviderFields(chat) };
});
app.post("/v1/searches/:searchId/run", async (req) => {
@@ -1166,7 +1060,6 @@ export async function registerRoutes(app: FastifyInstance) {
where: { id: searchId },
data: {
query,
queryNormalized: normalizeSearchQuery(query),
title: normalizedTitle,
requestId: searchResponse?.requestId ?? null,
rawResponse: searchResponse as any,
@@ -1191,13 +1084,10 @@ export async function registerRoutes(app: FastifyInstance) {
const search = await prisma.search.findUnique({
where: { id: searchId },
include: {
results: { orderBy: { rank: "asc" } },
projectItems: starredProjectItemsSelect,
},
include: { results: { orderBy: { rank: "asc" } } },
});
if (!search) return app.httpErrors.notFound("search not found");
return { search: serializeSearchLike(search) };
return { search };
} catch (err: any) {
await prisma.search.update({
where: { id: searchId },
@@ -1252,14 +1142,10 @@ export async function registerRoutes(app: FastifyInstance) {
const chat = await prisma.chat.findUnique({
where: { id: chatId },
include: {
messages: { orderBy: { createdAt: "asc" } },
calls: { orderBy: { createdAt: "desc" } },
projectItems: starredProjectItemsSelect,
},
include: { messages: { orderBy: { createdAt: "asc" } }, calls: { orderBy: { createdAt: "desc" } } },
});
if (!chat) return app.httpErrors.notFound("chat not found");
return { chat: serializeChatLike(chat) };
return { chat: serializeProviderFields(chat) };
});
app.post("/v1/chats/:chatId/messages", async (req) => {

29
server/src/search-cache.ts
View File

@@ -1,29 +0,0 @@
export const SEARCH_QUERY_CACHE_TTL_MS = 24 * 60 * 60 * 1000;
export function normalizeSearchQuery(value: string | null | undefined) {
const normalized = value?.trim().toLowerCase() ?? "";
return normalized || null;
}
export function hasReusableSearchPayload(candidate: { resultCount: number; answerText?: string | null }) {
return candidate.resultCount > 0 || Boolean(candidate.answerText?.trim());
}
export function isFreshSearchCacheHit(
candidate: {
updatedAt: Date | string;
resultCount: number;
answerText?: string | null;
isActive?: boolean;
},
now = new Date(),
ttlMs = SEARCH_QUERY_CACHE_TTL_MS
) {
if (candidate.isActive) return false;
if (!hasReusableSearchPayload(candidate)) return false;
const updatedAtMs = new Date(candidate.updatedAt).getTime();
if (!Number.isFinite(updatedAtMs)) return false;
return now.getTime() - updatedAtMs <= ttlMs;
}

6
server/src/search/searxng.ts
View File

@@ -1,4 +1,3 @@
import { buildBrowserLikeRequestHeaders } from "../browser-fetch-headers.js";
import { env } from "../env.js";
const SEARXNG_TIMEOUT_MS = 12_000;
@@ -107,7 +106,10 @@ async function fetchSearxng(url: URL, accept: string) {
return await fetch(url, {
redirect: "follow",
signal: controller.signal,
headers: buildBrowserLikeRequestHeaders(accept),
headers: {
"User-Agent": "SybilBot/1.0 (+https://sybil.local)",
Accept: accept,
},
});
} finally {
clearTimeout(timeout);

353
server/tests/chat-tools-streaming.test.ts
View File

@@ -1,9 +1,11 @@
import assert from "node:assert/strict";
import test from "node:test";
import { type ToolAwareStreamingEvent } from "../src/llm/chat-tools.js";
import { completeWithChatCompletionsApi, streamWithChatCompletionsApi } from "../src/llm/protocols/chat-completions-api.js";
import { completeWithMessagesApi, streamWithMessagesApi } from "../src/llm/protocols/messages-api.js";
import { streamWithResponsesApi } from "../src/llm/protocols/responses-api.js";
import {
runPlainChatCompletionsStream,
runToolAwareChatCompletionsStream,
runToolAwareOpenAIChatStream,
type ToolAwareStreamingEvent,
} from "../src/llm/chat-tools.js";
async function* streamFrom(events: any[]) {
for (const event of events) {
@@ -20,7 +22,7 @@ async function collectEvents(iterable: AsyncIterable<ToolAwareStreamingEvent>) {
return events;
}
test("Responses API stream emits text deltas as they arrive", async () => {
test("OpenAI Responses stream emits text deltas as they arrive", async () => {
const outputMessage = {
id: "msg_1",
type: "message",
@@ -50,7 +52,7 @@ test("Responses API stream emits text deltas as they arrive", async () => {
};
const events = await collectEvents(
streamWithResponsesApi({
runToolAwareOpenAIChatStream({
client: client as any,
model: "gpt-test",
messages: [{ role: "user", content: "Say hello" }],
@@ -68,7 +70,7 @@ test("Responses API stream emits text deltas as they arrive", async () => {
assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.text : null, "Hello");
});
test("Chat Completions API stream emits text deltas as they arrive", async () => {
test("OpenAI-compatible Chat Completions stream emits text deltas as they arrive", async () => {
const client = {
chat: {
completions: {
@@ -87,7 +89,7 @@ test("Chat Completions API stream emits text deltas as they arrive", async () =>
};
const events = await collectEvents(
streamWithChatCompletionsApi({
runToolAwareChatCompletionsStream({
client: client as any,
model: "grok-test",
messages: [{ role: "user", content: "Say hello" }],
@@ -122,11 +124,10 @@ test("plain Chat Completions stream does not send Sybil-managed tools", async ()
};
const events = await collectEvents(
streamWithChatCompletionsApi({
runPlainChatCompletionsStream({
client: client as any,
model: "hermes-agent",
messages: [{ role: "user", content: "Say hi" }],
enabledTools: [],
})
);
@@ -139,335 +140,3 @@ test("plain Chat Completions stream does not send Sybil-managed tools", async ()
);
assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.text : null, "Hi");
});
test("fetch_url sends browser-like navigation headers", async () => {
const originalFetch = globalThis.fetch;
const fetchCalls: Array<{ input: RequestInfo | URL; init?: RequestInit }> = [];
globalThis.fetch = (async (input: RequestInfo | URL, init?: RequestInit) => {
fetchCalls.push({ input, init });
return new Response("<!doctype html><title>CPI</title><main>Consumer price index</main>", {
status: 200,
headers: { "content-type": "text/html; charset=utf-8" },
});
}) as typeof fetch;
try {
let requestCount = 0;
const client = {
chat: {
completions: {
create: async () => {
requestCount += 1;
if (requestCount === 1) {
return {
choices: [
{
message: {
tool_calls: [
{
id: "call_1",
type: "function",
function: {
name: "fetch_url",
arguments: JSON.stringify({ url: "https://www.bls.gov/news.release/pdf/cpi.pdf" }),
},
},
],
},
},
],
};
}
return {
choices: [{ message: { content: "Fetched" } }],
};
},
},
},
};
const result = await completeWithChatCompletionsApi({
client: client as any,
model: "grok-test",
messages: [{ role: "user", content: "Fetch CPI PDF" }],
});
assert.equal(result.text, "Fetched");
assert.equal(fetchCalls.length, 1);
assert.equal(String(fetchCalls[0]?.input), "https://www.bls.gov/news.release/pdf/cpi.pdf");
assert.deepEqual(fetchCalls[0]?.init?.headers, {
"User-Agent":
"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/130.0.0.0 Safari/537.36",
Accept: "text/html,application/xhtml+xml,application/xml;q=0.9,application/pdf;q=0.9,*/*;q=0.8",
"Accept-Language": "en-US,en;q=0.9",
"Upgrade-Insecure-Requests": "1",
"Sec-Fetch-Dest": "document",
"Sec-Fetch-Mode": "navigate",
"Sec-Fetch-Site": "none",
"Sec-Fetch-User": "?1",
});
assert.equal(result.toolEvents[0]?.status, "completed");
} finally {
globalThis.fetch = originalFetch;
}
});
test("Messages API executes tool_use blocks and sends tool_result follow-up", async () => {
const originalFetch = globalThis.fetch;
const fetchCalls: Array<{ input: RequestInfo | URL; init?: RequestInit }> = [];
globalThis.fetch = (async (input: RequestInfo | URL, init?: RequestInit) => {
fetchCalls.push({ input, init });
return new Response("<!doctype html><title>Example</title><main>Tool result body</main>", {
status: 200,
headers: { "content-type": "text/html; charset=utf-8" },
});
}) as typeof fetch;
try {
const requestBodies: any[] = [];
const client = {
messages: {
create: async (body: any) => {
requestBodies.push(body);
if (requestBodies.length === 1) {
return {
content: [
{
type: "tool_use",
id: "toolu_1",
name: "fetch_url",
input: { url: "https://example.com/article" },
},
],
usage: { input_tokens: 3, output_tokens: 2 },
};
}
return {
content: [{ type: "text", text: "Fetched" }],
usage: { input_tokens: 5, output_tokens: 1 },
};
},
},
};
const result = await completeWithMessagesApi({
client: client as any,
model: "claude-test",
messages: [{ role: "user", content: "Fetch the article" }],
});
assert.equal(result.text, "Fetched");
assert.equal(fetchCalls.length, 1);
assert.equal(String(fetchCalls[0]?.input), "https://example.com/article");
assert.equal(requestBodies.length, 2);
assert.equal(requestBodies[0]?.model, "claude-test");
assert.equal(requestBodies[0]?.tool_choice?.type, "auto");
const fetchTool = requestBodies[0]?.tools?.find((tool: any) => tool.name === "fetch_url");
assert.equal(fetchTool?.input_schema?.type, "object");
assert.equal(fetchTool?.input_schema?.properties?.url?.type, "string");
const secondMessages = requestBodies[1]?.messages ?? [];
assert.equal(secondMessages.at(-2)?.role, "assistant");
assert.equal(secondMessages.at(-2)?.content?.[0]?.type, "tool_use");
assert.equal(secondMessages.at(-1)?.role, "user");
const toolResult = secondMessages.at(-1)?.content?.[0];
assert.equal(toolResult?.type, "tool_result");
assert.equal(toolResult?.tool_use_id, "toolu_1");
assert.equal(toolResult?.is_error, false);
assert.equal(JSON.parse(toolResult?.content ?? "{}").ok, true);
assert.equal(result.toolEvents[0]?.toolCallId, "toolu_1");
assert.equal(result.toolEvents[0]?.status, "completed");
assert.equal(result.usage?.inputTokens, 8);
assert.equal(result.usage?.outputTokens, 3);
assert.equal(result.usage?.totalTokens, 11);
} finally {
globalThis.fetch = originalFetch;
}
});
test("Chat Completions API stream emits initiated and terminal tool call updates", async () => {
let requestCount = 0;
const client = {
chat: {
completions: {
create: async () => {
requestCount += 1;
if (requestCount === 1) {
return streamFrom([
{
choices: [
{
delta: {
tool_calls: [
{
index: 0,
id: "call_1",
function: {
name: "unknown_tool",
arguments: "{\"query\":\"current weather\"}",
},
},
],
},
finish_reason: "tool_calls",
},
],
},
]);
}
return streamFrom([
{ choices: [{ delta: { content: "Done" } }] },
{ choices: [{ delta: {}, finish_reason: "stop" }] },
]);
},
},
},
};
const events = await collectEvents(
streamWithChatCompletionsApi({
client: client as any,
model: "grok-test",
messages: [{ role: "user", content: "Use a tool" }],
})
);
assert.deepEqual(
events.map((event) => event.type),
["tool_call", "tool_call", "delta", "done"]
);
const toolEvents = events.flatMap((event) => (event.type === "tool_call" ? [event.event] : []));
assert.equal(toolEvents[0]?.toolCallId, "call_1");
assert.equal(toolEvents[0]?.status, "initiated");
assert.equal(toolEvents[0]?.completedAt, undefined);
assert.equal(toolEvents[0]?.durationMs, undefined);
assert.equal(toolEvents[1]?.toolCallId, "call_1");
assert.equal(toolEvents[1]?.status, "failed");
assert.match(toolEvents[1]?.error ?? "", /Unknown tool: unknown_tool/);
assert.equal(typeof toolEvents[1]?.completedAt, "string");
assert.equal(typeof toolEvents[1]?.durationMs, "number");
assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.text : null, "Done");
});
test("Messages API stream emits initiated and terminal tool call updates", async () => {
let requestCount = 0;
const requestBodies: any[] = [];
const client = {
messages: {
create: async (body: any) => {
requestCount += 1;
requestBodies.push(body);
if (requestCount === 1) {
return streamFrom([
{
type: "message_start",
message: {
usage: { input_tokens: 3, output_tokens: 0 },
},
},
{
type: "content_block_start",
index: 0,
content_block: { type: "text", text: "" },
},
{
type: "content_block_delta",
index: 0,
delta: { type: "text_delta", text: "I'll check that." },
},
{ type: "content_block_stop", index: 0 },
{
type: "content_block_start",
index: 1,
content_block: {
type: "tool_use",
id: "toolu_1",
name: "unknown_tool",
input: {},
},
},
{
type: "content_block_delta",
index: 1,
delta: { type: "input_json_delta", partial_json: "{\"query\":\"current weather\"}" },
},
{ type: "content_block_stop", index: 1 },
{
type: "message_delta",
delta: { stop_reason: "tool_use", stop_sequence: null },
usage: { output_tokens: 2 },
},
{ type: "message_stop" },
]);
}
return streamFrom([
{
type: "message_start",
message: {
usage: { input_tokens: 4, output_tokens: 0 },
},
},
{
type: "content_block_start",
index: 0,
content_block: { type: "text", text: "" },
},
{
type: "content_block_delta",
index: 0,
delta: { type: "text_delta", text: "Done" },
},
{ type: "content_block_stop", index: 0 },
{
type: "message_delta",
delta: { stop_reason: "end_turn", stop_sequence: null },
usage: { output_tokens: 1 },
},
{ type: "message_stop" },
]);
},
},
};
const events = await collectEvents(
streamWithMessagesApi({
client: client as any,
model: "claude-test",
messages: [{ role: "user", content: "Use a tool" }],
})
);
assert.deepEqual(
events.map((event) => event.type),
["tool_call", "tool_call", "delta", "done"]
);
assert.equal(requestBodies[0]?.stream, true);
assert.equal(requestBodies[0]?.tools?.some((tool: any) => tool.name === "fetch_url"), true);
const secondMessages = requestBodies[1]?.messages ?? [];
assert.equal(secondMessages.at(-2)?.role, "assistant");
assert.equal(secondMessages.at(-2)?.content?.[0]?.type, "text");
assert.equal(secondMessages.at(-2)?.content?.[0]?.text, "I'll check that.");
assert.equal(secondMessages.at(-2)?.content?.[1]?.type, "tool_use");
assert.deepEqual(secondMessages.at(-2)?.content?.[1]?.input, { query: "current weather" });
const toolResult = secondMessages.at(-1)?.content?.[0];
assert.equal(toolResult?.type, "tool_result");
assert.equal(toolResult?.tool_use_id, "toolu_1");
assert.equal(toolResult?.is_error, true);
assert.match(JSON.parse(toolResult?.content ?? "{}").error ?? "", /Unknown tool: unknown_tool/);
const toolEvents = events.flatMap((event) => (event.type === "tool_call" ? [event.event] : []));
assert.equal(toolEvents[0]?.toolCallId, "toolu_1");
assert.equal(toolEvents[0]?.status, "initiated");
assert.equal(toolEvents[1]?.toolCallId, "toolu_1");
assert.equal(toolEvents[1]?.status, "failed");
assert.match(toolEvents[1]?.error ?? "", /Unknown tool: unknown_tool/);
assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.text : null, "Done");
assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.usage?.inputTokens : null, 7);
assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.usage?.outputTokens : null, 3);
});

6
server/tests/message-content.test.ts
View File

@@ -1,6 +1,6 @@
import assert from "node:assert/strict";
import test from "node:test";
import { buildSystemPromptAugmentation, buildTopLevelSystemPrompt } from "../src/llm/message-content.js";
import { buildSystemPromptAugmentation, getAnthropicSystemPrompt } from "../src/llm/message-content.js";
test("system prompt augmentation includes date and default location", () => {
const prompt = buildSystemPromptAugmentation(undefined, new Date("2026-05-24T15:30:00Z"));
@@ -14,8 +14,8 @@ test("system prompt augmentation uses provided user location", () => {
assert.equal(prompt, "Current date: 2026-05-24.\nUser location: New York, NY.");
});
test("top-level system prompt includes runtime context with existing system messages", () => {
const prompt = buildTopLevelSystemPrompt(
test("Anthropic system prompt includes runtime context with existing system messages", () => {
const prompt = getAnthropicSystemPrompt(
[{ role: "system", content: "Use concise answers." }],
"Los Angeles, CA"
);

36
server/tests/provider-adapters.test.ts
View File

@@ -1,36 +0,0 @@
import assert from "node:assert/strict";
import test from "node:test";
import { describeProviderChatBackend } from "../src/llm/provider-adapters.js";
test("provider backend registry selects chat protocol and managed-tool mode", () => {
assert.deepEqual(describeProviderChatBackend("openai", []), {
provider: "openai",
protocol: "chat-completions",
managedTools: false,
enabledTools: [],
});
assert.deepEqual(describeProviderChatBackend("openai", ["web_search"]), {
provider: "openai",
protocol: "responses",
managedTools: true,
enabledTools: ["web_search"],
});
assert.deepEqual(describeProviderChatBackend("anthropic", ["web_search"]), {
provider: "anthropic",
protocol: "messages",
managedTools: true,
enabledTools: ["web_search"],
});
assert.deepEqual(describeProviderChatBackend("xai", ["web_search"]), {
provider: "xai",
protocol: "chat-completions",
managedTools: true,
enabledTools: ["web_search"],
});
assert.deepEqual(describeProviderChatBackend("hermes-agent", ["web_search"]), {
provider: "hermes-agent",
protocol: "chat-completions",
managedTools: false,
enabledTools: [],
});
});

25
server/tests/search-cache.test.ts
View File

@@ -1,25 +0,0 @@
import assert from "node:assert/strict";
import test from "node:test";
import { SEARCH_QUERY_CACHE_TTL_MS, isFreshSearchCacheHit, normalizeSearchQuery } from "../src/search-cache.js";
test("normalizeSearchQuery trims and lowercases query text", () => {
assert.equal(normalizeSearchQuery(" Bitcoin PRICE "), "bitcoin price");
assert.equal(normalizeSearchQuery(" "), null);
assert.equal(normalizeSearchQuery(null), null);
});
test("isFreshSearchCacheHit requires fresh persisted payload and no active stream", () => {
const now = new Date("2026-05-31T12:00:00.000Z");
assert.equal(
isFreshSearchCacheHit({ updatedAt: new Date(now.getTime() - SEARCH_QUERY_CACHE_TTL_MS + 1), resultCount: 1 }, now),
true
);
assert.equal(
isFreshSearchCacheHit({ updatedAt: new Date(now.getTime() - SEARCH_QUERY_CACHE_TTL_MS - 1), resultCount: 1 }, now),
false
);
assert.equal(isFreshSearchCacheHit({ updatedAt: now, resultCount: 0, answerText: "" }, now), false);
assert.equal(isFreshSearchCacheHit({ updatedAt: now, resultCount: 0, answerText: "answer" }, now), true);
assert.equal(isFreshSearchCacheHit({ updatedAt: now, resultCount: 1, isActive: true }, now), false);
});

24
tui/src/api.ts
View File

@@ -60,22 +60,6 @@ export class SybilApiClient {
return data.chat;
}
async updateChatTitle(chatId: string, title: string) {
const data = await this.request<{ chat: ChatSummary }>(`/v1/chats/${chatId}`, {
method: "PATCH",
body: { title },
});
return data.chat;
}
async updateChatStar(chatId: string, starred: boolean) {
const data = await this.request<{ chat: ChatSummary }>(`/v1/chats/${chatId}/star`, {
method: "PATCH",
body: { starred },
});
return data.chat;
}
async suggestChatTitle(body: { chatId: string; content: string }) {
const data = await this.request<{ chat: ChatSummary }>("/v1/chats/title/suggest", {
method: "POST",
@@ -106,14 +90,6 @@ export class SybilApiClient {
return data.search;
}
async updateSearchStar(searchId: string, starred: boolean) {
const data = await this.request<{ search: SearchSummary }>(`/v1/searches/${searchId}/star`, {
method: "PATCH",
body: { starred },
});
return data.search;
}
async deleteSearch(searchId: string) {
await this.request<{ deleted: true }>(`/v1/searches/${searchId}`, { method: "DELETE" });
}

257
tui/src/index.ts
View File

@@ -20,8 +20,6 @@ type SidebarItem = SidebarSelection & {
title: string;
updatedAt: string;
createdAt: string;
starred: boolean;
starredAt: string | null;
initiatedProvider: Provider | null;
initiatedModel: string | null;
lastUsedProvider: Provider | null;
@@ -32,7 +30,7 @@ type ToolLogMetadata = {
kind: "tool_call";
toolCallId?: string;
toolName?: string;
status?: "initiated" | "completed" | "failed";
status?: "completed" | "failed";
summary?: string;
args?: Record<string, unknown>;
startedAt?: string;
@@ -133,8 +131,6 @@ function buildSidebarItems(items: WorkspaceItem[]): SidebarItem[] {
title: getChatTitle(chat),
updatedAt: chat.updatedAt,
createdAt: chat.createdAt,
starred: chat.starred,
starredAt: chat.starredAt,
initiatedProvider: chat.initiatedProvider,
initiatedModel: chat.initiatedModel,
lastUsedProvider: chat.lastUsedProvider,
@@ -149,8 +145,6 @@ function buildSidebarItems(items: WorkspaceItem[]): SidebarItem[] {
title: getSearchTitle(search),
updatedAt: search.updatedAt,
createdAt: search.createdAt,
starred: search.starred,
starredAt: search.starredAt,
initiatedProvider: null,
initiatedModel: null,
lastUsedProvider: null,
@@ -171,7 +165,13 @@ function isToolCallLogMessage(message: Message) {
}
function buildOptimisticToolMessage(event: ToolCallEvent): Message {
const metadata: ToolLogMetadata = {
return {
id: `temp-tool-${event.toolCallId}`,
createdAt: event.completedAt ?? new Date().toISOString(),
role: "tool",
content: event.summary,
name: event.name,
metadata: {
kind: "tool_call",
toolCallId: event.toolCallId,
toolName: event.name,
@@ -179,37 +179,12 @@ function buildOptimisticToolMessage(event: ToolCallEvent): Message {
summary: event.summary,
args: event.args,
startedAt: event.startedAt,
completedAt: event.completedAt,
durationMs: event.durationMs,
error: event.error ?? null,
resultPreview: event.resultPreview ?? null,
} satisfies ToolLogMetadata,
};
if (event.completedAt) metadata.completedAt = event.completedAt;
if (typeof event.durationMs === "number") metadata.durationMs = event.durationMs;
return {
id: `temp-tool-${event.toolCallId}`,
createdAt: event.completedAt ?? event.startedAt ?? new Date().toISOString(),
role: "tool",
content: event.summary,
name: event.name,
metadata,
};
}
function upsertOptimisticToolMessage(messages: Message[], event: ToolCallEvent) {
const toolMessage = buildOptimisticToolMessage(event);
const existingIndex = messages.findIndex(
(message) => asToolLogMetadata(message.metadata)?.toolCallId === event.toolCallId || message.id === `temp-tool-${event.toolCallId}`
);
if (existingIndex >= 0) {
return messages.map((message, index) => (index === existingIndex ? { ...toolMessage, id: message.id } : message));
}
const assistantIndex = messages.findIndex(
(message, index, all) => index === all.length - 1 && message.id.startsWith("temp-assistant-")
);
if (assistantIndex < 0) return messages.concat(toolMessage);
return [...messages.slice(0, assistantIndex), toolMessage, ...messages.slice(assistantIndex)];
}
function getModelOptions(catalog: ModelCatalogResponse["providers"], provider: Provider) {
@@ -279,7 +254,6 @@ async function main() {
let renderedSidebarItems: SidebarItem[] = [];
let renderedSidebarLines: string[] = [];
let suppressedSidebarSelectEvents = 0;
let isRenamePromptOpen = false;
const screen = blessed.screen({
smartCSR: true,
@@ -387,26 +361,6 @@ async function main() {
},
});
const renamePrompt = (blessed as any).prompt({
parent: screen,
label: " Rename chat ",
border: "line",
tags: true,
keys: true,
vi: true,
mouse: true,
top: "center",
left: "center",
width: "50%",
height: "shrink",
hidden: true,
style: {
border: { fg: "cyan" },
label: { fg: "cyan" },
fg: "white",
},
});
const focusables = [sidebar, transcript, composer] as const;
function getTranscriptViewportHeight() {
@@ -546,13 +500,12 @@ async function main() {
? ["No chats/searches yet. Press n or /. "]
: items.map((item) => {
const kind = item.kind === "chat" ? "C" : "S";
const star = item.starred ? "{yellow-fg}★{/yellow-fg} " : " ";
const title = truncate(item.title, 36);
const initiatedLabel =
item.kind === "chat" && item.initiatedModel
? ` | ${getProviderLabel(item.initiatedProvider)} ${truncate(item.initiatedModel, 16)}`
: "";
return `${star}${kind} ${title} {gray-fg}${formatDate(item.updatedAt)}${escapeTags(initiatedLabel)}{/gray-fg}`;
return `${kind} ${title} {gray-fg}${formatDate(item.updatedAt)}${escapeTags(initiatedLabel)}{/gray-fg}`;
});
const linesChanged =
@@ -621,12 +574,7 @@ async function main() {
for (const message of messages) {
const toolMeta = asToolLogMetadata(message.metadata);
if (message.role === "tool" && toolMeta) {
const prefix =
toolMeta.status === "failed"
? "{red-fg}[tool failed]{/red-fg}"
: toolMeta.status === "initiated"
? "{yellow-fg}[tool running]{/yellow-fg}"
: "{cyan-fg}[tool]{/cyan-fg}";
const prefix = toolMeta.status === "failed" ? "{red-fg}[tool failed]{/red-fg}" : "{cyan-fg}[tool]{/cyan-fg}";
const summary = toolMeta.summary?.trim() || message.content.trim() || "Tool call executed.";
parts.push(`${prefix} ${escapeTags(summary)}`);
continue;
@@ -732,7 +680,7 @@ async function main() {
const top = `{bold}${escapeTags(getSelectedTitle())}{/bold} {gray-fg}- Sybil TUI${modeLabel}${isSearchMode ? " • Exa Search" : ""}{/gray-fg}`;
let controls =
"{gray-fg}Controls:{/gray-fg} [tab] focus [esc] command mode [↑/↓] highlight [enter] send/select [n] new chat [/] new search [s] star [r] rename [d] delete [C-r] refresh [q] quit";
"{gray-fg}Controls:{/gray-fg} [tab] focus [esc] command mode [↑/↓] highlight [enter] send/select [n] new chat [/] new search [d] delete [q] quit";
if (!isSearchMode) {
controls += `\n{gray-fg}Model:{/gray-fg} provider {cyan-fg}${provider}{/cyan-fg} [p] model {cyan-fg}${escapeTags(model)}{/cyan-fg} [m]`;
controls += providerModelOptions.length === 0 ? " {red-fg}(no models){/red-fg}" : "";
@@ -894,27 +842,6 @@ async function main() {
composer.readInput();
}
function shouldIgnoreGlobalShortcut() {
return isRenamePromptOpen || isTextInputFocused(screen, composer);
}
function promptForChatTitle(currentTitle: string) {
isRenamePromptOpen = true;
updateUI();
return new Promise<string | null>((resolve) => {
renamePrompt.input("Title:", currentTitle, (err: Error | null, value: string | null) => {
isRenamePromptOpen = false;
renamePrompt.hide();
screen.render();
if (err || value === null || value === undefined) {
resolve(null);
return;
}
resolve(value);
});
});
}
function cycleFocus(step: 1 | -1) {
const focused = screen.focused;
const currentIndex = focusables.findIndex((node) => node === focused);
@@ -983,20 +910,10 @@ async function main() {
pendingTitleGeneration.add(chatId);
try {
const updated = await api.suggestChatTitle({ chatId, content });
chats = chats.map((chat) => (chat.id === updated.id ? updated : chat));
chats = chats.map((chat) => (chat.id === updated.id ? { ...chat, title: updated.title, updatedAt: updated.updatedAt } : chat));
workspaceItems = workspaceItems.map((item) => (item.type === "chat" && item.id === updated.id ? chatWorkspaceItem(updated) : item));
if (selectedChat?.id === updated.id) {
selectedChat = {
...selectedChat,
title: updated.title,
updatedAt: updated.updatedAt,
starred: updated.starred,
starredAt: updated.starredAt,
initiatedProvider: updated.initiatedProvider,
initiatedModel: updated.initiatedModel,
lastUsedProvider: updated.lastUsedProvider,
lastUsedModel: updated.lastUsedModel,
};
selectedChat = { ...selectedChat, title: updated.title, updatedAt: updated.updatedAt };
}
updateUI();
} catch {
@@ -1047,8 +964,6 @@ async function main() {
title: chat.title,
createdAt: chat.createdAt,
updatedAt: chat.updatedAt,
starred: chat.starred,
starredAt: chat.starredAt,
initiatedProvider: chat.initiatedProvider,
initiatedModel: chat.initiatedModel,
lastUsedProvider: chat.lastUsedProvider,
@@ -1107,7 +1022,29 @@ async function main() {
},
onToolCall: (payload) => {
if (!pendingChatState) return;
pendingChatState = { ...pendingChatState, messages: upsertOptimisticToolMessage(pendingChatState.messages, payload) };
const alreadyPresent = pendingChatState.messages.some(
(message) =>
asToolLogMetadata(message.metadata)?.toolCallId === payload.toolCallId || message.id === `temp-tool-${payload.toolCallId}`
);
if (alreadyPresent) return;
const toolMessage = buildOptimisticToolMessage(payload);
const assistantIndex = pendingChatState.messages.findIndex(
(message, index, all) => index === all.length - 1 && message.id.startsWith("temp-assistant-")
);
if (assistantIndex < 0) {
pendingChatState = { ...pendingChatState, messages: pendingChatState.messages.concat(toolMessage) };
} else {
pendingChatState = {
...pendingChatState,
messages: [
...pendingChatState.messages.slice(0, assistantIndex),
toolMessage,
...pendingChatState.messages.slice(assistantIndex),
],
};
}
queueTranscriptScrollToBottomIfFollowing();
updateUI();
@@ -1203,8 +1140,6 @@ async function main() {
query,
createdAt: nowIso,
updatedAt: nowIso,
starred: false,
starredAt: null,
requestId: null,
latencyMs: null,
error: null,
@@ -1367,88 +1302,6 @@ async function main() {
await refreshCollections({ loadSelection: true, scrollToBottomOnLoad: true });
}
async function handleRenameSelection() {
if (!selectedItem || selectedItem.kind !== "chat") return;
const chatId = selectedItem.id;
const summary = chats.find((chat) => chat.id === chatId);
const currentTitle = selectedChat?.id === chatId ? getChatTitle(selectedChat, selectedChat.messages) : summary ? getChatTitle(summary) : "New chat";
const value = await promptForChatTitle(currentTitle);
const title = value?.trim();
if (!title) {
updateUI();
return;
}
setError(null);
const updated = await api.updateChatTitle(chatId, title);
chats = [updated, ...chats.filter((chat) => chat.id !== updated.id)];
workspaceItems = upsertWorkspaceItem(workspaceItems, chatWorkspaceItem(updated));
if (selectedChat?.id === updated.id) {
selectedChat = {
...selectedChat,
title: updated.title,
updatedAt: updated.updatedAt,
initiatedProvider: updated.initiatedProvider,
initiatedModel: updated.initiatedModel,
lastUsedProvider: updated.lastUsedProvider,
lastUsedModel: updated.lastUsedModel,
};
}
updateUI();
}
async function handleToggleStarSelection() {
if (!selectedItem) return;
const currentItem = getSidebarItems().find((item) => item.kind === selectedItem?.kind && item.id === selectedItem?.id);
const nextStarred = !currentItem?.starred;
setError(null);
if (selectedItem.kind === "chat") {
const updated = await api.updateChatStar(selectedItem.id, nextStarred);
chats = chats.map((chat) => (chat.id === updated.id ? updated : chat));
if (!chats.some((chat) => chat.id === updated.id)) chats = [updated, ...chats];
workspaceItems = workspaceItems.map((item) => (item.type === "chat" && item.id === updated.id ? chatWorkspaceItem(updated) : item));
if (!workspaceItems.some((item) => item.type === "chat" && item.id === updated.id)) {
workspaceItems = [chatWorkspaceItem(updated), ...workspaceItems];
}
if (selectedChat?.id === updated.id) {
selectedChat = {
...selectedChat,
title: updated.title,
updatedAt: updated.updatedAt,
starred: updated.starred,
starredAt: updated.starredAt,
initiatedProvider: updated.initiatedProvider,
initiatedModel: updated.initiatedModel,
lastUsedProvider: updated.lastUsedProvider,
lastUsedModel: updated.lastUsedModel,
};
}
} else {
const updated = await api.updateSearchStar(selectedItem.id, nextStarred);
searches = searches.map((search) => (search.id === updated.id ? updated : search));
if (!searches.some((search) => search.id === updated.id)) searches = [updated, ...searches];
workspaceItems = workspaceItems.map((item) => (item.type === "search" && item.id === updated.id ? searchWorkspaceItem(updated) : item));
if (!workspaceItems.some((item) => item.type === "search" && item.id === updated.id)) {
workspaceItems = [searchWorkspaceItem(updated), ...workspaceItems];
}
if (selectedSearch?.id === updated.id) {
selectedSearch = {
...selectedSearch,
title: updated.title,
query: updated.query,
updatedAt: updated.updatedAt,
starred: updated.starred,
starredAt: updated.starredAt,
};
}
}
updateUI();
}
function cycleProvider() {
const visibleProviders = getVisibleProviders(modelCatalog);
const cycleProviders = visibleProviders.length ? visibleProviders : BASE_PROVIDERS;
@@ -1534,18 +1387,18 @@ async function main() {
});
screen.key(["q"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
screen.destroy();
process.exit(0);
});
screen.key(["tab"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
cycleFocus(1);
});
screen.key(["S-tab", "backtab"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
cycleFocus(-1);
});
@@ -1562,50 +1415,36 @@ async function main() {
});
screen.key(["n"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
handleCreateChat();
});
screen.key(["/"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
handleCreateSearch();
});
screen.key(["d"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
void runAction(async () => {
await handleDeleteSelection();
});
});
screen.key(["s"], () => {
if (shouldIgnoreGlobalShortcut()) return;
void runAction(async () => {
await handleToggleStarSelection();
});
});
screen.key(["p"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
if (getIsSearchMode() || isSending) return;
cycleProvider();
});
screen.key(["m"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
if (getIsSearchMode() || isSending) return;
cycleModel();
});
screen.key(["r"], () => {
if (shouldIgnoreGlobalShortcut()) return;
void runAction(async () => {
await handleRenameSelection();
});
});
screen.key(["C-r"], () => {
if (shouldIgnoreGlobalShortcut()) return;
if (isTextInputFocused(screen, composer)) return;
void runAction(async () => {
await refreshCollections({ loadSelection: true });
await refreshModels();

14
tui/src/types.ts
View File

@@ -15,8 +15,6 @@ export type ChatSummary = {
title: string | null;
createdAt: string;
updatedAt: string;
starred: boolean;
starredAt: string | null;
initiatedProvider: Provider | null;
initiatedModel: string | null;
lastUsedProvider: Provider | null;
@@ -29,8 +27,6 @@ export type SearchSummary = {
query: string | null;
createdAt: string;
updatedAt: string;
starred: boolean;
starredAt: string | null;
};
export type ChatWorkspaceItem = ChatSummary & {
@@ -55,12 +51,12 @@ export type Message = {
export type ToolCallEvent = {
toolCallId: string;
name: string;
status: "initiated" | "completed" | "failed";
status: "completed" | "failed";
summary: string;
args: Record<string, unknown>;
startedAt: string;
completedAt?: string;
durationMs?: number;
completedAt: string;
durationMs: number;
error?: string;
resultPreview?: string;
};
@@ -70,8 +66,6 @@ export type ChatDetail = {
title: string | null;
createdAt: string;
updatedAt: string;
starred: boolean;
starredAt: string | null;
initiatedProvider: Provider | null;
initiatedModel: string | null;
lastUsedProvider: Provider | null;
@@ -101,8 +95,6 @@ export type SearchDetail = {
query: string | null;
createdAt: string;
updatedAt: string;
starred: boolean;
starredAt: string | null;
requestId: string | null;
latencyMs: number | null;
error: string | null;

8
web/index.html
View File

@@ -3,18 +3,12 @@
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no, viewport-fit=cover, interactive-widget=resizes-content" />
<meta name="description" content="Sybil chat and search workspace" />
<meta name="application-name" content="Sybil" />
<meta name="theme-color" content="#0f172a" />
<meta name="mobile-web-app-capable" content="yes" />
<meta name="apple-mobile-web-app-capable" content="yes" />
<meta name="apple-mobile-web-app-status-bar-style" content="black-translucent" />
<meta name="apple-mobile-web-app-status-bar-style" content="default" />
<meta name="apple-mobile-web-app-title" content="Sybil" />
<meta name="format-detection" content="telephone=no" />
<link rel="manifest" href="/manifest.webmanifest" />
<link rel="apple-touch-icon" sizes="180x180" href="/icons/apple-touch-icon.png" />
<link rel="icon" type="image/png" sizes="32x32" href="/icons/favicon-32.png" />
<link rel="icon" type="image/png" sizes="192x192" href="/icons/icon-192.png" />
<link rel="search" type="application/opensearchdescription+xml" title="Sybil Search" href="/opensearch.xml" />
<title>Sybil</title>
</head>

BIN
web/public/icons/apple-touch-icon.png
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 49 KiB

BIN
web/public/icons/favicon-32.png
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 2.4 KiB

BIN
web/public/icons/icon-192.png
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 56 KiB

BIN
web/public/icons/icon-512.png
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 258 KiB

BIN
web/public/icons/icon-maskable-512.png
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 258 KiB

29
web/public/manifest.webmanifest
View File

@@ -1,32 +1,9 @@
{
"id": "/",
"name": "Sybil",
"short_name": "Sybil",
"description": "Sybil chat and search workspace",
"start_url": "/",
"scope": "/",
"display": "fullscreen",
"display_override": ["fullscreen", "standalone"],
"background_color": "#0b0718",
"theme_color": "#0f172a",
"icons": [
{
"src": "/icons/icon-192.png",
"sizes": "192x192",
"type": "image/png",
"purpose": "any"
},
{
"src": "/icons/icon-512.png",
"sizes": "512x512",
"type": "image/png",
"purpose": "any"
},
{
"src": "/icons/icon-maskable-512.png",
"sizes": "512x512",
"type": "image/png",
"purpose": "maskable"
}
]
"display": "standalone",
"background_color": "#ffffff",
"theme_color": "#0f172a"
}

12
web/public/sw.js
View File

@@ -1,12 +0,0 @@
self.addEventListener("install", () => {
self.skipWaiting();
});
self.addEventListener("activate", (event) => {
event.waitUntil(self.clients.claim());
});
self.addEventListener("fetch", (event) => {
if (event.request.mode !== "navigate") return;
event.respondWith(fetch(event.request));
});

807
web/src/App.tsx
View File

File diff suppressed because it is too large Load Diff

2
web/src/components/auth/auth-screen.tsx
View File

@@ -12,7 +12,7 @@ type Props = {
export function AuthScreen({ authTokenInput, setAuthTokenInput, isSigningIn, authError, onSignIn }: Props) {
return (
<div className="app-grid-surface app-safe-pad flex h-full items-center justify-center">
<div className="app-grid-surface flex h-full items-center justify-center p-4">
<div className="glass-panel w-full max-w-md rounded-2xl border border-violet-300/18 p-6">
<div className="mb-6">
<div className="sybil-wordmark bg-[linear-gradient(90deg,#ff8df8,#9a6dff_54%,#67dfff)] bg-clip-text text-3xl text-transparent">

379
web/src/components/chat/chat-messages-panel.tsx
View File

@@ -1,10 +1,8 @@
import { useEffect, useMemo, useRef, useState } from "preact/hooks";
import type { ComponentChildren, JSX } from "preact";
import { cn } from "@/lib/utils";
import { ChatAttachmentList } from "@/components/chat/chat-attachment-list";
import { getMessageAttachments, type Message } from "@/lib/api";
import { MarkdownContent } from "@/components/markdown/markdown-content";
import { ChevronDown, ChevronUp, Globe2, Link2, Wrench } from "lucide-preact";
import { Globe2, Link2, Wrench } from "lucide-preact";
type Props = {
messages: Message[];
@@ -16,7 +14,7 @@ type ToolLogMetadata = {
kind: "tool_call";
toolCallId?: string;
toolName?: string;
status?: "initiated" | "completed" | "failed";
status?: "completed" | "failed";
summary?: string;
args?: Record<string, unknown>;
startedAt?: string;
@@ -73,40 +71,9 @@ function formatToolTimestamp(...values: Array<string | null | undefined>) {
return new Intl.DateTimeFormat(undefined, { hour: "numeric", minute: "2-digit" }).format(new Date(value));
}
type ToolCallVisualState = "initiated" | "completed" | "failed";
type MessageRenderItem = { kind: "message"; message: Message } | { kind: "tool_group"; key: string; messages: Message[] };
type ToolStackStyle = JSX.CSSProperties & {
"--tool-stack-x"?: string;
"--tool-stack-y"?: string;
"--tool-stack-z"?: string;
"--tool-stack-scale"?: string;
"--tool-stack-opacity"?: string;
"--tool-stack-delay"?: string;
"--tool-stack-from-transform"?: string;
"--tool-stack-to-transform"?: string;
"--tool-stack-from-opacity"?: string;
"--tool-stack-to-opacity"?: string;
};
type ToolStackContainerStyle = JSX.CSSProperties & {
"--tool-stack-from-height"?: string;
"--tool-stack-to-height"?: string;
};
type ToolStackMotionDirection = "expand" | "collapse" | null;
const COLLAPSED_TOOL_STACK_LIMIT = 4;
const TOOL_STACK_CARD_HEIGHT = 62;
const TOOL_STACK_CARD_GAP = 10;
const TOOL_STACK_LAYOUT_ANIMATION_MS = 340;
function getToolVisualState(metadata: ToolLogMetadata): ToolCallVisualState {
if (metadata.status === "failed") return "failed";
if (metadata.status === "initiated") return "initiated";
return "completed";
}
function getToolDetailLabel(message: Message, metadata: ToolLogMetadata, state: ToolCallVisualState) {
function getToolDetailLabel(message: Message, metadata: ToolLogMetadata, isFailed: boolean) {
return [
state === "failed" ? "Failed" : state === "initiated" ? "Running" : "Completed",
isFailed ? "Failed" : "Completed",
formatDuration(metadata.durationMs),
formatToolTimestamp(message.createdAt, metadata.completedAt, metadata.startedAt),
]
@@ -114,343 +81,53 @@ function getToolDetailLabel(message: Message, metadata: ToolLogMetadata, state:
.join(" • ");
}
function buildMessageRenderItems(messages: Message[]) {
const items: MessageRenderItem[] = [];
let toolRun: Message[] = [];
const flushToolRun = () => {
if (!toolRun.length) return;
if (toolRun.length === 1) {
items.push({ kind: "message", message: toolRun[0] });
} else {
items.push({ kind: "tool_group", key: toolRun[0].id, messages: toolRun });
}
toolRun = [];
};
for (const message of messages) {
if (message.role === "tool" && asToolLogMetadata(message.metadata)) {
toolRun.push(message);
continue;
}
flushToolRun();
items.push({ kind: "message", message });
}
flushToolRun();
return items;
}
function getToolCallMessageIDs(messages: Message[]) {
const ids = new Set<string>();
for (const message of messages) {
if (message.role === "tool" && asToolLogMetadata(message.metadata)) ids.add(message.id);
}
return ids;
}
function getToolStackHeight(messageCount: number, expanded: boolean) {
const visibleCount = Math.min(messageCount, COLLAPSED_TOOL_STACK_LIMIT);
return expanded
? `${TOOL_STACK_CARD_HEIGHT + Math.max(0, messageCount - 1) * (TOOL_STACK_CARD_HEIGHT + TOOL_STACK_CARD_GAP)}px`
: `${TOOL_STACK_CARD_HEIGHT + Math.max(0, visibleCount - 1) * TOOL_STACK_CARD_GAP}px`;
}
function getToolStackContainerStyle(messageCount: number, expanded: boolean, motionDirection: ToolStackMotionDirection): ToolStackContainerStyle {
const collapsedHeight = getToolStackHeight(messageCount, false);
const expandedHeight = getToolStackHeight(messageCount, true);
const targetHeight = expanded ? expandedHeight : collapsedHeight;
const fromHeight = motionDirection === "expand" ? collapsedHeight : motionDirection === "collapse" ? expandedHeight : targetHeight;
return {
"--tool-stack-from-height": fromHeight,
"--tool-stack-to-height": targetHeight,
height: targetHeight,
};
}
function getExpandedToolLayout(index: number, messageCount: number) {
const y = `${index * (TOOL_STACK_CARD_HEIGHT + TOOL_STACK_CARD_GAP)}px`;
return {
opacity: "1",
transform: `translate3d(0px, ${y}, 0px) scale(1)`,
x: "0px",
y,
z: "0px",
scale: "1",
zIndex: messageCount - index,
};
}
function getCollapsedToolLayout(index: number, messageCount: number) {
const depth = messageCount - index - 1;
const visibleDepth = Math.min(depth, COLLAPSED_TOOL_STACK_LIMIT - 1);
const isHidden = depth >= COLLAPSED_TOOL_STACK_LIMIT;
const visibleCount = Math.min(messageCount, COLLAPSED_TOOL_STACK_LIMIT);
const x = `${visibleDepth * 11}px`;
const y = `${visibleDepth * TOOL_STACK_CARD_GAP}px`;
const z = `${visibleDepth * -36}px`;
const scale = `${Math.max(0.88, 1 - visibleDepth * 0.035)}`;
const opacity = isHidden ? "0" : `${Math.max(0.34, 1 - visibleDepth * 0.22)}`;
return {
opacity,
transform: `translate3d(${x}, ${y}, ${z}) scale(${scale})`,
x,
y,
z,
scale,
zIndex: isHidden ? 0 : visibleCount - visibleDepth,
};
}
function getToolStackStyle(index: number, messageCount: number, expanded: boolean, motionDirection: ToolStackMotionDirection): ToolStackStyle {
const expandedLayout = getExpandedToolLayout(index, messageCount);
const collapsedLayout = getCollapsedToolLayout(index, messageCount);
const targetLayout = expanded ? expandedLayout : collapsedLayout;
const fromLayout = motionDirection === "expand" ? collapsedLayout : motionDirection === "collapse" ? expandedLayout : targetLayout;
return {
"--tool-stack-x": targetLayout.x,
"--tool-stack-y": targetLayout.y,
"--tool-stack-z": targetLayout.z,
"--tool-stack-scale": targetLayout.scale,
"--tool-stack-opacity": targetLayout.opacity,
"--tool-stack-delay": `${Math.min(messageCount - index - 1, COLLAPSED_TOOL_STACK_LIMIT - 1) * 34}ms`,
"--tool-stack-from-transform": fromLayout.transform,
"--tool-stack-to-transform": targetLayout.transform,
"--tool-stack-from-opacity": fromLayout.opacity,
"--tool-stack-to-opacity": targetLayout.opacity,
opacity: targetLayout.opacity,
transform: targetLayout.transform,
zIndex: targetLayout.zIndex,
};
}
function ToolCallCard({
message,
className,
style,
}: {
message: Message;
className?: string;
style?: JSX.CSSProperties;
}) {
const toolLogMetadata = asToolLogMetadata(message.metadata);
if (!toolLogMetadata) return null;
const iconKind = getToolIconName(toolLogMetadata.toolName ?? message.name);
const Icon = iconKind === "search" ? Globe2 : iconKind === "fetch" ? Link2 : Wrench;
const toolState = getToolVisualState(toolLogMetadata);
const isFailed = toolState === "failed";
const isInitiated = toolState === "initiated";
const toolSummary = getToolSummary(message, toolLogMetadata);
const toolLabel = getToolLabel(message, toolLogMetadata);
const toolDetailLabel = getToolDetailLabel(message, toolLogMetadata, toolState);
export function ChatMessagesPanel({ messages, isLoading, isSending }: Props) {
const hasPendingAssistant = messages.some((message) => message.id.startsWith("temp-assistant-") && message.content.trim().length === 0);
return (
<>
{isLoading && messages.length === 0 ? <p className="text-sm text-muted-foreground">Loading messages...</p> : null}
<div className="mx-auto max-w-4xl space-y-6">
{messages.map((message) => {
const toolLogMetadata = asToolLogMetadata(message.metadata);
if (message.role === "tool" && toolLogMetadata) {
const iconKind = getToolIconName(toolLogMetadata.toolName ?? message.name);
const Icon = iconKind === "search" ? Globe2 : iconKind === "fetch" ? Link2 : Wrench;
const isFailed = toolLogMetadata.status === "failed";
const toolSummary = getToolSummary(message, toolLogMetadata);
const toolLabel = getToolLabel(message, toolLogMetadata);
const toolDetailLabel = getToolDetailLabel(message, toolLogMetadata, isFailed);
return (
<div key={message.id} className="flex justify-start">
<div
className={cn(
"inline-flex min-w-0 items-start gap-3 overflow-hidden rounded-xl border px-3 py-2.5 shadow-[inset_0_1px_0_hsl(180_100%_88%_/_0.06)]",
"inline-flex max-w-[85%] min-w-0 items-start gap-3 overflow-hidden rounded-xl border px-3 py-2.5 shadow-[inset_0_1px_0_hsl(180_100%_88%_/_0.06)]",
isFailed
? "border-rose-400/44 bg-[linear-gradient(90deg,hsl(350_64%_20%),hsl(342_58%_9%))]"
: isInitiated
? "border-amber-300/44 bg-[linear-gradient(90deg,hsl(43_72%_20%),hsl(260_48%_13%))]"
: "border-cyan-400/44 bg-[linear-gradient(90deg,hsl(184_82%_14%),hsl(208_66%_10%))]",
className
? "border-rose-400/34 bg-[linear-gradient(90deg,hsl(350_72%_44%_/_0.18),hsl(342_66%_9%_/_0.72))]"
: "border-cyan-400/34 bg-[linear-gradient(90deg,hsl(184_89%_21%_/_0.70),hsl(208_66%_12%_/_0.78))]"
)}
style={style}
title={`${toolSummary}\n${toolLabel} • ${toolDetailLabel}`}
>
<span
className={cn(
"mt-0.5 flex h-[30px] w-[30px] shrink-0 items-center justify-center rounded-lg border",
isFailed
? "border-rose-400/34 bg-rose-400/13 text-rose-300"
: isInitiated
? "border-amber-300/34 bg-amber-300/13 text-amber-200"
: "border-cyan-300/34 bg-cyan-300/13 text-cyan-300"
isFailed ? "border-rose-400/34 bg-rose-400/13 text-rose-300" : "border-cyan-300/34 bg-cyan-300/13 text-cyan-300"
)}
>
<Icon className="h-4 w-4" />
</span>
<span className="min-w-0 flex-1 space-y-1">
<span className={cn("block truncate text-sm leading-5", isFailed ? "text-rose-200" : "text-violet-50/95")}>{toolSummary}</span>
<span className={cn("block truncate text-sm leading-5", isFailed ? "text-rose-200" : "text-violet-50/95")}>
{toolSummary}
</span>
<span className="flex min-w-0 items-center gap-1.5 text-[11px] leading-4">
<span className={cn("min-w-0 truncate font-semibold", isFailed ? "text-rose-300/85" : isInitiated ? "text-amber-200/90" : "text-cyan-200/90")}>
<span className={cn("min-w-0 truncate font-semibold", isFailed ? "text-rose-300/85" : "text-cyan-200/90")}>
{toolLabel}
</span>
<span className="min-w-0 truncate text-violet-200/64">{toolDetailLabel}</span>
</span>
</span>
</div>
);
}
function ToolCallStackCardSurface({
messageID,
animateEntry,
isHidden,
children,
}: {
messageID: string;
animateEntry: boolean;
isHidden: boolean;
children: ComponentChildren;
}) {
const [shouldAnimateEntry] = useState(() => animateEntry);
return (
<div
className={cn("tool-call-stack-card-surface", shouldAnimateEntry && !isHidden && "tool-call-stack-card-enter")}
data-tool-stack-card-id={messageID}
>
{children}
</div>
);
}
function ToolCallStack({
groupKey,
messages,
expanded,
entryMessageIDs,
onToggle,
}: {
groupKey: string;
messages: Message[];
expanded: boolean;
entryMessageIDs: Set<string>;
onToggle: (groupKey: string) => void;
}) {
const hiddenCount = Math.max(0, messages.length - COLLAPSED_TOOL_STACK_LIMIT);
const countLabel = `${messages.length} tool ${messages.length === 1 ? "call" : "calls"}`;
const [motionDirection, setMotionDirection] = useState<ToolStackMotionDirection>(null);
const [motionRevision, setMotionRevision] = useState(0);
const motionResetTimerRef = useRef<number | null>(null);
const handleToggle = () => {
setMotionDirection(expanded ? "collapse" : "expand");
setMotionRevision((current) => current + 1);
if (typeof window !== "undefined") {
if (motionResetTimerRef.current !== null) window.clearTimeout(motionResetTimerRef.current);
motionResetTimerRef.current = window.setTimeout(() => {
setMotionDirection(null);
motionResetTimerRef.current = null;
}, TOOL_STACK_LAYOUT_ANIMATION_MS + 60);
}
onToggle(groupKey);
};
return (
<div className="flex justify-start">
<div
className={cn(
"tool-call-stack-shell relative w-full max-w-[85%] min-w-0 pr-10",
motionDirection && (motionRevision % 2 === 0 ? "tool-call-stack-shell-layout-a" : "tool-call-stack-shell-layout-b")
)}
data-tool-stack-group={groupKey}
data-expanded={expanded ? "true" : "false"}
style={getToolStackContainerStyle(messages.length, expanded, motionDirection)}
>
{messages.map((message, index) => {
const depth = messages.length - index - 1;
const isHidden = !expanded && depth >= COLLAPSED_TOOL_STACK_LIMIT;
const shouldAnimateEntry = entryMessageIDs.has(message.id) && !isHidden;
return (
<div
key={message.id}
className={cn(
"tool-call-stack-card absolute left-0 right-10 top-0 w-auto max-w-none",
motionDirection && (motionRevision % 2 === 0 ? "tool-call-stack-card-layout-a" : "tool-call-stack-card-layout-b"),
isHidden && "pointer-events-none"
)}
style={getToolStackStyle(index, messages.length, expanded, motionDirection)}
aria-hidden={isHidden ? "true" : undefined}
>
<ToolCallStackCardSurface messageID={message.id} animateEntry={shouldAnimateEntry} isHidden={isHidden}>
<ToolCallCard message={message} className="tool-call-stack-card-glass w-full max-w-full" />
</ToolCallStackCardSurface>
</div>
);
})}
{!expanded && hiddenCount ? (
<span className="absolute bottom-1 right-10 z-20 rounded-full border border-cyan-300/30 bg-slate-950/86 px-2 py-0.5 text-[10px] font-semibold leading-none text-cyan-100 shadow-sm">
+{hiddenCount}
</span>
) : null}
<button
type="button"
className="tool-call-stack-toggle absolute right-0 top-2 z-20 flex h-8 w-8 items-center justify-center rounded-full"
aria-expanded={expanded ? "true" : "false"}
aria-label={`${expanded ? "Collapse" : "Expand"} ${countLabel}`}
title={`${expanded ? "Collapse" : "Expand"} ${countLabel}`}
onClick={handleToggle}
>
{expanded ? <ChevronUp className="h-4 w-4" /> : <ChevronDown className="h-4 w-4" />}
</button>
</div>
</div>
);
}
export function ChatMessagesPanel({ messages, isLoading, isSending }: Props) {
const hasPendingAssistant = messages.some((message) => message.id.startsWith("temp-assistant-") && message.content.trim().length === 0);
const renderItems = useMemo(() => buildMessageRenderItems(messages), [messages]);
const toolCallMessageIDs = useMemo(() => getToolCallMessageIDs(messages), [messages]);
const seenToolCallMessageIDsRef = useRef<Set<string> | null>(null);
const entryToolCallMessageIDs = useMemo(() => {
const seenIDs = seenToolCallMessageIDsRef.current;
if (!seenIDs) return new Set<string>();
const entryIDs = new Set<string>();
for (const id of toolCallMessageIDs) {
if (!seenIDs.has(id)) entryIDs.add(id);
}
return entryIDs;
}, [toolCallMessageIDs]);
const [expandedToolGroups, setExpandedToolGroups] = useState<Set<string>>(() => new Set());
useEffect(() => {
if (!toolCallMessageIDs.size) return;
const seenIDs = seenToolCallMessageIDsRef.current ?? new Set<string>();
for (const id of toolCallMessageIDs) seenIDs.add(id);
seenToolCallMessageIDsRef.current = seenIDs;
}, [toolCallMessageIDs]);
const toggleToolGroup = (groupKey: string) => {
setExpandedToolGroups((current) => {
const next = new Set(current);
if (next.has(groupKey)) next.delete(groupKey);
else next.add(groupKey);
return next;
});
};
return (
<>
{isLoading && messages.length === 0 ? <p className="text-sm text-muted-foreground">Loading messages...</p> : null}
<div className="mx-auto max-w-4xl space-y-6">
{renderItems.map((item) => {
if (item.kind === "tool_group") {
return (
<ToolCallStack
key={`tool-group-${item.key}`}
groupKey={item.key}
messages={item.messages}
expanded={expandedToolGroups.has(item.key)}
entryMessageIDs={entryToolCallMessageIDs}
onToggle={toggleToolGroup}
/>
);
}
const { message } = item;
const toolLogMetadata = asToolLogMetadata(message.metadata);
if (message.role === "tool" && toolLogMetadata) {
return (
<div key={message.id} className="flex justify-start">
<ToolCallCard message={message} className="max-w-[85%]" />
</div>
);
}

195
web/src/index.css
View File

@@ -14,10 +14,6 @@
:root {
color-scheme: dark;
--safe-area-top: env(safe-area-inset-top, 0px);
--safe-area-right: env(safe-area-inset-right, 0px);
--safe-area-bottom: env(safe-area-inset-bottom, 0px);
--safe-area-left: env(safe-area-inset-left, 0px);
--background: 235 45% 4%;
--foreground: 258 36% 96%;
--muted: 246 30% 13%;
@@ -44,15 +40,6 @@ html,
body,
#app {
height: 100%;
width: 100%;
}
@supports (height: 100dvh) {
html,
body,
#app {
height: 100dvh;
}
}
body {
@@ -62,8 +49,6 @@ body {
linear-gradient(90deg, hsl(187 92% 49% / 0.08), transparent 24%, hsl(264 92% 59% / 0.12) 74%, transparent),
linear-gradient(180deg, hsl(250 60% 16% / 0.68), hsl(235 45% 4%) 48%, hsl(235 54% 3%));
font-family: "Inter", "Avenir Next", "Segoe UI", sans-serif;
overflow: hidden;
overscroll-behavior: none;
}
button,
@@ -93,44 +78,6 @@ textarea {
background-size: 48px 48px;
}
.app-safe-frame {
padding: var(--safe-area-top) var(--safe-area-right) var(--safe-area-bottom) var(--safe-area-left);
}
.app-safe-pad {
padding:
max(1rem, var(--safe-area-top))
max(1rem, var(--safe-area-right))
max(1rem, var(--safe-area-bottom))
max(1rem, var(--safe-area-left));
}
.app-search-safe-pad {
padding:
max(1.5rem, var(--safe-area-top))
max(0.75rem, var(--safe-area-right))
max(1.5rem, var(--safe-area-bottom))
max(0.75rem, var(--safe-area-left));
}
@media (min-width: 768px) {
.app-safe-frame {
padding:
max(0.5rem, var(--safe-area-top))
max(0.5rem, var(--safe-area-right))
max(0.5rem, var(--safe-area-bottom))
max(0.5rem, var(--safe-area-left));
}
.app-search-safe-pad {
padding:
max(1.5rem, var(--safe-area-top))
max(1.5rem, var(--safe-area-right))
max(1.5rem, var(--safe-area-bottom))
max(1.5rem, var(--safe-area-left));
}
}
.glass-panel {
background:
linear-gradient(180deg, hsl(243 42% 12% / 0.88), hsl(236 48% 5% / 0.92)),
@@ -140,148 +87,6 @@ textarea {
0 14px 36px hsl(240 80% 2% / 0.28);
}
.tool-call-stack-shell {
perspective: 900px;
transform-style: preserve-3d;
isolation: isolate;
}
.tool-call-stack-card {
transform: translate3d(var(--tool-stack-x, 0), var(--tool-stack-y, 0), var(--tool-stack-z, 0)) scale(var(--tool-stack-scale, 1));
transform-origin: top left;
opacity: var(--tool-stack-opacity, 1);
transition:
opacity 180ms ease,
transform 300ms cubic-bezier(0.2, 0.8, 0.22, 1);
will-change: transform, opacity;
}
.tool-call-stack-shell-layout-a {
animation: tool-call-stack-height-a 340ms cubic-bezier(0.22, 0.61, 0.36, 1) both;
}
.tool-call-stack-shell-layout-b {
animation: tool-call-stack-height-b 340ms cubic-bezier(0.22, 0.61, 0.36, 1) both;
}
.tool-call-stack-card-layout-a {
animation: tool-call-stack-layout-a 340ms cubic-bezier(0.22, 0.61, 0.36, 1) both;
}
.tool-call-stack-card-layout-b {
animation: tool-call-stack-layout-b 340ms cubic-bezier(0.22, 0.61, 0.36, 1) both;
}
.tool-call-stack-card-surface {
transform-origin: top left;
}
.tool-call-stack-card-glass {
backdrop-filter: none;
}
.tool-call-stack-card-enter {
animation: tool-call-stack-drop-in 320ms cubic-bezier(0.18, 0.95, 0.28, 1) backwards;
animation-delay: var(--tool-stack-delay, 0ms);
}
.tool-call-stack-toggle {
border: 1px solid hsl(188 82% 70% / 0.36);
background:
linear-gradient(180deg, hsl(230 36% 16% / 0.96), hsl(238 48% 7% / 0.96)),
hsl(236 48% 8%);
color: hsl(186 92% 86%);
box-shadow:
inset 0 1px 0 hsl(180 100% 88% / 0.08),
0 8px 22px hsl(235 72% 2% / 0.42);
transition:
border-color 160ms ease,
color 160ms ease,
transform 160ms ease,
filter 160ms ease;
}
.tool-call-stack-toggle:hover {
border-color: hsl(188 92% 74% / 0.62);
color: hsl(184 100% 92%);
filter: brightness(1.08);
}
.tool-call-stack-toggle:focus-visible {
outline: 2px solid hsl(188 92% 72% / 0.9);
outline-offset: 2px;
}
@keyframes tool-call-stack-height-a {
from {
height: var(--tool-stack-from-height);
}
to {
height: var(--tool-stack-to-height);
}
}
@keyframes tool-call-stack-height-b {
from {
height: var(--tool-stack-from-height);
}
to {
height: var(--tool-stack-to-height);
}
}
@keyframes tool-call-stack-layout-a {
from {
opacity: var(--tool-stack-from-opacity, 1);
transform: var(--tool-stack-from-transform);
}
to {
opacity: var(--tool-stack-to-opacity, 1);
transform: var(--tool-stack-to-transform);
}
}
@keyframes tool-call-stack-layout-b {
from {
opacity: var(--tool-stack-from-opacity, 1);
transform: var(--tool-stack-from-transform);
}
to {
opacity: var(--tool-stack-to-opacity, 1);
transform: var(--tool-stack-to-transform);
}
}
@keyframes tool-call-stack-drop-in {
from {
opacity: 0.72;
transform: translate3d(0, -0.65rem, 120px) scale(1.025) rotateX(3deg);
}
to {
opacity: 1;
transform: translate3d(0, 0, 0) scale(1) rotateX(0);
}
}
@media (prefers-reduced-motion: reduce) {
.tool-call-stack-card {
transition: none;
}
.tool-call-stack-shell-layout-a,
.tool-call-stack-shell-layout-b,
.tool-call-stack-card-layout-a,
.tool-call-stack-card-layout-b,
.tool-call-stack-card-enter {
animation: none;
}
}
.md-content {
word-break: break-word;
}

59
web/src/lib/api.ts
View File

@@ -3,8 +3,6 @@ export type ChatSummary = {
title: string | null;
createdAt: string;
updatedAt: string;
starred: boolean;
starredAt: string | null;
initiatedProvider: Provider | null;
initiatedModel: string | null;
lastUsedProvider: Provider | null;
@@ -19,8 +17,6 @@ export type SearchSummary = {
query: string | null;
createdAt: string;
updatedAt: string;
starred: boolean;
starredAt: string | null;
};
export type ChatWorkspaceItem = ChatSummary & {
@@ -45,12 +41,12 @@ export type Message = {
export type ToolCallEvent = {
toolCallId: string;
name: string;
status: "initiated" | "completed" | "failed";
status: "completed" | "failed";
summary: string;
args: Record<string, unknown>;
startedAt: string;
completedAt?: string;
durationMs?: number;
completedAt: string;
durationMs: number;
error?: string;
resultPreview?: string;
};
@@ -60,8 +56,6 @@ export type ChatDetail = {
title: string | null;
createdAt: string;
updatedAt: string;
starred: boolean;
starredAt: string | null;
initiatedProvider: Provider | null;
initiatedModel: string | null;
lastUsedProvider: Provider | null;
@@ -93,8 +87,6 @@ export type SearchDetail = {
query: string | null;
createdAt: string;
updatedAt: string;
starred: boolean;
starredAt: string | null;
requestId: string | null;
latencyMs: number | null;
error: string | null;
@@ -196,18 +188,6 @@ type CreateChatRequest = {
messages?: CompletionRequestMessage[];
};
type CreateSearchRequest = {
title?: string;
query?: string;
reuseByQuery?: boolean;
};
type CreateSearchResponse = {
search: SearchSummary;
reused: boolean;
cacheHit: boolean;
};
const API_BASE_URL = import.meta.env.VITE_API_BASE_URL ?? "/api";
const ENV_ADMIN_TOKEN = (import.meta.env.VITE_ADMIN_TOKEN as string | undefined)?.trim() || null;
let authToken: string | null = ENV_ADMIN_TOKEN;
@@ -299,18 +279,7 @@ export async function updateChatTitle(chatId: string, title: string) {
return data.chat;
}
export async function updateChatStar(chatId: string, starred: boolean) {
const data = await api<{ chat: ChatSummary }>(`/v1/chats/${chatId}/star`, {
method: "PATCH",
body: JSON.stringify({ starred }),
});
return data.chat;
}
export async function updateChatSettings(
chatId: string,
body: { title?: string; additionalSystemPrompt?: string | null; enabledTools?: string[] }
) {
export async function updateChatSettings(chatId: string, body: { additionalSystemPrompt?: string | null; enabledTools?: string[] }) {
const data = await api<{ chat: ChatSummary }>(`/v1/chats/${chatId}`, {
method: "PATCH",
body: JSON.stringify(body),
@@ -335,35 +304,19 @@ export async function listSearches() {
return data.searches;
}
async function postSearch(body?: CreateSearchRequest) {
return api<CreateSearchResponse>("/v1/searches", {
export async function createSearch(body?: { title?: string; query?: string }) {
const data = await api<{ search: SearchSummary }>("/v1/searches", {
method: "POST",
body: JSON.stringify(body ?? {}),
});
}
export async function createSearch(body?: CreateSearchRequest) {
const data = await postSearch(body);
return data.search;
}
export async function createReusableSearch(body: Omit<CreateSearchRequest, "reuseByQuery">) {
return postSearch({ ...body, reuseByQuery: true });
}
export async function getSearch(searchId: string) {
const data = await api<{ search: SearchDetail }>(`/v1/searches/${searchId}`);
return data.search;
}
export async function updateSearchStar(searchId: string, starred: boolean) {
const data = await api<{ search: SearchSummary }>(`/v1/searches/${searchId}/star`, {
method: "PATCH",
body: JSON.stringify({ starred }),
});
return data.search;
}
export async function createChatFromSearch(searchId: string, body?: { title?: string }) {
const data = await api<{ chat: ChatSummary }>(`/v1/searches/${searchId}/chat`, {
method: "POST",

3
web/src/main.tsx
View File

@@ -1,8 +1,5 @@
import { render } from "preact";
import { RootRouter } from "@/root-router";
import { registerServiceWorker } from "@/pwa";
import "./index.css";
registerServiceWorker();
render(<RootRouter />, document.getElementById("app")!);

24
web/src/pages/search-route-page.tsx
View File

@@ -4,7 +4,7 @@ import { AuthScreen } from "@/components/auth/auth-screen";
import { SearchResultsPanel } from "@/components/search/search-results-panel";
import { Button } from "@/components/ui/button";
import { Input } from "@/components/ui/input";
import { createReusableSearch, getSearch, runSearchStream, type SearchDetail } from "@/lib/api";
import { createSearch, runSearchStream, type SearchDetail } from "@/lib/api";
import { useSessionAuth } from "@/hooks/use-session-auth";
function readQueryFromUrl() {
@@ -85,16 +85,14 @@ export default function SearchRoutePage() {
const runQuery = async (query: string) => {
const trimmed = query.trim();
const requestId = ++requestCounterRef.current;
streamAbortRef.current?.abort();
if (!trimmed) {
setSearch(null);
setError(null);
setIsRunning(false);
return;
}
const requestId = ++requestCounterRef.current;
streamAbortRef.current?.abort();
const abortController = new AbortController();
streamAbortRef.current = abortController;
let wasInterrupted = false;
@@ -108,8 +106,6 @@ export default function SearchRoutePage() {
query: trimmed,
createdAt: nowIso,
updatedAt: nowIso,
starred: false,
starredAt: null,
requestId: null,
latencyMs: null,
error: null,
@@ -121,11 +117,10 @@ export default function SearchRoutePage() {
});
try {
const createdResult = await createReusableSearch({
const created = await createSearch({
query: trimmed,
title: trimmed.slice(0, 80),
});
const created = createdResult.search;
if (requestId !== requestCounterRef.current) return;
setSearch((current) =>
@@ -137,19 +132,10 @@ export default function SearchRoutePage() {
query: created.query,
createdAt: created.createdAt,
updatedAt: created.updatedAt,
starred: created.starred,
starredAt: created.starredAt,
}
: current
);
if (createdResult.cacheHit) {
const cached = await getSearch(created.id);
if (requestId !== requestCounterRef.current) return;
setSearch(cached);
return;
}
await runSearchStream(
created.id,
{
@@ -262,7 +248,7 @@ export default function SearchRoutePage() {
}
return (
<div className="app-search-safe-pad h-full overflow-y-auto">
<div className="h-full overflow-y-auto px-3 py-6 md:px-6">
<div className="mx-auto w-full max-w-4xl space-y-5">
<form
className="flex items-center gap-2 rounded-xl border bg-background p-2 shadow-sm"

9
web/src/pwa.ts
View File

@@ -1,9 +0,0 @@
export function registerServiceWorker() {
if (!import.meta.env.PROD || !("serviceWorker" in navigator)) return;
window.addEventListener("load", () => {
void navigator.serviceWorker.register("/sw.js").catch((error: unknown) => {
console.warn("Sybil service worker registration failed", error);
});
});
}

2
web/tsconfig.tsbuildinfo
View File

@@ -1 +1 @@
{"root":["./src/App.tsx","./src/main.tsx","./src/pwa.ts","./src/root-router.tsx","./src/vite-env.d.ts","./src/components/sybil-character.tsx","./src/components/auth/auth-screen.tsx","./src/components/chat/chat-attachment-list.tsx","./src/components/chat/chat-messages-panel.tsx","./src/components/markdown/markdown-content.tsx","./src/components/search/search-results-panel.tsx","./src/components/ui/button.tsx","./src/components/ui/input.tsx","./src/components/ui/scroll-area.tsx","./src/components/ui/separator.tsx","./src/components/ui/textarea.tsx","./src/hooks/use-session-auth.ts","./src/lib/api.ts","./src/lib/utils.ts","./src/pages/search-route-page.tsx"],"version":"5.9.3"}
{"root":["./src/app.tsx","./src/main.tsx","./src/root-router.tsx","./src/vite-env.d.ts","./src/components/sybil-character.tsx","./src/components/auth/auth-screen.tsx","./src/components/chat/chat-attachment-list.tsx","./src/components/chat/chat-messages-panel.tsx","./src/components/markdown/markdown-content.tsx","./src/components/search/search-results-panel.tsx","./src/components/ui/button.tsx","./src/components/ui/input.tsx","./src/components/ui/scroll-area.tsx","./src/components/ui/separator.tsx","./src/components/ui/textarea.tsx","./src/hooks/use-session-auth.ts","./src/lib/api.ts","./src/lib/utils.ts","./src/pages/search-route-page.tsx"],"version":"5.9.3"}