Preserve provider changes before sending

dist/default.conf

@@ -12,17 +12,43 @@ server {
   location /api/ {
     proxy_pass http://server:8787/;
     proxy_http_version 1.1;
+    proxy_buffering off;
+    proxy_cache off;
+    proxy_read_timeout 3600s;
+    proxy_send_timeout 3600s;
     proxy_set_header Host $host;
     proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
     proxy_set_header X-Forwarded-Proto $scheme;
+    proxy_set_header Connection "";
+  }
+
+  location = /sw.js {
+    add_header Cache-Control "no-store, no-cache, must-revalidate" always;
+    expires -1;
+    try_files $uri =404;
   }
 
   location = /manifest.webmanifest {
     default_type application/manifest+json;
+    add_header Cache-Control "no-store, no-cache, must-revalidate" always;
+    expires -1;
+    try_files $uri =404;
+  }
+
+  location = /index.html {
+    add_header Cache-Control "no-store, no-cache, must-revalidate" always;
+    expires -1;
+    try_files $uri =404;
+  }
+
+  location /assets/ {
+    add_header Cache-Control "public, max-age=31536000, immutable" always;
     try_files $uri =404;
   }
 
   location / {
+    add_header Cache-Control "no-store, no-cache, must-revalidate" always;
+    expires -1;
     try_files $uri $uri/ /index.html;
   }
 }

docker-compose.example.yml

@@ -17,6 +17,7 @@ services:
       HERMES_AGENT_API_KEY: ${HERMES_AGENT_API_KEY:-}
       HERMES_AGENT_MODEL: ${HERMES_AGENT_MODEL:-}
       EXA_API_KEY: ${EXA_API_KEY:-}
+      BRAVE_SEARCH_API_KEY: ${BRAVE_SEARCH_API_KEY:-}
       CHAT_WEB_SEARCH_ENGINE: ${CHAT_WEB_SEARCH_ENGINE:-exa}
       SEARXNG_BASE_URL: ${SEARXNG_BASE_URL:-}
       CHAT_MAX_TOOL_ROUNDS: ${CHAT_MAX_TOOL_ROUNDS:-100}

docs/api/rest.md

@@ -186,6 +186,7 @@ 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.
+- If the title provider is unavailable or rejects the request, server still persists a deterministic title derived from the first line of `content` instead of leaving the chat untitled.
 
 ### `DELETE /v1/chats/:chatId`
 - Response: `{ "deleted": true }`
@@ -304,7 +305,8 @@ Behavior notes:
 - 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`, `xai`, and `gemini`: `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`.
+- `web_search` returns ranked results with per-result summaries/snippets. Its backend engine is selected by `CHAT_WEB_SEARCH_ENGINE`: `exa` (default), `brave` (requires `BRAVE_SEARCH_API_KEY`), or `searxng` (requires `SEARXNG_BASE_URL`; the instance must allow `format=json`).
+- Brave searches are queued and evenly paced according to the shortest window in Brave's `X-RateLimit-Policy` response header. The backend also honors `X-RateLimit-Remaining`/`X-RateLimit-Reset` and retries `429` responses up to three times with reset-aware exponential backoff; quota resets beyond the bounded retry window fail immediately.
 - `fetch_url` fetches a URL with browser-like navigation headers 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.
@@ -322,6 +324,16 @@ Behavior notes:
   - `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.
 
+## Streaming Chat
+
+### `POST /v1/chat-completions/stream`
+
+- The request accepts the chat-completion fields above plus optional `persist` and `clientRequestId` fields.
+- `clientRequestId` is only valid for a persisted request with a `chatId`, may be up to 128 characters, and should be a stable unique value generated once per user submission.
+- Retrying with the same `chatId` and `clientRequestId` replays the matching active or completed stream rather than starting a duplicate provider call.
+- The server persists the ID in `metadata.clientRequestId` on the submitted user message and completed assistant message.
+- The complete request, SSE event, persistence, retry, and attach contracts are defined in `docs/api/streaming-chat.md`.
+
 ## Searches
 
 ### `GET /v1/searches`

docs/api/streaming-chat.md

@@ -21,6 +21,7 @@ Authentication:
 {
   "chatId": "optional-chat-id",
   "persist": true,
+  "clientRequestId": "optional-client-generated-id",
   "provider": "openai|anthropic|xai|gemini|hermes-agent",
   "model": "string",
   "messages": [
@@ -61,6 +62,9 @@ Notes:
 - If `persist` is `true` and `chatId` is omitted, backend creates a new chat.
 - 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.
+- `clientRequestId` is optional and is only valid for a persisted stream with a `chatId`. Clients should generate one stable, unique value per user submission and reuse it when retrying a disconnected request.
+- A retry with the same `chatId` and `clientRequestId` attaches to and replays the matching active stream. If that submission already completed, the endpoint replays `meta` and `done` without invoking the provider again. This makes retrying the initial streaming `POST` idempotent.
+- `clientRequestId` values may be up to 128 characters. The server stores the value in `metadata.clientRequestId` on the submitted user message and completed assistant message.
 - 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.
@@ -70,7 +74,7 @@ Notes:
 Persisted chat streams with a `chatId` are backend-owned active runs:
 - Once started, the backend keeps the stream running even if the HTTP client disconnects or refreshes.
 - While running, `GET /v1/active-runs` includes the `chatId`.
-- Starting a second persisted stream for the same active `chatId` returns `409`.
+- Starting a second persisted stream for the same active `chatId` returns `409`, unless its `clientRequestId` matches the active submission, in which case the existing stream is replayed.
 - Clients can reattach with `POST /v1/chats/:chatId/stream/attach`.
 
 ## Attach Endpoint
@@ -182,7 +186,8 @@ Terminal tool-call event:
 - `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.
-- `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.
+- `web_search` uses `CHAT_WEB_SEARCH_ENGINE`: `exa` (default), `brave` (requires `BRAVE_SEARCH_API_KEY`), or `searxng` (requires `SEARXNG_BASE_URL`; the instance must allow `format=json`). This only affects chat-mode tool calls, not search-mode endpoints.
+- Brave searches are queued and evenly paced according to the shortest window in Brave's `X-RateLimit-Policy` response header. The backend also honors `X-RateLimit-Remaining`/`X-RateLimit-Reset` and retries `429` responses up to three times with reset-aware exponential backoff; quota resets beyond the bounded retry window fail immediately.
 - `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.

ios/Packages/Sybil/Sources/Sybil/SybilChatTranscriptView.swift

@@ -67,7 +67,6 @@ struct SybilChatTranscriptView: View {
             .scrollDismissesKeyboard(.interactively)
             .onAppear {
                 syncKnownToolCallMessageIDs()
-                scrollToBottom(with: proxy, animated: false)
             }
             .onChange(of: toolCallMessageIDSignature) { _, _ in
                 syncKnownToolCallMessageIDs()

ios/Packages/Sybil/Sources/Sybil/SybilViewModel.swift

@@ -1752,13 +1752,16 @@ final class SybilViewModel {
                 switch target {
                 case let .chat(chatID):
                     SybilLog.debug(SybilLog.app, "Refreshing chat \(chatID)")
+                    let isSelectingDifferentChat = selectedChat?.id != chatID
                     let chat = try await client.getChat(chatID: chatID)
                     guard selectedItem == target, draftKind == nil else {
                         return
                     }
                     selectedChat = chat
                     selectedSearch = nil
-                    requestChatBottomPin()
+                    if isSelectingDifferentChat {
+                        requestChatBottomPin()
+                    }
 
                     if let provider = chat.lastUsedProvider,
                        let model = chat.lastUsedModel,

ios/Packages/Sybil/Tests/SybilTests/SybilTests.swift

@@ -544,12 +544,14 @@ private func makeToolCallMessage(id: String, date: Date, summary: String = "Ran
 @MainActor
 @Test func foregroundChatRefreshReloadsSelectedTranscript() async throws {
     let date = Date(timeIntervalSince1970: 1_700_000_100)
+    let staleDetail = makeChatDetail(id: "chat-2", date: date, body: "stale transcript")
     let detail = makeChatDetail(id: "chat-2", date: date, body: "refreshed transcript")
     let client = MockSybilClient(chatDetails: ["chat-2": detail])
     let viewModel = SybilViewModel(settings: testSettings(named: #function)) { _ in client }
     viewModel.isAuthenticated = true
     viewModel.isCheckingSession = false
     viewModel.selectedItem = .chat("chat-2")
+    viewModel.selectedChat = staleDetail
 
     await viewModel.refreshVisibleContent(refreshCollections: false, refreshSelection: true)
 
@@ -559,7 +561,7 @@ 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)
+    #expect(viewModel.chatBottomPinRequestID == 0)
 }
 
 @MainActor
@@ -675,6 +677,7 @@ private func makeToolCallMessage(id: String, date: Date, summary: String = "Ran
 
     #expect(viewModel.displayedMessages.first?.content == "fresh transcript")
     #expect(!viewModel.isLoadingSelection)
+    #expect(viewModel.chatBottomPinRequestID == 1)
 }
 
 @MainActor

server/README.md

@@ -48,7 +48,8 @@ If `ADMIN_TOKEN` is not set, the server runs in open mode (dev).
 - `HERMES_AGENT_API_KEY` (enables the Hermes Agent provider; set to Hermes `API_SERVER_KEY`, or any non-empty value if that local server does not require auth)
 - `HERMES_AGENT_MODEL` (optional fallback/override model id; defaults client-side to `hermes-agent`)
 - `EXA_API_KEY`
-- `CHAT_WEB_SEARCH_ENGINE` (`exa` by default, or `searxng` for chat tool calls only)
+- `BRAVE_SEARCH_API_KEY` (required when `CHAT_WEB_SEARCH_ENGINE=brave`)
+- `CHAT_WEB_SEARCH_ENGINE` (`exa` by default; `brave` and `searxng` are also supported for chat tool calls only)
 - `SEARXNG_BASE_URL` (required when `CHAT_WEB_SEARCH_ENGINE=searxng`; instance must allow `format=json`)
 - `CHAT_MAX_TOOL_ROUNDS` (`100` by default; maximum model/tool result cycles per chat completion)
 - `CHAT_CODEX_TOOL_ENABLED` (`false` by default; enables the `codex_exec` chat tool for managed-tool providers)

server/src/env.ts

@@ -24,7 +24,7 @@ const ChatWebSearchEngineSchema = z.preprocess(
     const trimmed = value.trim();
     return trimmed ? trimmed.toLowerCase() : undefined;
   },
-  z.enum(["exa", "searxng"]).default("exa")
+  z.enum(["exa", "searxng", "brave"]).default("exa")
 );
 
 const BooleanFlagSchema = z.preprocess((value) => {
@@ -71,6 +71,7 @@ const EnvSchema = z.object({
   HERMES_AGENT_API_KEY: OptionalTrimmedStringSchema,
   HERMES_AGENT_MODEL: OptionalTrimmedStringSchema,
   EXA_API_KEY: z.string().optional(),
+  BRAVE_SEARCH_API_KEY: OptionalTrimmedStringSchema,
 
   // Chat-mode web_search tool configuration. Search mode remains Exa-only for now.
   CHAT_WEB_SEARCH_ENGINE: ChatWebSearchEngineSchema,
@@ -100,6 +101,14 @@ const EnvSchema = z.object({
     });
   }
 
+  if (value.CHAT_WEB_SEARCH_ENGINE === "brave" && !value.BRAVE_SEARCH_API_KEY) {
+    ctx.addIssue({
+      code: "custom",
+      path: ["BRAVE_SEARCH_API_KEY"],
+      message: "BRAVE_SEARCH_API_KEY is required when CHAT_WEB_SEARCH_ENGINE=brave",
+    });
+  }
+
   if ((value.CHAT_CODEX_TOOL_ENABLED || value.CHAT_SHELL_TOOL_ENABLED) && !value.CHAT_CODEX_REMOTE_HOST) {
     ctx.addIssue({
       code: "custom",

server/src/llm/chat-tools.ts

@@ -7,6 +7,7 @@ import { convert as htmlToText } from "html-to-text";
 import { z } from "zod";
 import { buildBrowserLikeNavigationHeaders } from "../browser-fetch-headers.js";
 import { env } from "../env.js";
+import { searchBrave } from "../search/brave.js";
 import { exaClient } from "../search/exa.js";
 import { searchSearxng } from "../search/searxng.js";
 import type { ChatMessage } from "./types.js";
@@ -507,11 +508,33 @@ async function runSearxngWebSearchTool(args: WebSearchArgs): Promise<ToolRunOutc
   };
 }
 
+async function runBraveWebSearchTool(args: WebSearchArgs): Promise<ToolRunOutcome> {
+  const response = await searchBrave(args.query, {
+    numResults: args.numResults ?? DEFAULT_WEB_RESULTS,
+    includeDomains: args.includeDomains,
+    excludeDomains: args.excludeDomains,
+  });
+
+  return {
+    ok: true,
+    searchEngine: "brave",
+    query: args.query,
+    requestId: response.requestId,
+    results: response.results.map((result, index) => ({
+      rank: index + 1,
+      ...result,
+    })),
+  };
+}
+
 async function runWebSearchTool(input: unknown): Promise<ToolRunOutcome> {
   const args = WebSearchArgsSchema.parse(input);
   if (env.CHAT_WEB_SEARCH_ENGINE === "searxng") {
     return runSearxngWebSearchTool(args);
   }
+  if (env.CHAT_WEB_SEARCH_ENGINE === "brave") {
+    return runBraveWebSearchTool(args);
+  }
   return runExaWebSearchTool(args);
 }
 

server/src/llm/streaming.ts

@@ -119,7 +119,12 @@ export async function* runMultiplexStream(req: MultiplexRequest): AsyncGenerator
     if (shouldPersist && chatId && call) {
       await prisma.$transaction(async (tx) => {
         await tx.message.create({
-          data: { chatId, role: "assistant" as any, content: text },
+          data: {
+            chatId,
+            role: "assistant" as any,
+            content: text,
+            metadata: req.clientRequestId ? ({ clientRequestId: req.clientRequestId } as any) : undefined,
+          },
         });
         await tx.llmCall.update({
           where: { id: call.id },

server/src/llm/types.ts

@@ -33,6 +33,7 @@ export type ChatMessage = {
 export type MultiplexRequest = {
   chatId?: string;
   persist?: boolean;
+  clientRequestId?: string;
   provider: Provider;
   model: string;
   messages: ChatMessage[];

server/src/routes.ts

@@ -88,7 +88,7 @@ function withRequestUserLocation<T extends { userLocation?: string }>(body: T, r
   return body.userLocation ? body : { ...body, userLocation: inferRequestUserLocation(req) };
 }
 
-async function storeNonAssistantMessages(chatId: string, messages: IncomingChatMessage[]) {
+async function storeNonAssistantMessages(chatId: string, messages: IncomingChatMessage[], clientRequestId?: string) {
   const incoming = messages.filter((m) => m.role !== "assistant");
   if (!incoming.length) return;
 
@@ -109,14 +109,21 @@ async function storeNonAssistantMessages(chatId: string, messages: IncomingChatM
   const toInsert = sharedPrefix === existingNonAssistant.length ? incoming.slice(existingNonAssistant.length) : incoming;
   if (!toInsert.length) return;
 
+  const finalUserMessageIndex = toInsert.map((message) => message.role).lastIndexOf("user");
   await prisma.message.createMany({
-    data: toInsert.map((m) => ({
+    data: toInsert.map((m, index) => {
-      chatId,
+      const metadata = {
-      role: m.role as any,
+        ...(m.attachments?.length ? { attachments: m.attachments } : {}),
-      content: m.content,
+        ...(clientRequestId && index === finalUserMessageIndex ? { clientRequestId } : {}),
-      name: m.name,
+      };
-      metadata: m.attachments?.length ? ({ attachments: m.attachments } as any) : undefined,
+      return {
-    })),
+        chatId,
+        role: m.role as any,
+        content: m.content,
+        name: m.name,
+        metadata: Object.keys(metadata).length ? (metadata as any) : undefined,
+      };
+    }),
   });
 }
 
@@ -169,6 +176,7 @@ const CompletionStreamBody = z
   .object({
     chatId: z.string().optional(),
     persist: z.boolean().optional(),
+    clientRequestId: z.string().trim().min(1).max(128).optional(),
     provider: ProviderSchema,
     model: z.string().min(1),
     messages: z.array(CompletionMessageSchema),
@@ -186,6 +194,13 @@ const CompletionStreamBody = z
         path: ["chatId"],
       });
     }
+    if (value.clientRequestId && (value.persist === false || !value.chatId)) {
+      ctx.addIssue({
+        code: z.ZodIssueCode.custom,
+        message: "clientRequestId requires a persisted stream with chatId",
+        path: ["clientRequestId"],
+      });
+    }
   });
 
 function mergeAttachmentsIntoMetadata(metadata: unknown, attachments?: ChatAttachment[]) {
@@ -399,6 +414,7 @@ function buildSseHeaders(originHeader: string | undefined) {
 type SearchRunRequest = z.infer<typeof SearchRunBody>;
 
 const activeChatStreams = new Map<string, ActiveSseStream>();
+const activeChatStreamRequestIds = new Map<string, string>();
 const activeSearchStreams = new Map<string, ActiveSseStream>();
 const STARRED_PROJECT_ID = "starred";
 
@@ -554,6 +570,7 @@ function writeSseEvent(reply: FastifyReply, event: SseStreamEvent) {
 }
 
 async function streamActiveRun(req: FastifyRequest, reply: FastifyReply, stream: ActiveSseStream) {
+  if (reply.raw.destroyed || reply.raw.writableEnded) return reply;
   reply.raw.writeHead(200, buildSseHeaders(typeof req.headers.origin === "string" ? req.headers.origin : undefined));
   reply.raw.flushHeaders?.();
 
@@ -588,10 +605,24 @@ function mapChatStreamEvent(ev: StreamEvent): SseStreamEvent {
   return { event: ev.type, data: ev };
 }
 
-function startActiveChatStream(chatId: string, body: z.infer<typeof CompletionStreamBody>) {
+function registerActiveChatStream(chatId: string, clientRequestId?: string) {
   const stream = new ActiveSseStream();
   activeChatStreams.set(chatId, stream);
+  if (clientRequestId) {
+    activeChatStreamRequestIds.set(chatId, clientRequestId);
+  } else {
+    activeChatStreamRequestIds.delete(chatId);
+  }
+  return stream;
+}
 
+function clearActiveChatStream(chatId: string, stream: ActiveSseStream) {
+  if (activeChatStreams.get(chatId) !== stream) return;
+  activeChatStreams.delete(chatId);
+  activeChatStreamRequestIds.delete(chatId);
+}
+
+function executeActiveChatStream(chatId: string, body: z.infer<typeof CompletionStreamBody>, stream: ActiveSseStream) {
   void (async () => {
     let sawTerminalEvent = false;
     try {
@@ -611,13 +642,54 @@ function startActiveChatStream(chatId: string, body: z.infer<typeof CompletionSt
     } catch (err) {
       stream.complete({ event: "error", data: { message: getErrorMessage(err) } });
     } finally {
-      activeChatStreams.delete(chatId);
+      clearActiveChatStream(chatId, stream);
     }
   })();
+}
 
+function startActiveChatStream(chatId: string, body: z.infer<typeof CompletionStreamBody>) {
+  const stream = registerActiveChatStream(chatId, body.clientRequestId);
+  executeActiveChatStream(chatId, body, stream);
   return stream;
 }
 
+function getMetadataClientRequestId(metadata: unknown) {
+  if (!metadata || typeof metadata !== "object" || Array.isArray(metadata)) return null;
+  const clientRequestId = (metadata as Record<string, unknown>).clientRequestId;
+  return typeof clientRequestId === "string" ? clientRequestId : null;
+}
+
+async function findCompletedChatSubmission(chatId: string, clientRequestId: string) {
+  const assistantMessages = await prisma.message.findMany({
+    where: { chatId, role: "assistant" as any },
+    orderBy: { createdAt: "desc" },
+    select: { content: true, metadata: true },
+  });
+  return assistantMessages.find((message) => getMetadataClientRequestId(message.metadata) === clientRequestId) ?? null;
+}
+
+function completeChatSubmissionStream(
+  stream: ActiveSseStream,
+  chatId: string,
+  body: z.infer<typeof CompletionStreamBody>,
+  assistantText: string
+) {
+  stream.emit("meta", {
+    type: "meta",
+    chatId,
+    callId: null,
+    provider: body.provider,
+    model: body.model,
+  });
+  stream.complete({
+    event: "done",
+    data: {
+      type: "done",
+      text: assistantText,
+    },
+  });
+}
+
 async function executeSearchRunStream(searchId: string, body: SearchRunRequest, stream: ActiveSseStream) {
   const startedAt = performance.now();
   const query = body.query?.trim();
@@ -935,7 +1007,18 @@ export async function registerRoutes(app: FastifyInstance) {
     if (existing.title?.trim()) return { chat: serializeChatLike(existing) };
 
     const fallback = body.content.split(/\r?\n/)[0]?.trim().slice(0, 48) || "New chat";
-    const suggestedRaw = await generateChatTitle(body.content);
+    let suggestedRaw = "";
+    try {
+      suggestedRaw = await generateChatTitle(body.content);
+    } catch (err) {
+      req.log.warn(
+        {
+          chatId: body.chatId,
+          err: getErrorMessage(err),
+        },
+        "chat title generation failed; using fallback"
+      );
+    }
     const title = normalizeSuggestedTitle(suggestedRaw, fallback);
 
     await prisma.chat.updateMany({
@@ -1353,15 +1436,39 @@ export async function registerRoutes(app: FastifyInstance) {
       if (!exists) return app.httpErrors.notFound("chat not found");
     }
 
-    // Store only new non-assistant messages to avoid duplicate history entries.
     if (body.persist !== false && body.chatId) {
-      await storeNonAssistantMessages(body.chatId, body.messages);
+      const activeStream = activeChatStreams.get(body.chatId);
-    }
+      if (activeStream) {
-
+        if (body.clientRequestId && activeChatStreamRequestIds.get(body.chatId) === body.clientRequestId) {
-    if (body.persist !== false && body.chatId) {
+          return streamActiveRun(req, reply, activeStream);
-      if (activeChatStreams.has(body.chatId)) {
+        }
         return app.httpErrors.conflict("chat completion already running");
       }
+
+      if (body.clientRequestId) {
+        const reservedStream = registerActiveChatStream(body.chatId, body.clientRequestId);
+        try {
+          const completedSubmission = await findCompletedChatSubmission(body.chatId, body.clientRequestId);
+          if (completedSubmission) {
+            completeChatSubmissionStream(reservedStream, body.chatId, body, completedSubmission.content);
+            clearActiveChatStream(body.chatId, reservedStream);
+            return streamActiveRun(req, reply, reservedStream);
+          }
+
+          // Store only new non-assistant messages to avoid duplicate history entries.
+          await storeNonAssistantMessages(body.chatId, body.messages, body.clientRequestId);
+          const configuredBody = await applyStoredChatSettings(body);
+          executeActiveChatStream(body.chatId, configuredBody, reservedStream);
+          return streamActiveRun(req, reply, reservedStream);
+        } catch (err) {
+          reservedStream.complete({ event: "error", data: { message: getErrorMessage(err) } });
+          clearActiveChatStream(body.chatId, reservedStream);
+          throw err;
+        }
+      }
+
+      // Legacy requests without an idempotency key retain the original behavior.
+      await storeNonAssistantMessages(body.chatId, body.messages);
       const stream = startActiveChatStream(body.chatId, await applyStoredChatSettings(body));
       return streamActiveRun(req, reply, stream);
     }

server/src/search/brave.ts

@@ -0,0 +1,305 @@
+import { buildBrowserLikeRequestHeaders } from "../browser-fetch-headers.js";
+import { env } from "../env.js";
+
+const BRAVE_WEB_SEARCH_URL = "https://api.search.brave.com/res/v1/web/search";
+const BRAVE_SEARCH_TIMEOUT_MS = 12_000;
+const DEFAULT_BRAVE_REQUEST_INTERVAL_MS = 1_000;
+const RATE_LIMIT_INTERVAL_SAFETY_RATIO = 0.05;
+const MIN_RATE_LIMIT_INTERVAL_SAFETY_MS = 2;
+const RATE_LIMIT_RESET_SAFETY_MS = 50;
+const MAX_RATE_LIMIT_RETRIES = 3;
+const MAX_RATE_LIMIT_RETRY_DELAY_MS = 8_000;
+
+type RateLimitPolicy = {
+  limit: number;
+  windowSeconds: number;
+};
+
+let requestIntervalMs = addIntervalSafety(DEFAULT_BRAVE_REQUEST_INTERVAL_MS);
+let lastRequestAtMs = 0;
+let nextRequestAtMs = 0;
+let quotaUnavailableUntilMs = 0;
+let requestQueue = Promise.resolve();
+
+export type BraveSearchOptions = {
+  numResults: number;
+  includeDomains?: string[];
+  excludeDomains?: string[];
+};
+
+export type BraveSearchResult = {
+  title: string | null;
+  url: string | null;
+  publishedDate: string | null;
+  author: string | null;
+  summary: string | null;
+  text: string | null;
+  highlights: string[];
+};
+
+export type BraveSearchResponse = {
+  query: string;
+  requestId: string | null;
+  results: BraveSearchResult[];
+};
+
+function clipText(input: string, maxCharacters: number) {
+  return input.length <= maxCharacters ? input : `${input.slice(0, maxCharacters)}...`;
+}
+
+function compactWhitespace(input: string) {
+  return input.replace(/\r/g, "").replace(/[ \t]+\n/g, "\n").replace(/\n{3,}/g, "\n\n").replace(/\s+/g, " ").trim();
+}
+
+function requireBraveSearchApiKey() {
+  if (!env.BRAVE_SEARCH_API_KEY) {
+    throw new Error("BRAVE_SEARCH_API_KEY not set");
+  }
+  return env.BRAVE_SEARCH_API_KEY;
+}
+
+function sleep(milliseconds: number) {
+  return new Promise<void>((resolve) => setTimeout(resolve, milliseconds));
+}
+
+function addIntervalSafety(intervalMs: number) {
+  return intervalMs + Math.max(MIN_RATE_LIMIT_INTERVAL_SAFETY_MS, Math.ceil(intervalMs * RATE_LIMIT_INTERVAL_SAFETY_RATIO));
+}
+
+function parseCommaSeparatedNumbers(value: string | null) {
+  if (!value) return [];
+  return value.split(",").map((part) => Number(part.trim())).map((number) => (Number.isFinite(number) ? number : null));
+}
+
+function parseRateLimitPolicy(value: string | null): RateLimitPolicy[] {
+  if (!value) return [];
+  return value.split(",").flatMap((part) => {
+    const match = part.trim().match(/^(\d+)\s*;\s*w=(\d+)$/i);
+    if (!match) return [];
+    const limit = Number(match[1]);
+    const windowSeconds = Number(match[2]);
+    return limit > 0 && windowSeconds > 0 ? [{ limit, windowSeconds }] : [];
+  });
+}
+
+function getBurstPolicyIndex(policies: RateLimitPolicy[]) {
+  if (!policies.length) return null;
+  let burstIndex = 0;
+  for (let index = 1; index < policies.length; index += 1) {
+    if (policies[index]!.windowSeconds < policies[burstIndex]!.windowSeconds) burstIndex = index;
+  }
+  return burstIndex;
+}
+
+function updateRateLimitState(headers: Headers) {
+  const policies = parseRateLimitPolicy(headers.get("x-ratelimit-policy"));
+  const burstIndex = getBurstPolicyIndex(policies);
+  if (burstIndex === null) return;
+
+  const burstPolicy = policies[burstIndex]!;
+  const learnedIntervalMs = addIntervalSafety(Math.ceil((burstPolicy.windowSeconds * 1_000) / burstPolicy.limit));
+  if (learnedIntervalMs < requestIntervalMs && lastRequestAtMs > 0) {
+    nextRequestAtMs = Math.min(nextRequestAtMs, lastRequestAtMs + learnedIntervalMs);
+  }
+  requestIntervalMs = learnedIntervalMs;
+
+  const remaining = parseCommaSeparatedNumbers(headers.get("x-ratelimit-remaining"));
+  const resetSeconds = parseCommaSeparatedNumbers(headers.get("x-ratelimit-reset"));
+  for (let index = 0; index < policies.length; index += 1) {
+    if ((remaining[index] ?? null) === null || remaining[index]! >= 1 || (resetSeconds[index] ?? 0) <= 0) continue;
+    const unavailableUntilMs = Date.now() + resetSeconds[index]! * 1_000 + RATE_LIMIT_RESET_SAFETY_MS;
+    if (index === burstIndex) {
+      nextRequestAtMs = Math.max(nextRequestAtMs, unavailableUntilMs);
+    } else {
+      quotaUnavailableUntilMs = Math.max(quotaUnavailableUntilMs, unavailableUntilMs);
+    }
+  }
+}
+
+function assertLongTermQuotaAvailable() {
+  if (quotaUnavailableUntilMs <= Date.now()) {
+    quotaUnavailableUntilMs = 0;
+    return;
+  }
+  const resetSeconds = Math.ceil((quotaUnavailableUntilMs - Date.now()) / 1_000);
+  throw new Error(`Brave Search API long-term quota is exhausted; reset is expected in ${resetSeconds} seconds.`);
+}
+
+async function waitForRateLimitSlot() {
+  const reservation = requestQueue.then(async () => {
+    while (true) {
+      assertLongTermQuotaAvailable();
+      const waitMs = nextRequestAtMs - Date.now();
+      if (waitMs <= 0) break;
+      await sleep(waitMs);
+    }
+    lastRequestAtMs = Date.now();
+    nextRequestAtMs = lastRequestAtMs + requestIntervalMs;
+  });
+  requestQueue = reservation.catch(() => undefined);
+  await reservation;
+}
+
+function get429RetryDelayMs(headers: Headers, retryNumber: number) {
+  const remaining = parseCommaSeparatedNumbers(headers.get("x-ratelimit-remaining"));
+  const resetSeconds = parseCommaSeparatedNumbers(headers.get("x-ratelimit-reset"));
+  const exhaustedResetSeconds = resetSeconds.filter((reset, index): reset is number => reset !== null && (remaining[index] ?? 0) < 1);
+  const headerDelayMs = exhaustedResetSeconds.length ? Math.max(...exhaustedResetSeconds) * 1_000 : 0;
+  const exponentialDelayMs = 2 ** retryNumber * 1_000;
+  const delayMs = Math.max(headerDelayMs + RATE_LIMIT_RESET_SAFETY_MS, exponentialDelayMs);
+  return delayMs <= MAX_RATE_LIMIT_RETRY_DELAY_MS ? delayMs : null;
+}
+
+async function fetchBrave(url: URL) {
+  const apiKey = requireBraveSearchApiKey();
+  for (let attempt = 0; attempt <= MAX_RATE_LIMIT_RETRIES; attempt += 1) {
+    await waitForRateLimitSlot();
+
+    const controller = new AbortController();
+    const timeout = setTimeout(() => controller.abort(), BRAVE_SEARCH_TIMEOUT_MS);
+    let response: Response;
+    try {
+      response = await fetch(url, {
+        signal: controller.signal,
+        headers: {
+          ...buildBrowserLikeRequestHeaders("application/json"),
+          "X-Subscription-Token": apiKey,
+        },
+      });
+    } finally {
+      clearTimeout(timeout);
+    }
+
+    updateRateLimitState(response.headers);
+    if (response.status !== 429 || attempt === MAX_RATE_LIMIT_RETRIES) return response;
+
+    const retryDelayMs = get429RetryDelayMs(response.headers, attempt);
+    await response.arrayBuffer();
+    if (retryDelayMs === null) {
+      throw new Error("Brave Search API rate limit quota is exhausted beyond the retry window.");
+    }
+    await sleep(retryDelayMs);
+  }
+
+  throw new Error("Brave Search API request failed after rate-limit retries.");
+}
+
+function normalizeDomain(input: string) {
+  const trimmed = input.trim().toLowerCase();
+  if (!trimmed) return null;
+
+  try {
+    const parsed = new URL(trimmed.includes("://") ? trimmed : `https://${trimmed}`);
+    return parsed.hostname.replace(/^www\./, "");
+  } catch {
+    return trimmed.split(/[/?#]/, 1)[0]?.replace(/^www\./, "") || null;
+  }
+}
+
+function normalizeDomains(input: string[] | undefined) {
+  return Array.from(new Set((input ?? []).map(normalizeDomain).filter((domain): domain is string => Boolean(domain))));
+}
+
+function hostnameMatchesDomain(urlRaw: string | null, domain: string) {
+  if (!urlRaw) return false;
+  try {
+    const hostname = new URL(urlRaw).hostname.toLowerCase().replace(/^www\./, "");
+    return hostname === domain || hostname.endsWith(`.${domain}`);
+  } catch {
+    return false;
+  }
+}
+
+function filterResultsByDomains(results: BraveSearchResult[], options: BraveSearchOptions) {
+  const includeDomains = normalizeDomains(options.includeDomains);
+  const excludeDomains = normalizeDomains(options.excludeDomains);
+  return results.filter((result) => {
+    if (includeDomains.length && !includeDomains.some((domain) => hostnameMatchesDomain(result.url, domain))) return false;
+    if (excludeDomains.some((domain) => hostnameMatchesDomain(result.url, domain))) return false;
+    return true;
+  });
+}
+
+function buildBraveQuery(query: string, options: BraveSearchOptions) {
+  const includeDomains = normalizeDomains(options.includeDomains);
+  const excludeDomains = normalizeDomains(options.excludeDomains);
+  const includeClause =
+    includeDomains.length === 0
+      ? ""
+      : includeDomains.length === 1
+        ? `site:${includeDomains[0]}`
+        : `(${includeDomains.map((domain) => `site:${domain}`).join(" OR ")})`;
+  const excludeClause = excludeDomains.map((domain) => `-site:${domain}`).join(" ");
+  return [query, includeClause, excludeClause].filter(Boolean).join(" ");
+}
+
+function buildSearchUrl(query: string, options: BraveSearchOptions) {
+  const url = new URL(BRAVE_WEB_SEARCH_URL);
+  url.searchParams.set("q", buildBraveQuery(query, options));
+  url.searchParams.set("count", String(options.numResults));
+  url.searchParams.set("safesearch", "moderate");
+  url.searchParams.set("result_filter", "web");
+  url.searchParams.set("text_decorations", "false");
+  url.searchParams.set("extra_snippets", "true");
+  return url;
+}
+
+function stringOrNull(value: unknown) {
+  if (typeof value !== "string") return null;
+  const normalized = compactWhitespace(value);
+  return normalized || null;
+}
+
+function stringArray(value: unknown) {
+  if (!Array.isArray(value)) return [];
+  return value.filter((item): item is string => typeof item === "string").map(compactWhitespace).filter(Boolean);
+}
+
+function mapWebResult(result: any): BraveSearchResult {
+  const description = stringOrNull(result?.description);
+  const extraSnippets = stringArray(result?.extra_snippets);
+  const snippets = [description, ...extraSnippets].filter((snippet): snippet is string => Boolean(snippet));
+  const combinedText = snippets.join("\n\n");
+
+  return {
+    title: stringOrNull(result?.title),
+    url: stringOrNull(result?.url),
+    publishedDate: stringOrNull(result?.page_age),
+    author: stringOrNull(result?.profile?.name) ?? stringOrNull(result?.article?.author),
+    summary: description ? clipText(description, 1_400) : null,
+    text: combinedText ? clipText(combinedText, 700) : null,
+    highlights: snippets.slice(0, 3).map((snippet) => clipText(snippet, 280)),
+  };
+}
+
+export async function searchBrave(query: string, options: BraveSearchOptions): Promise<BraveSearchResponse> {
+  const url = buildSearchUrl(query, options);
+  const response = await fetchBrave(url);
+
+  if (!response.ok) {
+    await response.arrayBuffer();
+    throw new Error(`Brave Search API request failed with status ${response.status}.`);
+  }
+
+  const contentType = response.headers.get("content-type")?.toLowerCase() ?? "";
+  if (!contentType.includes("application/json")) {
+    await response.arrayBuffer();
+    throw new Error(`Brave Search API returned ${contentType || "unknown content type"}.`);
+  }
+
+  const data: any = await response.json();
+  const results = Array.isArray(data?.web?.results) ? data.web.results.map(mapWebResult) : [];
+  return {
+    query,
+    requestId: response.headers.get("x-request-id"),
+    results: filterResultsByDomains(results, options).slice(0, options.numResults),
+  };
+}
+
+export function resetBraveRateLimitStateForTests() {
+  requestIntervalMs = addIntervalSafety(DEFAULT_BRAVE_REQUEST_INTERVAL_MS);
+  lastRequestAtMs = 0;
+  nextRequestAtMs = 0;
+  quotaUnavailableUntilMs = 0;
+  requestQueue = Promise.resolve();
+}

server/tests/brave-search.test.ts

@@ -0,0 +1,284 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+import { env } from "../src/env.js";
+import { resetBraveRateLimitStateForTests, searchBrave } from "../src/search/brave.js";
+
+test("searchBrave authenticates, builds filters, and normalizes web results", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalApiKey = env.BRAVE_SEARCH_API_KEY;
+  const fetchCalls: Array<{ input: RequestInfo | URL; init?: RequestInit }> = [];
+  resetBraveRateLimitStateForTests();
+  env.BRAVE_SEARCH_API_KEY = "test-brave-key";
+  globalThis.fetch = (async (input: RequestInfo | URL, init?: RequestInit) => {
+    fetchCalls.push({ input, init });
+    return new Response(
+      JSON.stringify({
+        web: {
+          results: [
+            {
+              title: " Brave result ",
+              url: "https://docs.example.com/article",
+              description: "Main\n snippet",
+              extra_snippets: ["Extra snippet one", "Extra snippet two"],
+              page_age: "2026-07-18T12:00:00Z",
+              profile: { name: "Example Docs" },
+            },
+            {
+              title: "Excluded result",
+              url: "https://blocked.example.com/article",
+              description: "Should be filtered",
+            },
+          ],
+        },
+      }),
+      {
+        status: 200,
+        headers: {
+          "content-type": "application/json; charset=utf-8",
+          "x-request-id": "brave-request-1",
+        },
+      }
+    );
+  }) as typeof fetch;
+
+  try {
+    const response = await searchBrave("latest docs", {
+      numResults: 5,
+      includeDomains: ["https://example.com/path"],
+      excludeDomains: ["blocked.example.com"],
+    });
+
+    assert.equal(fetchCalls.length, 1);
+    const requestUrl = new URL(String(fetchCalls[0]?.input));
+    assert.equal(requestUrl.origin + requestUrl.pathname, "https://api.search.brave.com/res/v1/web/search");
+    assert.equal(requestUrl.searchParams.get("q"), "latest docs site:example.com -site:blocked.example.com");
+    assert.equal(requestUrl.searchParams.get("count"), "5");
+    assert.equal(requestUrl.searchParams.get("safesearch"), "moderate");
+    assert.equal(requestUrl.searchParams.get("result_filter"), "web");
+    assert.equal(requestUrl.searchParams.get("text_decorations"), "false");
+    assert.equal(requestUrl.searchParams.get("extra_snippets"), "true");
+    assert.equal((fetchCalls[0]?.init?.headers as Record<string, string>)["X-Subscription-Token"], "test-brave-key");
+
+    assert.deepEqual(response, {
+      query: "latest docs",
+      requestId: "brave-request-1",
+      results: [
+        {
+          title: "Brave result",
+          url: "https://docs.example.com/article",
+          publishedDate: "2026-07-18T12:00:00Z",
+          author: "Example Docs",
+          summary: "Main snippet",
+          text: "Main snippet\n\nExtra snippet one\n\nExtra snippet two",
+          highlights: ["Main snippet", "Extra snippet one", "Extra snippet two"],
+        },
+      ],
+    });
+  } finally {
+    globalThis.fetch = originalFetch;
+    env.BRAVE_SEARCH_API_KEY = originalApiKey;
+  }
+});
+
+test("searchBrave rejects requests without an API key", async () => {
+  const originalApiKey = env.BRAVE_SEARCH_API_KEY;
+  resetBraveRateLimitStateForTests();
+  env.BRAVE_SEARCH_API_KEY = undefined;
+  try {
+    await assert.rejects(() => searchBrave("test", { numResults: 1 }), /BRAVE_SEARCH_API_KEY not set/);
+  } finally {
+    env.BRAVE_SEARCH_API_KEY = originalApiKey;
+  }
+});
+
+test("searchBrave reports non-JSON responses", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalApiKey = env.BRAVE_SEARCH_API_KEY;
+  resetBraveRateLimitStateForTests();
+  env.BRAVE_SEARCH_API_KEY = "test-brave-key";
+  globalThis.fetch = (async () =>
+    new Response("upstream error", {
+      status: 200,
+      headers: { "content-type": "text/plain" },
+    })) as typeof fetch;
+
+  try {
+    await assert.rejects(
+      () => searchBrave("test", { numResults: 1 }),
+      /Brave Search API returned text\/plain/
+    );
+  } finally {
+    globalThis.fetch = originalFetch;
+    env.BRAVE_SEARCH_API_KEY = originalApiKey;
+  }
+});
+
+test("searchBrave evenly paces concurrent bursts using Brave's shortest policy window", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalApiKey = env.BRAVE_SEARCH_API_KEY;
+  const requestStartedAt: number[] = [];
+  resetBraveRateLimitStateForTests();
+  env.BRAVE_SEARCH_API_KEY = "test-brave-key";
+  globalThis.fetch = (async () => {
+    requestStartedAt.push(Date.now());
+    return new Response(JSON.stringify({ web: { results: [] } }), {
+      status: 200,
+      headers: {
+        "content-type": "application/json",
+        "x-ratelimit-policy": "1;w=1, 2000;w=2678400",
+        "x-ratelimit-remaining": "1, 1999",
+        "x-ratelimit-reset": "1, 2678400",
+      },
+    });
+  }) as typeof fetch;
+
+  try {
+    await Promise.all([
+      searchBrave("burst one", { numResults: 1 }),
+      searchBrave("burst two", { numResults: 1 }),
+      searchBrave("burst three", { numResults: 1 }),
+    ]);
+
+    assert.equal(requestStartedAt.length, 3);
+    assert.ok(requestStartedAt[1]! - requestStartedAt[0]! >= 1_000);
+    assert.ok(requestStartedAt[2]! - requestStartedAt[1]! >= 1_000);
+  } finally {
+    globalThis.fetch = originalFetch;
+    env.BRAVE_SEARCH_API_KEY = originalApiKey;
+    resetBraveRateLimitStateForTests();
+  }
+});
+
+test("searchBrave adapts its pacing to a 50 request-per-second Search plan", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalApiKey = env.BRAVE_SEARCH_API_KEY;
+  const requestStartedAt: number[] = [];
+  resetBraveRateLimitStateForTests();
+  env.BRAVE_SEARCH_API_KEY = "test-brave-key";
+  globalThis.fetch = (async () => {
+    requestStartedAt.push(Date.now());
+    return new Response(JSON.stringify({ web: { results: [] } }), {
+      status: 200,
+      headers: {
+        "content-type": "application/json",
+        "x-ratelimit-policy": "50;w=1, 0;w=2678400",
+        "x-ratelimit-remaining": "49, 0",
+        "x-ratelimit-reset": "1, 2678400",
+      },
+    });
+  }) as typeof fetch;
+
+  try {
+    await searchBrave("learn upgraded policy", { numResults: 1 });
+    await Promise.all(Array.from({ length: 8 }, (_, index) => searchBrave(`fast burst ${index}`, { numResults: 1 })));
+
+    assert.equal(requestStartedAt.length, 9);
+    const burstStartedAt = requestStartedAt.slice(1);
+    for (let index = 1; index < burstStartedAt.length; index += 1) {
+      assert.ok(burstStartedAt[index]! - burstStartedAt[index - 1]! >= 18);
+    }
+    assert.ok(burstStartedAt.at(-1)! - burstStartedAt[0]! < 500);
+  } finally {
+    globalThis.fetch = originalFetch;
+    env.BRAVE_SEARCH_API_KEY = originalApiKey;
+    resetBraveRateLimitStateForTests();
+  }
+});
+
+test("searchBrave retries 429 responses after the burst window resets", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalApiKey = env.BRAVE_SEARCH_API_KEY;
+  let fetchCount = 0;
+  resetBraveRateLimitStateForTests();
+  env.BRAVE_SEARCH_API_KEY = "test-brave-key";
+  globalThis.fetch = (async () => {
+    fetchCount += 1;
+    const rateLimitHeaders = {
+      "content-type": "application/json",
+      "x-ratelimit-policy": "1;w=1, 2000;w=2678400",
+      "x-ratelimit-remaining": fetchCount === 1 ? "0, 1999" : "1, 1998",
+      "x-ratelimit-reset": "1, 2678400",
+    };
+    if (fetchCount === 1) {
+      return new Response(JSON.stringify({ error: { detail: "Rate limit exceeded" } }), {
+        status: 429,
+        headers: rateLimitHeaders,
+      });
+    }
+    return new Response(JSON.stringify({ web: { results: [] } }), { status: 200, headers: rateLimitHeaders });
+  }) as typeof fetch;
+
+  try {
+    const startedAt = Date.now();
+    await searchBrave("retry burst", { numResults: 1 });
+    assert.equal(fetchCount, 2);
+    assert.ok(Date.now() - startedAt >= 1_000);
+  } finally {
+    globalThis.fetch = originalFetch;
+    env.BRAVE_SEARCH_API_KEY = originalApiKey;
+    resetBraveRateLimitStateForTests();
+  }
+});
+
+test("searchBrave does not wait for exhausted long-term quotas", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalApiKey = env.BRAVE_SEARCH_API_KEY;
+  resetBraveRateLimitStateForTests();
+  env.BRAVE_SEARCH_API_KEY = "test-brave-key";
+  globalThis.fetch = (async () =>
+    new Response(JSON.stringify({ error: { detail: "Quota exceeded" } }), {
+      status: 429,
+      headers: {
+        "content-type": "application/json",
+        "x-ratelimit-policy": "1;w=1, 2000;w=2678400",
+        "x-ratelimit-remaining": "0, 0",
+        "x-ratelimit-reset": "1, 100000",
+      },
+    })) as typeof fetch;
+
+  try {
+    const startedAt = Date.now();
+    await assert.rejects(
+      () => searchBrave("quota exhausted", { numResults: 1 }),
+      /rate limit quota is exhausted beyond the retry window/
+    );
+    assert.ok(Date.now() - startedAt < 1_000);
+  } finally {
+    globalThis.fetch = originalFetch;
+    env.BRAVE_SEARCH_API_KEY = originalApiKey;
+    resetBraveRateLimitStateForTests();
+  }
+});
+
+test("searchBrave blocks locally after a successful request exhausts the long-term quota", async () => {
+  const originalFetch = globalThis.fetch;
+  const originalApiKey = env.BRAVE_SEARCH_API_KEY;
+  let fetchCount = 0;
+  resetBraveRateLimitStateForTests();
+  env.BRAVE_SEARCH_API_KEY = "test-brave-key";
+  globalThis.fetch = (async () => {
+    fetchCount += 1;
+    return new Response(JSON.stringify({ web: { results: [] } }), {
+      status: 200,
+      headers: {
+        "content-type": "application/json",
+        "x-ratelimit-policy": "1;w=1, 2000;w=2678400",
+        "x-ratelimit-remaining": "0, 0",
+        "x-ratelimit-reset": "1, 100000",
+      },
+    });
+  }) as typeof fetch;
+
+  try {
+    await searchBrave("last allowed query", { numResults: 1 });
+    await assert.rejects(
+      () => searchBrave("over quota query", { numResults: 1 }),
+      /long-term quota is exhausted/
+    );
+    assert.equal(fetchCount, 1);
+  } finally {
+    globalThis.fetch = originalFetch;
+    env.BRAVE_SEARCH_API_KEY = originalApiKey;
+    resetBraveRateLimitStateForTests();
+  }
+});

web/package.json

@@ -7,6 +7,7 @@
     "dev": "vite",
     "build": "tsc -b && vite build",
     "preview": "vite preview",
+    "test": "node --test --experimental-strip-types tests/*.test.mjs",
     "typecheck": "tsc --noEmit"
   },
   "dependencies": {

web/public/sw.js

@@ -3,10 +3,24 @@ self.addEventListener("install", () => {
 });
 
 self.addEventListener("activate", (event) => {
-  event.waitUntil(self.clients.claim());
+  event.waitUntil(
+    (async () => {
+      await self.clients.claim();
+      const windows = await self.clients.matchAll({ type: "window", includeUncontrolled: true });
+      await Promise.all(
+        windows.map(async (client) => {
+          try {
+            await client.navigate(client.url);
+          } catch {
+            // The client may have closed while the new worker was activating.
+          }
+        })
+      );
+    })()
+  );
 });
 
 self.addEventListener("fetch", (event) => {
   if (event.request.mode !== "navigate") return;
-  event.respondWith(fetch(event.request));
+  event.respondWith(fetch(new Request(event.request, { cache: "no-store" })));
 });

web/src/components/chat/chat-messages-panel.tsx

@@ -478,6 +478,7 @@ export function ChatMessagesPanel({ messages, isLoading, isSending }: Props) {
                 ) : message.content.trim() ? (
                   <MarkdownContent
                     markdown={message.content}
+                    openLinksInNewTab
                     className={cn("[&_a]:text-inherit [&_a]:underline", isUser ? "leading-[1.78] text-fuchsia-50" : "leading-[1.82] text-violet-50")}
                   />
                 ) : null}

web/src/components/markdown/markdown-content.tsx

@@ -10,6 +10,7 @@ type Props = {
   className?: string;
   mode?: MarkdownMode;
   resolveCitationIndex?: (href: string) => number | undefined;
+  openLinksInNewTab?: boolean;
 };
 
 function replaceMarkdownLinksWithCitationTokens(markdown: string, resolveCitationIndex?: (href: string) => number | undefined) {
@@ -28,17 +29,30 @@ markdownRenderer.table = (token) => {
   return `<div class="md-table-scroll">${renderTable(token)}</div>`;
 };
 
-function renderMarkdown(markdown: string) {
+function setNewTabLinkAttributes(currentNode: Element) {
-  const rawHtml = marked.parse(markdown, { gfm: true, breaks: true, renderer: markdownRenderer }) as string;
+  if (currentNode.tagName !== "A") return;
-  return DOMPurify.sanitize(rawHtml, { ADD_ATTR: ["class", "target", "rel"] });
+  currentNode.setAttribute("target", "_blank");
+  currentNode.setAttribute("rel", "noopener noreferrer");
 }
 
-export function MarkdownContent({ markdown, className, mode = "default", resolveCitationIndex }: Props) {
+function renderMarkdown(markdown: string, openLinksInNewTab: boolean) {
+  const rawHtml = marked.parse(markdown, { gfm: true, breaks: true, renderer: markdownRenderer }) as string;
+  if (!openLinksInNewTab) return DOMPurify.sanitize(rawHtml, { ADD_ATTR: ["class", "target", "rel"] });
+
+  DOMPurify.addHook("afterSanitizeAttributes", setNewTabLinkAttributes);
+  try {
+    return DOMPurify.sanitize(rawHtml, { ADD_ATTR: ["class", "target", "rel"] });
+  } finally {
+    DOMPurify.removeHook("afterSanitizeAttributes", setNewTabLinkAttributes);
+  }
+}
+
+export function MarkdownContent({ markdown, className, mode = "default", resolveCitationIndex, openLinksInNewTab = false }: Props) {
   const html = useMemo(() => {
     const prepared =
       mode === "citationTokens" ? replaceMarkdownLinksWithCitationTokens(markdown, resolveCitationIndex) : markdown;
-    return renderMarkdown(prepared);
+    return renderMarkdown(prepared, openLinksInNewTab);
-  }, [markdown, mode, resolveCitationIndex]);
+  }, [markdown, mode, openLinksInNewTab, resolveCitationIndex]);
 
   return <div className={cn("md-content", className)} dangerouslySetInnerHTML={{ __html: html }} />;
 }

web/src/lib/api.ts

@@ -450,6 +450,7 @@ async function readSseStream(response: Response, dispatch: (eventName: string, p
   let buffer = "";
   let eventName = "message";
   let dataLines: string[] = [];
+  let sawTerminalEvent = false;
 
   const flushEvent = () => {
     if (!dataLines.length) {
@@ -466,6 +467,9 @@ async function readSseStream(response: Response, dispatch: (eventName: string, p
     }
 
     dispatch(eventName, payload);
+    if (eventName === "done" || eventName === "error") {
+      sawTerminalEvent = true;
+    }
 
     dataLines = [];
     eventName = "message";
@@ -505,6 +509,10 @@ async function readSseStream(response: Response, dispatch: (eventName: string, p
     }
   }
   flushEvent();
+
+  if (!sawTerminalEvent) {
+    throw new Error("Stream disconnected before completion");
+  }
 }
 
 export async function runSearchStream(
@@ -528,87 +536,14 @@ export async function runSearchStream(
     signal: options?.signal,
   });
 
-  if (!response.ok) {
+  await readSseStream(response, (eventName, payload) => {
-    const fallback = `${response.status} ${response.statusText}`;
-    let message = fallback;
-    try {
-      const body = (await response.json()) as { message?: string };
-      if (body.message) message = body.message;
-    } catch {
-      // keep fallback message
-    }
-    throw new Error(message);
-  }
-
-  if (!response.body) {
-    throw new Error("No response stream");
-  }
-
-  const reader = response.body.getReader();
-  const decoder = new TextDecoder();
-  let buffer = "";
-  let eventName = "message";
-  let dataLines: string[] = [];
-
-  const flushEvent = () => {
-    if (!dataLines.length) {
-      eventName = "message";
-      return;
-    }
-
-    const dataText = dataLines.join("\n");
-    let payload: any = null;
-    try {
-      payload = JSON.parse(dataText);
-    } catch {
-      payload = { message: dataText };
-    }
-
     if (eventName === "search_results") handlers.onSearchResults?.(payload);
     else if (eventName === "search_error") handlers.onSearchError?.(payload);
     else if (eventName === "answer") handlers.onAnswer?.(payload);
     else if (eventName === "answer_error") handlers.onAnswerError?.(payload);
     else if (eventName === "done") handlers.onDone?.(payload);
     else if (eventName === "error") handlers.onError?.(payload);
-
+  });
-    dataLines = [];
-    eventName = "message";
-  };
-
-  while (true) {
-    const { value, done } = await reader.read();
-    if (done) break;
-
-    buffer += decoder.decode(value, { stream: true });
-    let newlineIndex = buffer.indexOf("\n");
-
-    while (newlineIndex >= 0) {
-      const rawLine = buffer.slice(0, newlineIndex);
-      buffer = buffer.slice(newlineIndex + 1);
-      const line = rawLine.endsWith("\r") ? rawLine.slice(0, -1) : rawLine;
-
-      if (!line) {
-        flushEvent();
-      } else if (line.startsWith("event:")) {
-        eventName = line.slice("event:".length).trim();
-      } else if (line.startsWith("data:")) {
-        dataLines.push(line.slice("data:".length).trimStart());
-      }
-
-      newlineIndex = buffer.indexOf("\n");
-    }
-  }
-
-  buffer += decoder.decode();
-  if (buffer.length) {
-    const line = buffer.endsWith("\r") ? buffer.slice(0, -1) : buffer;
-    if (line.startsWith("event:")) {
-      eventName = line.slice("event:".length).trim();
-    } else if (line.startsWith("data:")) {
-      dataLines.push(line.slice("data:".length).trimStart());
-    }
-  }
-  flushEvent();
 }
 
 export async function attachSearchStream(searchId: string, handlers: RunSearchStreamHandlers, options?: { signal?: AbortSignal }) {
@@ -654,6 +589,7 @@ export async function runCompletionStream(
   body: {
     chatId?: string | null;
     persist?: boolean;
+    clientRequestId?: string;
     provider: Provider;
     model: string;
     messages: CompletionRequestMessage[];
@@ -679,86 +615,13 @@ export async function runCompletionStream(
     signal: options?.signal,
   });
 
-  if (!response.ok) {
+  await readSseStream(response, (eventName, payload) => {
-    const fallback = `${response.status} ${response.statusText}`;
-    let message = fallback;
-    try {
-      const body = (await response.json()) as { message?: string };
-      if (body.message) message = body.message;
-    } catch {
-      // keep fallback message
-    }
-    throw new Error(message);
-  }
-
-  if (!response.body) {
-    throw new Error("No response stream");
-  }
-
-  const reader = response.body.getReader();
-  const decoder = new TextDecoder();
-  let buffer = "";
-  let eventName = "message";
-  let dataLines: string[] = [];
-
-  const flushEvent = () => {
-    if (!dataLines.length) {
-      eventName = "message";
-      return;
-    }
-
-    const dataText = dataLines.join("\n");
-    let payload: any = null;
-    try {
-      payload = JSON.parse(dataText);
-    } catch {
-      payload = { message: dataText };
-    }
-
     if (eventName === "meta") handlers.onMeta?.(payload);
     else if (eventName === "tool_call") handlers.onToolCall?.(payload);
     else if (eventName === "delta") handlers.onDelta?.(payload);
     else if (eventName === "done") handlers.onDone?.(payload);
     else if (eventName === "error") handlers.onError?.(payload);
-
+  });
-    dataLines = [];
-    eventName = "message";
-  };
-
-  while (true) {
-    const { value, done } = await reader.read();
-    if (done) break;
-
-    buffer += decoder.decode(value, { stream: true });
-    let newlineIndex = buffer.indexOf("\n");
-
-    while (newlineIndex >= 0) {
-      const rawLine = buffer.slice(0, newlineIndex);
-      buffer = buffer.slice(newlineIndex + 1);
-      const line = rawLine.endsWith("\r") ? rawLine.slice(0, -1) : rawLine;
-
-      if (!line) {
-        flushEvent();
-      } else if (line.startsWith("event:")) {
-        eventName = line.slice("event:".length).trim();
-      } else if (line.startsWith("data:")) {
-        dataLines.push(line.slice("data:".length).trimStart());
-      }
-
-      newlineIndex = buffer.indexOf("\n");
-    }
-  }
-
-  buffer += decoder.decode();
-  if (buffer.length) {
-    const line = buffer.endsWith("\r") ? buffer.slice(0, -1) : buffer;
-    if (line.startsWith("event:")) {
-      eventName = line.slice("event:".length).trim();
-    } else if (line.startsWith("data:")) {
-      dataLines.push(line.slice("data:".length).trimStart());
-    }
-  }
-  flushEvent();
 }
 
 export async function attachCompletionStream(chatId: string, handlers: CompletionStreamHandlers, options?: { signal?: AbortSignal }) {

web/src/lib/chat-model-selection.ts

@@ -0,0 +1,24 @@
+import type { Provider } from "./api";
+
+type PersistedChatModel = {
+  lastUsedProvider: Provider | null;
+  lastUsedModel: string | null;
+};
+
+export type ChatModelSelection = {
+  provider: Provider;
+  model: string;
+};
+
+export function getChatModelSelection(chat: PersistedChatModel | null): ChatModelSelection | null {
+  if (!chat?.lastUsedProvider || !chat.lastUsedModel?.trim()) return null;
+  return {
+    provider: chat.lastUsedProvider,
+    model: chat.lastUsedModel.trim(),
+  };
+}
+
+export function getChatModelSelectionSyncKey(chatId: string | null, selection: ChatModelSelection | null) {
+  if (!chatId || !selection) return null;
+  return JSON.stringify([chatId, selection.provider, selection.model]);
+}

web/src/lib/sidebar-selection.ts

@@ -0,0 +1,31 @@
+export type SidebarSelection = { kind: "chat" | "search"; id: string };
+
+type WorkspaceSelectionItem = { type: SidebarSelection["kind"]; id: string };
+
+type ResolveSidebarSelectionOptions = {
+  initialSelection?: SidebarSelection;
+  selectFallback?: boolean;
+};
+
+export function resolveSidebarSelectionAfterRefresh(
+  current: SidebarSelection | null,
+  workspaceItems: WorkspaceSelectionItem[],
+  { initialSelection, selectFallback = false }: ResolveSidebarSelectionOptions = {}
+): SidebarSelection | null {
+  const hasItem = (candidate: SidebarSelection | null | undefined) => {
+    if (!candidate) return false;
+    return workspaceItems.some((item) => item.type === candidate.kind && item.id === candidate.id);
+  };
+
+  if (hasItem(current)) {
+    return current;
+  }
+  if (hasItem(initialSelection)) {
+    return initialSelection ?? null;
+  }
+  if (!selectFallback) {
+    return null;
+  }
+  const first = workspaceItems[0];
+  return first ? { kind: first.type, id: first.id } : null;
+}

web/src/pwa.ts

@@ -2,8 +2,11 @@ export function registerServiceWorker() {
   if (!import.meta.env.PROD || !("serviceWorker" in navigator)) return;
 
   window.addEventListener("load", () => {
-    void navigator.serviceWorker.register("/sw.js").catch((error: unknown) => {
+    void navigator.serviceWorker
-      console.warn("Sybil service worker registration failed", error);
+      .register("/sw.js", { updateViaCache: "none" })
-    });
+      .then((registration) => registration.update())
+      .catch((error: unknown) => {
+        console.warn("Sybil service worker registration failed", error);
+      });
   });
 }

web/tests/chat-model-selection.test.mjs

@@ -0,0 +1,49 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+import {
+  getChatModelSelection,
+  getChatModelSelectionSyncKey,
+} from "../src/lib/chat-model-selection.ts";
+
+test("chat model selections are normalized from persisted metadata", () => {
+  assert.deepEqual(
+    getChatModelSelection({
+      lastUsedProvider: "anthropic",
+      lastUsedModel: "  claude-sonnet-4-5  ",
+    }),
+    {
+      provider: "anthropic",
+      model: "claude-sonnet-4-5",
+    }
+  );
+});
+
+test("unrelated chat updates do not change the model synchronization key", () => {
+  const beforeSettingsSave = getChatModelSelection({
+    lastUsedProvider: "openai",
+    lastUsedModel: "gpt-4.1-mini",
+  });
+  const afterSettingsSave = getChatModelSelection({
+    lastUsedProvider: "openai",
+    lastUsedModel: "gpt-4.1-mini",
+  });
+
+  assert.equal(
+    getChatModelSelectionSyncKey("chat-1", beforeSettingsSave),
+    getChatModelSelectionSyncKey("chat-1", afterSettingsSave)
+  );
+});
+
+test("switching chats or persisted models changes the synchronization key", () => {
+  const original = { provider: "openai", model: "gpt-4.1-mini" };
+  const updated = { provider: "gemini", model: "gemini-3.5-flash" };
+
+  assert.notEqual(
+    getChatModelSelectionSyncKey("chat-1", original),
+    getChatModelSelectionSyncKey("chat-2", original)
+  );
+  assert.notEqual(
+    getChatModelSelectionSyncKey("chat-1", original),
+    getChatModelSelectionSyncKey("chat-1", updated)
+  );
+});

web/tests/sidebar-selection.test.mjs

@@ -0,0 +1,45 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+import { resolveSidebarSelectionAfterRefresh } from "../src/lib/sidebar-selection.ts";
+
+const workspaceItems = [
+  { type: "chat", id: "completed-chat" },
+  { type: "chat", id: "selected-chat" },
+  { type: "search", id: "selected-search" },
+];
+
+test("a collection refresh preserves the current thread selection", () => {
+  assert.deepEqual(
+    resolveSidebarSelectionAfterRefresh({ kind: "chat", id: "selected-chat" }, workspaceItems),
+    { kind: "chat", id: "selected-chat" }
+  );
+});
+
+test("an initial route selection cannot override a current thread selection", () => {
+  assert.deepEqual(
+    resolveSidebarSelectionAfterRefresh(
+      { kind: "search", id: "selected-search" },
+      workspaceItems,
+      { initialSelection: { kind: "chat", id: "completed-chat" }, selectFallback: true }
+    ),
+    { kind: "search", id: "selected-search" }
+  );
+});
+
+test("a collection refresh preserves an intentionally empty selection", () => {
+  assert.equal(resolveSidebarSelectionAfterRefresh(null, workspaceItems), null);
+});
+
+test("initial load can select the URL thread or fall back to the first item", () => {
+  assert.deepEqual(
+    resolveSidebarSelectionAfterRefresh(null, workspaceItems, {
+      initialSelection: { kind: "search", id: "selected-search" },
+      selectFallback: true,
+    }),
+    { kind: "search", id: "selected-search" }
+  );
+  assert.deepEqual(resolveSidebarSelectionAfterRefresh(null, workspaceItems, { selectFallback: true }), {
+    kind: "chat",
+    id: "completed-chat",
+  });
+});

web/tsconfig.tsbuildinfo

@@ -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/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/chat-model-selection.ts","./src/lib/sidebar-selection.ts","./src/lib/utils.ts","./src/pages/search-route-page.tsx"],"version":"5.9.3"}