Sybil-2/server/tests/chat-tools-streaming.test.ts

import assert from "node:assert/strict";
import test from "node:test";
import {
  runPlainChatCompletionsStream,
  runToolAwareChatCompletionsStream,
  runToolAwareOpenAIChatStream,
  type ToolAwareStreamingEvent,
} from "../src/llm/chat-tools.js";

async function* streamFrom(events: any[]) {
  for (const event of events) {
    await Promise.resolve();
    yield event;
  }
}

async function collectEvents(iterable: AsyncIterable<ToolAwareStreamingEvent>) {
  const events: ToolAwareStreamingEvent[] = [];
  for await (const event of iterable) {
    events.push(event);
  }
  return events;
}

test("OpenAI Responses stream emits text deltas as they arrive", async () => {
  const outputMessage = {
    id: "msg_1",
    type: "message",
    role: "assistant",
    status: "completed",
    content: [{ type: "output_text", text: "Hello" }],
  };
  const client = {
    responses: {
      create: async () =>
        streamFrom([
          { type: "response.output_item.added", item: { ...outputMessage, content: [] }, output_index: 0 },
          { type: "response.output_text.delta", delta: "Hel", output_index: 0, content_index: 0 },
          { type: "response.output_text.delta", delta: "lo", output_index: 0, content_index: 0 },
          { type: "response.output_item.done", item: outputMessage, output_index: 0 },
          {
            type: "response.completed",
            response: {
              status: "completed",
              output_text: "Hello",
              output: [outputMessage],
              usage: { input_tokens: 2, output_tokens: 1, total_tokens: 3 },
            },
          },
        ]),
    },
  };

  const events = await collectEvents(
    runToolAwareOpenAIChatStream({
      client: client as any,
      model: "gpt-test",
      messages: [{ role: "user", content: "Say hello" }],
    })
  );

  assert.deepEqual(
    events.map((event) => event.type),
    ["delta", "delta", "done"]
  );
  assert.deepEqual(
    events.filter((event) => event.type === "delta").map((event) => event.text),
    ["Hel", "lo"]
  );
  assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.text : null, "Hello");
});

test("OpenAI-compatible Chat Completions stream emits text deltas as they arrive", async () => {
  const client = {
    chat: {
      completions: {
        create: async () =>
          streamFrom([
            { choices: [{ delta: { role: "assistant" } }] },
            { choices: [{ delta: { content: "Hel" } }] },
            { choices: [{ delta: { content: "lo" } }] },
            {
              choices: [{ delta: {}, finish_reason: "stop" }],
              usage: { prompt_tokens: 2, completion_tokens: 1, total_tokens: 3 },
            },
          ]),
      },
    },
  };

  const events = await collectEvents(
    runToolAwareChatCompletionsStream({
      client: client as any,
      model: "grok-test",
      messages: [{ role: "user", content: "Say hello" }],
    })
  );

  assert.deepEqual(
    events.map((event) => event.type),
    ["delta", "delta", "done"]
  );
  assert.deepEqual(
    events.filter((event) => event.type === "delta").map((event) => event.text),
    ["Hel", "lo"]
  );
  assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.text : null, "Hello");
});

test("plain Chat Completions stream does not send Sybil-managed tools", async () => {
  let requestBody: any = null;
  const client = {
    chat: {
      completions: {
        create: async (body: any) => {
          requestBody = body;
          return streamFrom([
            { choices: [{ delta: { content: "Hi" } }] },
            { choices: [{ delta: {}, finish_reason: "stop" }] },
          ]);
        },
      },
    },
  };

  const events = await collectEvents(
    runPlainChatCompletionsStream({
      client: client as any,
      model: "hermes-agent",
      messages: [{ role: "user", content: "Say hi" }],
    })
  );

  assert.equal(requestBody.model, "hermes-agent");
  assert.equal(requestBody.stream, true);
  assert.equal("tools" in requestBody, false);
  assert.deepEqual(
    events.map((event) => event.type),
    ["delta", "done"]
  );
  assert.equal(events.at(-1)?.type === "done" ? events.at(-1)?.result.text : null, "Hi");
});

test("OpenAI-compatible Chat Completions 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(
    runToolAwareChatCompletionsStream({
      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");
});