@livekit/agents-plugin-cerebras 1.2.6

package/dist/llm.test.cjs

@@ -0,0 +1,208 @@
+"use strict";
+var __create = Object.create;
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __getProtoOf = Object.getPrototypeOf;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
+  // If the importer is in node compatibility mode or this is not an ESM
+  // file that has been converted to a CommonJS file using a Babel-
+  // compatible transform (i.e. "__esModule" has not been set), then set
+  // "default" to the CommonJS "module.exports" for node compatibility.
+  isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
+  mod
+));
+var import_agents = require("@livekit/agents");
+var import_msgpack = require("@msgpack/msgpack");
+var import_node_zlib = require("node:zlib");
+var import_openai = __toESM(require("openai"), 1);
+var import_vitest = require("vitest");
+var import_zod = require("zod");
+var import_llm = require("./llm.cjs");
+(0, import_vitest.assert)(process.env.CEREBRAS_API_KEY, "CEREBRAS_API_KEY must be set");
+const CHAT_MODEL = "llama3.1-8b";
+const TOOL_MODEL = "qwen-3-235b-a22b-instruct-2507";
+function createCapturingFetch(opts) {
+  const capturedRequests = [];
+  const fetch = async (input, init) => {
+    if ((init == null ? void 0 : init.method) === "POST" && init.body && typeof init.body === "string") {
+      const headers = new Headers(init.headers);
+      let body;
+      if (opts.useMsgpack) {
+        body = (0, import_msgpack.encode)(JSON.parse(init.body));
+        headers.set("Content-Type", "application/vnd.msgpack");
+      } else {
+        body = new TextEncoder().encode(init.body);
+      }
+      if (opts.useGzip) {
+        body = (0, import_node_zlib.gzipSync)(body, { level: 5 });
+        headers.set("Content-Encoding", "gzip");
+      }
+      capturedRequests.push({ url: extractUrl(input), headers });
+      return globalThis.fetch(input, { ...init, body: Buffer.from(body), headers });
+    }
+    capturedRequests.push({ url: extractUrl(input), headers: new Headers(init == null ? void 0 : init.headers) });
+    return globalThis.fetch(input, init);
+  };
+  return { fetch, capturedRequests };
+}
+function extractUrl(input) {
+  if (typeof input === "string") return input;
+  if (input instanceof URL) return input.toString();
+  return input.url;
+}
+function cerebrasLLM(opts = {}) {
+  return new import_llm.LLM({ model: CHAT_MODEL, ...opts });
+}
+function cerebrasLLMWithCapture(opts) {
+  const { fetch, capturedRequests } = createCapturingFetch(opts);
+  const client = new import_openai.default({
+    apiKey: process.env.CEREBRAS_API_KEY,
+    baseURL: "https://api.cerebras.ai/v1",
+    fetch
+  });
+  return { llm: new import_llm.LLM({ model: CHAT_MODEL, client }), capturedRequests };
+}
+class WeatherAgent extends import_agents.voice.Agent {
+  constructor() {
+    super({
+      instructions: "You are a helpful assistant.",
+      tools: {
+        get_weather: import_agents.llm.tool({
+          description: "Get the current weather for a location.",
+          parameters: import_zod.z.object({
+            location: import_zod.z.string().describe("The city name")
+          }),
+          execute: async ({ location }) => {
+            return `The weather in ${location} is sunny, 72\xB0F.`;
+          }
+        })
+      }
+    });
+  }
+}
+(0, import_vitest.describe)("Cerebras", { timeout: 3e4 }, () => {
+  (0, import_vitest.it)("basic chat completion returns a non-empty assistant message", async () => {
+    const session = new import_agents.voice.AgentSession({ llm: cerebrasLLM() });
+    await session.start({
+      agent: new import_agents.voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+  });
+  (0, import_vitest.it)("LLM can invoke a tool and the result is returned", async () => {
+    const session = new import_agents.voice.AgentSession({ llm: new import_llm.LLM({ model: TOOL_MODEL }) });
+    await session.start({ agent: new WeatherAgent() });
+    const result = session.run({ userInput: "What is the weather in Tokyo?" });
+    await result.wait();
+    result.expect.nextEvent().isFunctionCall({
+      name: "get_weather",
+      args: { location: "Tokyo" }
+    });
+    result.expect.nextEvent().isFunctionCallOutput({
+      output: JSON.stringify("The weather in Tokyo is sunny, 72\xB0F.")
+    });
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+  });
+  (0, import_vitest.it)("gzip-only sends Content-Encoding: gzip with JSON content type", async () => {
+    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({
+      useGzip: true,
+      useMsgpack: false
+    });
+    const session = new import_agents.voice.AgentSession({ llm: model });
+    await session.start({
+      agent: new import_agents.voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+    const chatReqs = capturedRequests.filter((r) => r.url.includes("/chat/completions"));
+    (0, import_vitest.expect)(chatReqs.length).toBeGreaterThan(0);
+    (0, import_vitest.expect)(chatReqs[0].headers.get("content-type")).toBe("application/json");
+    (0, import_vitest.expect)(chatReqs[0].headers.get("content-encoding")).toBe("gzip");
+  });
+  (0, import_vitest.it)("msgpack-only sends Content-Type: application/vnd.msgpack without gzip", async () => {
+    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({
+      useGzip: false,
+      useMsgpack: true
+    });
+    const session = new import_agents.voice.AgentSession({ llm: model });
+    await session.start({
+      agent: new import_agents.voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+    const chatReqs = capturedRequests.filter((r) => r.url.includes("/chat/completions"));
+    (0, import_vitest.expect)(chatReqs.length).toBeGreaterThan(0);
+    (0, import_vitest.expect)(chatReqs[0].headers.get("content-type")).toBe("application/vnd.msgpack");
+    (0, import_vitest.expect)(chatReqs[0].headers.get("content-encoding")).toBeNull();
+  });
+  (0, import_vitest.it)("both flags send msgpack content type with gzip encoding", async () => {
+    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({
+      useGzip: true,
+      useMsgpack: true
+    });
+    const session = new import_agents.voice.AgentSession({ llm: model });
+    await session.start({
+      agent: new import_agents.voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+    const chatReqs = capturedRequests.filter((r) => r.url.includes("/chat/completions"));
+    (0, import_vitest.expect)(chatReqs.length).toBeGreaterThan(0);
+    (0, import_vitest.expect)(chatReqs[0].headers.get("content-type")).toBe("application/vnd.msgpack");
+    (0, import_vitest.expect)(chatReqs[0].headers.get("content-encoding")).toBe("gzip");
+  });
+  (0, import_vitest.it)("with both flags off sends standard JSON without gzip", async () => {
+    const session = new import_agents.voice.AgentSession({
+      llm: cerebrasLLM({ gzipCompression: false, msgpackEncoding: false })
+    });
+    await session.start({
+      agent: new import_agents.voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+  });
+  (0, import_vitest.it)("streaming chat returns content via the LLM directly", async () => {
+    var _a;
+    const model = cerebrasLLM();
+    const chatCtx = new import_agents.llm.ChatContext();
+    chatCtx.addMessage({ role: "system", content: "You are a helpful assistant." });
+    chatCtx.addMessage({ role: "user", content: "Count from 1 to 5." });
+    const stream = model.chat({ chatCtx });
+    let text = "";
+    for await (const chunk of stream) {
+      if ((_a = chunk.delta) == null ? void 0 : _a.content) {
+        text += chunk.delta.content;
+      }
+    }
+    (0, import_vitest.expect)(text.length).toBeGreaterThan(0);
+    (0, import_vitest.expect)(text).toContain("3");
+  });
+});
+//# sourceMappingURL=llm.test.cjs.map

package/dist/llm.test.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/llm.test.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2026 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport { llm, voice } from '@livekit/agents';\nimport { encode } from '@msgpack/msgpack';\nimport { gzipSync } from 'node:zlib';\nimport OpenAI from 'openai';\nimport { assert, describe, expect, it } from 'vitest';\nimport { z } from 'zod';\nimport type { LLMOptions } from './llm.js';\nimport { LLM } from './llm.js';\n\nassert(process.env.CEREBRAS_API_KEY, 'CEREBRAS_API_KEY must be set');\n\n// llama3.1-8b is fast and has generous rate limits but can't do tool calls reliably;\n// qwen-3-235b is needed for function calling but has tight per-minute token quotas.\nconst CHAT_MODEL = 'llama3.1-8b';\nconst TOOL_MODEL = 'qwen-3-235b-a22b-instruct-2507';\n\ninterface CapturedRequest {\n  url: string;\n  headers: Headers;\n}\n\n/**\n * Wraps a real fetch, applying msgpack/gzip compression and capturing outgoing\n * request metadata for assertion.  TypeScript equivalent of Python's\n * `HeaderCapturingTransport` + `_CerebrasClient`.\n */\nfunction createCapturingFetch(opts: { useMsgpack: boolean; useGzip: boolean }): {\n  fetch: typeof globalThis.fetch;\n  capturedRequests: CapturedRequest[];\n} {\n  const capturedRequests: CapturedRequest[] = [];\n\n  const fetch = async (input: RequestInfo | URL, init?: RequestInit): Promise<Response> => {\n    if (init?.method === 'POST' && init.body && typeof init.body === 'string') {\n      const headers = new Headers(init.headers);\n\n      let body: Uint8Array;\n      if (opts.useMsgpack) {\n        body = encode(JSON.parse(init.body));\n        headers.set('Content-Type', 'application/vnd.msgpack');\n      } else {\n        body = new TextEncoder().encode(init.body);\n      }\n\n      if (opts.useGzip) {\n        body = gzipSync(body, { level: 5 });\n        headers.set('Content-Encoding', 'gzip');\n      }\n\n      capturedRequests.push({ url: extractUrl(input), headers });\n      return globalThis.fetch(input, { ...init, body: Buffer.from(body), headers });\n    }\n\n    capturedRequests.push({ url: extractUrl(input), headers: new Headers(init?.headers) });\n    return globalThis.fetch(input, init);\n  };\n\n  return { fetch: fetch as typeof globalThis.fetch, capturedRequests };\n}\n\nfunction extractUrl(input: RequestInfo | URL): string {\n  if (typeof input === 'string') return input;\n  if (input instanceof URL) return input.toString();\n  return input.url;\n}\n\nfunction cerebrasLLM(opts: Partial<LLMOptions> = {}): LLM {\n  return new LLM({ model: CHAT_MODEL, ...opts });\n}\n\nfunction cerebrasLLMWithCapture(opts: { useGzip: boolean; useMsgpack: boolean }): {\n  llm: LLM;\n  capturedRequests: CapturedRequest[];\n} {\n  const { fetch, capturedRequests } = createCapturingFetch(opts);\n  const client = new OpenAI({\n    apiKey: process.env.CEREBRAS_API_KEY,\n    baseURL: 'https://api.cerebras.ai/v1',\n    fetch,\n  });\n  return { llm: new LLM({ model: CHAT_MODEL, client }), capturedRequests };\n}\n\nclass WeatherAgent extends voice.Agent {\n  constructor() {\n    super({\n      instructions: 'You are a helpful assistant.',\n      tools: {\n        get_weather: llm.tool({\n          description: 'Get the current weather for a location.',\n          parameters: z.object({\n            location: z.string().describe('The city name'),\n          }),\n          execute: async ({ location }) => {\n            return `The weather in ${location} is sunny, 72°F.`;\n          },\n        }),\n      },\n    });\n  }\n}\n\ndescribe('Cerebras', { timeout: 30_000 }, () => {\n  it('basic chat completion returns a non-empty assistant message', async () => {\n    const session = new voice.AgentSession({ llm: cerebrasLLM() });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n  });\n\n  it('LLM can invoke a tool and the result is returned', async () => {\n    const session = new voice.AgentSession({ llm: new LLM({ model: TOOL_MODEL }) });\n    await session.start({ agent: new WeatherAgent() });\n\n    const result = session.run({ userInput: 'What is the weather in Tokyo?' });\n    await result.wait();\n\n    result.expect.nextEvent().isFunctionCall({\n      name: 'get_weather',\n      args: { location: 'Tokyo' },\n    });\n    result.expect.nextEvent().isFunctionCallOutput({\n      output: JSON.stringify('The weather in Tokyo is sunny, 72°F.'),\n    });\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n  });\n\n  it('gzip-only sends Content-Encoding: gzip with JSON content type', async () => {\n    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({\n      useGzip: true,\n      useMsgpack: false,\n    });\n    const session = new voice.AgentSession({ llm: model });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n\n    const chatReqs = capturedRequests.filter((r) => r.url.includes('/chat/completions'));\n    expect(chatReqs.length).toBeGreaterThan(0);\n    expect(chatReqs[0]!.headers.get('content-type')).toBe('application/json');\n    expect(chatReqs[0]!.headers.get('content-encoding')).toBe('gzip');\n  });\n\n  it('msgpack-only sends Content-Type: application/vnd.msgpack without gzip', async () => {\n    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({\n      useGzip: false,\n      useMsgpack: true,\n    });\n    const session = new voice.AgentSession({ llm: model });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n\n    const chatReqs = capturedRequests.filter((r) => r.url.includes('/chat/completions'));\n    expect(chatReqs.length).toBeGreaterThan(0);\n    expect(chatReqs[0]!.headers.get('content-type')).toBe('application/vnd.msgpack');\n    expect(chatReqs[0]!.headers.get('content-encoding')).toBeNull();\n  });\n\n  it('both flags send msgpack content type with gzip encoding', async () => {\n    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({\n      useGzip: true,\n      useMsgpack: true,\n    });\n    const session = new voice.AgentSession({ llm: model });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n\n    const chatReqs = capturedRequests.filter((r) => r.url.includes('/chat/completions'));\n    expect(chatReqs.length).toBeGreaterThan(0);\n    expect(chatReqs[0]!.headers.get('content-type')).toBe('application/vnd.msgpack');\n    expect(chatReqs[0]!.headers.get('content-encoding')).toBe('gzip');\n  });\n\n  it('with both flags off sends standard JSON without gzip', async () => {\n    const session = new voice.AgentSession({\n      llm: cerebrasLLM({ gzipCompression: false, msgpackEncoding: false }),\n    });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n  });\n\n  it('streaming chat returns content via the LLM directly', async () => {\n    const model = cerebrasLLM();\n    const chatCtx = new llm.ChatContext();\n    chatCtx.addMessage({ role: 'system', content: 'You are a helpful assistant.' });\n    chatCtx.addMessage({ role: 'user', content: 'Count from 1 to 5.' });\n\n    const stream = model.chat({ chatCtx });\n    let text = '';\n    for await (const chunk of stream) {\n      if (chunk.delta?.content) {\n        text += chunk.delta.content;\n      }\n    }\n\n    expect(text.length).toBeGreaterThan(0);\n    expect(text).toContain('3');\n  });\n});\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;AAGA,oBAA2B;AAC3B,qBAAuB;AACvB,uBAAyB;AACzB,oBAAmB;AACnB,oBAA6C;AAC7C,iBAAkB;AAElB,iBAAoB;AAAA,IAEpB,sBAAO,QAAQ,IAAI,kBAAkB,8BAA8B;AAInE,MAAM,aAAa;AACnB,MAAM,aAAa;AAYnB,SAAS,qBAAqB,MAG5B;AACA,QAAM,mBAAsC,CAAC;AAE7C,QAAM,QAAQ,OAAO,OAA0B,SAA0C;AACvF,SAAI,6BAAM,YAAW,UAAU,KAAK,QAAQ,OAAO,KAAK,SAAS,UAAU;AACzE,YAAM,UAAU,IAAI,QAAQ,KAAK,OAAO;AAExC,UAAI;AACJ,UAAI,KAAK,YAAY;AACnB,mBAAO,uBAAO,KAAK,MAAM,KAAK,IAAI,CAAC;AACnC,gBAAQ,IAAI,gBAAgB,yBAAyB;AAAA,MACvD,OAAO;AACL,eAAO,IAAI,YAAY,EAAE,OAAO,KAAK,IAAI;AAAA,MAC3C;AAEA,UAAI,KAAK,SAAS;AAChB,mBAAO,2BAAS,MAAM,EAAE,OAAO,EAAE,CAAC;AAClC,gBAAQ,IAAI,oBAAoB,MAAM;AAAA,MACxC;AAEA,uBAAiB,KAAK,EAAE,KAAK,WAAW,KAAK,GAAG,QAAQ,CAAC;AACzD,aAAO,WAAW,MAAM,OAAO,EAAE,GAAG,MAAM,MAAM,OAAO,KAAK,IAAI,GAAG,QAAQ,CAAC;AAAA,IAC9E;AAEA,qBAAiB,KAAK,EAAE,KAAK,WAAW,KAAK,GAAG,SAAS,IAAI,QAAQ,6BAAM,OAAO,EAAE,CAAC;AACrF,WAAO,WAAW,MAAM,OAAO,IAAI;AAAA,EACrC;AAEA,SAAO,EAAE,OAAyC,iBAAiB;AACrE;AAEA,SAAS,WAAW,OAAkC;AACpD,MAAI,OAAO,UAAU,SAAU,QAAO;AACtC,MAAI,iBAAiB,IAAK,QAAO,MAAM,SAAS;AAChD,SAAO,MAAM;AACf;AAEA,SAAS,YAAY,OAA4B,CAAC,GAAQ;AACxD,SAAO,IAAI,eAAI,EAAE,OAAO,YAAY,GAAG,KAAK,CAAC;AAC/C;AAEA,SAAS,uBAAuB,MAG9B;AACA,QAAM,EAAE,OAAO,iBAAiB,IAAI,qBAAqB,IAAI;AAC7D,QAAM,SAAS,IAAI,cAAAA,QAAO;AAAA,IACxB,QAAQ,QAAQ,IAAI;AAAA,IACpB,SAAS;AAAA,IACT;AAAA,EACF,CAAC;AACD,SAAO,EAAE,KAAK,IAAI,eAAI,EAAE,OAAO,YAAY,OAAO,CAAC,GAAG,iBAAiB;AACzE;AAEA,MAAM,qBAAqB,oBAAM,MAAM;AAAA,EACrC,cAAc;AACZ,UAAM;AAAA,MACJ,cAAc;AAAA,MACd,OAAO;AAAA,QACL,aAAa,kBAAI,KAAK;AAAA,UACpB,aAAa;AAAA,UACb,YAAY,aAAE,OAAO;AAAA,YACnB,UAAU,aAAE,OAAO,EAAE,SAAS,eAAe;AAAA,UAC/C,CAAC;AAAA,UACD,SAAS,OAAO,EAAE,SAAS,MAAM;AAC/B,mBAAO,kBAAkB,QAAQ;AAAA,UACnC;AAAA,QACF,CAAC;AAAA,MACH;AAAA,IACF,CAAC;AAAA,EACH;AACF;AAAA,IAEA,wBAAS,YAAY,EAAE,SAAS,IAAO,GAAG,MAAM;AAC9C,wBAAG,+DAA+D,YAAY;AAC5E,UAAM,UAAU,IAAI,oBAAM,aAAa,EAAE,KAAK,YAAY,EAAE,CAAC;AAC7D,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,oBAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAAA,EACtB,CAAC;AAED,wBAAG,oDAAoD,YAAY;AACjE,UAAM,UAAU,IAAI,oBAAM,aAAa,EAAE,KAAK,IAAI,eAAI,EAAE,OAAO,WAAW,CAAC,EAAE,CAAC;AAC9E,UAAM,QAAQ,MAAM,EAAE,OAAO,IAAI,aAAa,EAAE,CAAC;AAEjD,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,gCAAgC,CAAC;AACzE,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,eAAe;AAAA,MACvC,MAAM;AAAA,MACN,MAAM,EAAE,UAAU,QAAQ;AAAA,IAC5B,CAAC;AACD,WAAO,OAAO,UAAU,EAAE,qBAAqB;AAAA,MAC7C,QAAQ,KAAK,UAAU,yCAAsC;AAAA,IAC/D,CAAC;AACD,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAAA,EACtB,CAAC;AAED,wBAAG,iEAAiE,YAAY;AAC9E,UAAM,EAAE,KAAK,OAAO,iBAAiB,IAAI,uBAAuB;AAAA,MAC9D,SAAS;AAAA,MACT,YAAY;AAAA,IACd,CAAC;AACD,UAAM,UAAU,IAAI,oBAAM,aAAa,EAAE,KAAK,MAAM,CAAC;AACrD,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,oBAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAEpB,UAAM,WAAW,iBAAiB,OAAO,CAAC,MAAM,EAAE,IAAI,SAAS,mBAAmB,CAAC;AACnF,8BAAO,SAAS,MAAM,EAAE,gBAAgB,CAAC;AACzC,8BAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,cAAc,CAAC,EAAE,KAAK,kBAAkB;AACxE,8BAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,kBAAkB,CAAC,EAAE,KAAK,MAAM;AAAA,EAClE,CAAC;AAED,wBAAG,yEAAyE,YAAY;AACtF,UAAM,EAAE,KAAK,OAAO,iBAAiB,IAAI,uBAAuB;AAAA,MAC9D,SAAS;AAAA,MACT,YAAY;AAAA,IACd,CAAC;AACD,UAAM,UAAU,IAAI,oBAAM,aAAa,EAAE,KAAK,MAAM,CAAC;AACrD,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,oBAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAEpB,UAAM,WAAW,iBAAiB,OAAO,CAAC,MAAM,EAAE,IAAI,SAAS,mBAAmB,CAAC;AACnF,8BAAO,SAAS,MAAM,EAAE,gBAAgB,CAAC;AACzC,8BAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,cAAc,CAAC,EAAE,KAAK,yBAAyB;AAC/E,8BAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,kBAAkB,CAAC,EAAE,SAAS;AAAA,EAChE,CAAC;AAED,wBAAG,2DAA2D,YAAY;AACxE,UAAM,EAAE,KAAK,OAAO,iBAAiB,IAAI,uBAAuB;AAAA,MAC9D,SAAS;AAAA,MACT,YAAY;AAAA,IACd,CAAC;AACD,UAAM,UAAU,IAAI,oBAAM,aAAa,EAAE,KAAK,MAAM,CAAC;AACrD,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,oBAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAEpB,UAAM,WAAW,iBAAiB,OAAO,CAAC,MAAM,EAAE,IAAI,SAAS,mBAAmB,CAAC;AACnF,8BAAO,SAAS,MAAM,EAAE,gBAAgB,CAAC;AACzC,8BAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,cAAc,CAAC,EAAE,KAAK,yBAAyB;AAC/E,8BAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,kBAAkB,CAAC,EAAE,KAAK,MAAM;AAAA,EAClE,CAAC;AAED,wBAAG,wDAAwD,YAAY;AACrE,UAAM,UAAU,IAAI,oBAAM,aAAa;AAAA,MACrC,KAAK,YAAY,EAAE,iBAAiB,OAAO,iBAAiB,MAAM,CAAC;AAAA,IACrE,CAAC;AACD,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,oBAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAAA,EACtB,CAAC;AAED,wBAAG,uDAAuD,YAAY;AAtOxE;AAuOI,UAAM,QAAQ,YAAY;AAC1B,UAAM,UAAU,IAAI,kBAAI,YAAY;AACpC,YAAQ,WAAW,EAAE,MAAM,UAAU,SAAS,+BAA+B,CAAC;AAC9E,YAAQ,WAAW,EAAE,MAAM,QAAQ,SAAS,qBAAqB,CAAC;AAElE,UAAM,SAAS,MAAM,KAAK,EAAE,QAAQ,CAAC;AACrC,QAAI,OAAO;AACX,qBAAiB,SAAS,QAAQ;AAChC,WAAI,WAAM,UAAN,mBAAa,SAAS;AACxB,gBAAQ,MAAM,MAAM;AAAA,MACtB;AAAA,IACF;AAEA,8BAAO,KAAK,MAAM,EAAE,gBAAgB,CAAC;AACrC,8BAAO,IAAI,EAAE,UAAU,GAAG;AAAA,EAC5B,CAAC;AACH,CAAC;","names":["OpenAI"]}

package/dist/llm.test.d.cts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=llm.test.d.ts.map

package/dist/llm.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=llm.test.d.ts.map

package/dist/llm.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm.test.d.ts","sourceRoot":"","sources":["../src/llm.test.ts"],"names":[],"mappings":""}

package/dist/llm.test.js

@@ -0,0 +1,185 @@
+import { llm, voice } from "@livekit/agents";
+import { encode } from "@msgpack/msgpack";
+import { gzipSync } from "node:zlib";
+import OpenAI from "openai";
+import { assert, describe, expect, it } from "vitest";
+import { z } from "zod";
+import { LLM } from "./llm.js";
+assert(process.env.CEREBRAS_API_KEY, "CEREBRAS_API_KEY must be set");
+const CHAT_MODEL = "llama3.1-8b";
+const TOOL_MODEL = "qwen-3-235b-a22b-instruct-2507";
+function createCapturingFetch(opts) {
+  const capturedRequests = [];
+  const fetch = async (input, init) => {
+    if ((init == null ? void 0 : init.method) === "POST" && init.body && typeof init.body === "string") {
+      const headers = new Headers(init.headers);
+      let body;
+      if (opts.useMsgpack) {
+        body = encode(JSON.parse(init.body));
+        headers.set("Content-Type", "application/vnd.msgpack");
+      } else {
+        body = new TextEncoder().encode(init.body);
+      }
+      if (opts.useGzip) {
+        body = gzipSync(body, { level: 5 });
+        headers.set("Content-Encoding", "gzip");
+      }
+      capturedRequests.push({ url: extractUrl(input), headers });
+      return globalThis.fetch(input, { ...init, body: Buffer.from(body), headers });
+    }
+    capturedRequests.push({ url: extractUrl(input), headers: new Headers(init == null ? void 0 : init.headers) });
+    return globalThis.fetch(input, init);
+  };
+  return { fetch, capturedRequests };
+}
+function extractUrl(input) {
+  if (typeof input === "string") return input;
+  if (input instanceof URL) return input.toString();
+  return input.url;
+}
+function cerebrasLLM(opts = {}) {
+  return new LLM({ model: CHAT_MODEL, ...opts });
+}
+function cerebrasLLMWithCapture(opts) {
+  const { fetch, capturedRequests } = createCapturingFetch(opts);
+  const client = new OpenAI({
+    apiKey: process.env.CEREBRAS_API_KEY,
+    baseURL: "https://api.cerebras.ai/v1",
+    fetch
+  });
+  return { llm: new LLM({ model: CHAT_MODEL, client }), capturedRequests };
+}
+class WeatherAgent extends voice.Agent {
+  constructor() {
+    super({
+      instructions: "You are a helpful assistant.",
+      tools: {
+        get_weather: llm.tool({
+          description: "Get the current weather for a location.",
+          parameters: z.object({
+            location: z.string().describe("The city name")
+          }),
+          execute: async ({ location }) => {
+            return `The weather in ${location} is sunny, 72\xB0F.`;
+          }
+        })
+      }
+    });
+  }
+}
+describe("Cerebras", { timeout: 3e4 }, () => {
+  it("basic chat completion returns a non-empty assistant message", async () => {
+    const session = new voice.AgentSession({ llm: cerebrasLLM() });
+    await session.start({
+      agent: new voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+  });
+  it("LLM can invoke a tool and the result is returned", async () => {
+    const session = new voice.AgentSession({ llm: new LLM({ model: TOOL_MODEL }) });
+    await session.start({ agent: new WeatherAgent() });
+    const result = session.run({ userInput: "What is the weather in Tokyo?" });
+    await result.wait();
+    result.expect.nextEvent().isFunctionCall({
+      name: "get_weather",
+      args: { location: "Tokyo" }
+    });
+    result.expect.nextEvent().isFunctionCallOutput({
+      output: JSON.stringify("The weather in Tokyo is sunny, 72\xB0F.")
+    });
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+  });
+  it("gzip-only sends Content-Encoding: gzip with JSON content type", async () => {
+    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({
+      useGzip: true,
+      useMsgpack: false
+    });
+    const session = new voice.AgentSession({ llm: model });
+    await session.start({
+      agent: new voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+    const chatReqs = capturedRequests.filter((r) => r.url.includes("/chat/completions"));
+    expect(chatReqs.length).toBeGreaterThan(0);
+    expect(chatReqs[0].headers.get("content-type")).toBe("application/json");
+    expect(chatReqs[0].headers.get("content-encoding")).toBe("gzip");
+  });
+  it("msgpack-only sends Content-Type: application/vnd.msgpack without gzip", async () => {
+    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({
+      useGzip: false,
+      useMsgpack: true
+    });
+    const session = new voice.AgentSession({ llm: model });
+    await session.start({
+      agent: new voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+    const chatReqs = capturedRequests.filter((r) => r.url.includes("/chat/completions"));
+    expect(chatReqs.length).toBeGreaterThan(0);
+    expect(chatReqs[0].headers.get("content-type")).toBe("application/vnd.msgpack");
+    expect(chatReqs[0].headers.get("content-encoding")).toBeNull();
+  });
+  it("both flags send msgpack content type with gzip encoding", async () => {
+    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({
+      useGzip: true,
+      useMsgpack: true
+    });
+    const session = new voice.AgentSession({ llm: model });
+    await session.start({
+      agent: new voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+    const chatReqs = capturedRequests.filter((r) => r.url.includes("/chat/completions"));
+    expect(chatReqs.length).toBeGreaterThan(0);
+    expect(chatReqs[0].headers.get("content-type")).toBe("application/vnd.msgpack");
+    expect(chatReqs[0].headers.get("content-encoding")).toBe("gzip");
+  });
+  it("with both flags off sends standard JSON without gzip", async () => {
+    const session = new voice.AgentSession({
+      llm: cerebrasLLM({ gzipCompression: false, msgpackEncoding: false })
+    });
+    await session.start({
+      agent: new voice.Agent({ instructions: "You are a helpful assistant." })
+    });
+    const result = session.run({ userInput: "Say hello in exactly one word." });
+    await result.wait();
+    result.expect.nextEvent().isMessage({ role: "assistant" });
+    result.expect.noMoreEvents();
+    await session.close();
+  });
+  it("streaming chat returns content via the LLM directly", async () => {
+    var _a;
+    const model = cerebrasLLM();
+    const chatCtx = new llm.ChatContext();
+    chatCtx.addMessage({ role: "system", content: "You are a helpful assistant." });
+    chatCtx.addMessage({ role: "user", content: "Count from 1 to 5." });
+    const stream = model.chat({ chatCtx });
+    let text = "";
+    for await (const chunk of stream) {
+      if ((_a = chunk.delta) == null ? void 0 : _a.content) {
+        text += chunk.delta.content;
+      }
+    }
+    expect(text.length).toBeGreaterThan(0);
+    expect(text).toContain("3");
+  });
+});
+//# sourceMappingURL=llm.test.js.map

package/dist/llm.test.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/llm.test.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2026 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport { llm, voice } from '@livekit/agents';\nimport { encode } from '@msgpack/msgpack';\nimport { gzipSync } from 'node:zlib';\nimport OpenAI from 'openai';\nimport { assert, describe, expect, it } from 'vitest';\nimport { z } from 'zod';\nimport type { LLMOptions } from './llm.js';\nimport { LLM } from './llm.js';\n\nassert(process.env.CEREBRAS_API_KEY, 'CEREBRAS_API_KEY must be set');\n\n// llama3.1-8b is fast and has generous rate limits but can't do tool calls reliably;\n// qwen-3-235b is needed for function calling but has tight per-minute token quotas.\nconst CHAT_MODEL = 'llama3.1-8b';\nconst TOOL_MODEL = 'qwen-3-235b-a22b-instruct-2507';\n\ninterface CapturedRequest {\n  url: string;\n  headers: Headers;\n}\n\n/**\n * Wraps a real fetch, applying msgpack/gzip compression and capturing outgoing\n * request metadata for assertion.  TypeScript equivalent of Python's\n * `HeaderCapturingTransport` + `_CerebrasClient`.\n */\nfunction createCapturingFetch(opts: { useMsgpack: boolean; useGzip: boolean }): {\n  fetch: typeof globalThis.fetch;\n  capturedRequests: CapturedRequest[];\n} {\n  const capturedRequests: CapturedRequest[] = [];\n\n  const fetch = async (input: RequestInfo | URL, init?: RequestInit): Promise<Response> => {\n    if (init?.method === 'POST' && init.body && typeof init.body === 'string') {\n      const headers = new Headers(init.headers);\n\n      let body: Uint8Array;\n      if (opts.useMsgpack) {\n        body = encode(JSON.parse(init.body));\n        headers.set('Content-Type', 'application/vnd.msgpack');\n      } else {\n        body = new TextEncoder().encode(init.body);\n      }\n\n      if (opts.useGzip) {\n        body = gzipSync(body, { level: 5 });\n        headers.set('Content-Encoding', 'gzip');\n      }\n\n      capturedRequests.push({ url: extractUrl(input), headers });\n      return globalThis.fetch(input, { ...init, body: Buffer.from(body), headers });\n    }\n\n    capturedRequests.push({ url: extractUrl(input), headers: new Headers(init?.headers) });\n    return globalThis.fetch(input, init);\n  };\n\n  return { fetch: fetch as typeof globalThis.fetch, capturedRequests };\n}\n\nfunction extractUrl(input: RequestInfo | URL): string {\n  if (typeof input === 'string') return input;\n  if (input instanceof URL) return input.toString();\n  return input.url;\n}\n\nfunction cerebrasLLM(opts: Partial<LLMOptions> = {}): LLM {\n  return new LLM({ model: CHAT_MODEL, ...opts });\n}\n\nfunction cerebrasLLMWithCapture(opts: { useGzip: boolean; useMsgpack: boolean }): {\n  llm: LLM;\n  capturedRequests: CapturedRequest[];\n} {\n  const { fetch, capturedRequests } = createCapturingFetch(opts);\n  const client = new OpenAI({\n    apiKey: process.env.CEREBRAS_API_KEY,\n    baseURL: 'https://api.cerebras.ai/v1',\n    fetch,\n  });\n  return { llm: new LLM({ model: CHAT_MODEL, client }), capturedRequests };\n}\n\nclass WeatherAgent extends voice.Agent {\n  constructor() {\n    super({\n      instructions: 'You are a helpful assistant.',\n      tools: {\n        get_weather: llm.tool({\n          description: 'Get the current weather for a location.',\n          parameters: z.object({\n            location: z.string().describe('The city name'),\n          }),\n          execute: async ({ location }) => {\n            return `The weather in ${location} is sunny, 72°F.`;\n          },\n        }),\n      },\n    });\n  }\n}\n\ndescribe('Cerebras', { timeout: 30_000 }, () => {\n  it('basic chat completion returns a non-empty assistant message', async () => {\n    const session = new voice.AgentSession({ llm: cerebrasLLM() });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n  });\n\n  it('LLM can invoke a tool and the result is returned', async () => {\n    const session = new voice.AgentSession({ llm: new LLM({ model: TOOL_MODEL }) });\n    await session.start({ agent: new WeatherAgent() });\n\n    const result = session.run({ userInput: 'What is the weather in Tokyo?' });\n    await result.wait();\n\n    result.expect.nextEvent().isFunctionCall({\n      name: 'get_weather',\n      args: { location: 'Tokyo' },\n    });\n    result.expect.nextEvent().isFunctionCallOutput({\n      output: JSON.stringify('The weather in Tokyo is sunny, 72°F.'),\n    });\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n  });\n\n  it('gzip-only sends Content-Encoding: gzip with JSON content type', async () => {\n    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({\n      useGzip: true,\n      useMsgpack: false,\n    });\n    const session = new voice.AgentSession({ llm: model });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n\n    const chatReqs = capturedRequests.filter((r) => r.url.includes('/chat/completions'));\n    expect(chatReqs.length).toBeGreaterThan(0);\n    expect(chatReqs[0]!.headers.get('content-type')).toBe('application/json');\n    expect(chatReqs[0]!.headers.get('content-encoding')).toBe('gzip');\n  });\n\n  it('msgpack-only sends Content-Type: application/vnd.msgpack without gzip', async () => {\n    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({\n      useGzip: false,\n      useMsgpack: true,\n    });\n    const session = new voice.AgentSession({ llm: model });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n\n    const chatReqs = capturedRequests.filter((r) => r.url.includes('/chat/completions'));\n    expect(chatReqs.length).toBeGreaterThan(0);\n    expect(chatReqs[0]!.headers.get('content-type')).toBe('application/vnd.msgpack');\n    expect(chatReqs[0]!.headers.get('content-encoding')).toBeNull();\n  });\n\n  it('both flags send msgpack content type with gzip encoding', async () => {\n    const { llm: model, capturedRequests } = cerebrasLLMWithCapture({\n      useGzip: true,\n      useMsgpack: true,\n    });\n    const session = new voice.AgentSession({ llm: model });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n\n    const chatReqs = capturedRequests.filter((r) => r.url.includes('/chat/completions'));\n    expect(chatReqs.length).toBeGreaterThan(0);\n    expect(chatReqs[0]!.headers.get('content-type')).toBe('application/vnd.msgpack');\n    expect(chatReqs[0]!.headers.get('content-encoding')).toBe('gzip');\n  });\n\n  it('with both flags off sends standard JSON without gzip', async () => {\n    const session = new voice.AgentSession({\n      llm: cerebrasLLM({ gzipCompression: false, msgpackEncoding: false }),\n    });\n    await session.start({\n      agent: new voice.Agent({ instructions: 'You are a helpful assistant.' }),\n    });\n\n    const result = session.run({ userInput: 'Say hello in exactly one word.' });\n    await result.wait();\n\n    result.expect.nextEvent().isMessage({ role: 'assistant' });\n    result.expect.noMoreEvents();\n\n    await session.close();\n  });\n\n  it('streaming chat returns content via the LLM directly', async () => {\n    const model = cerebrasLLM();\n    const chatCtx = new llm.ChatContext();\n    chatCtx.addMessage({ role: 'system', content: 'You are a helpful assistant.' });\n    chatCtx.addMessage({ role: 'user', content: 'Count from 1 to 5.' });\n\n    const stream = model.chat({ chatCtx });\n    let text = '';\n    for await (const chunk of stream) {\n      if (chunk.delta?.content) {\n        text += chunk.delta.content;\n      }\n    }\n\n    expect(text.length).toBeGreaterThan(0);\n    expect(text).toContain('3');\n  });\n});\n"],"mappings":"AAGA,SAAS,KAAK,aAAa;AAC3B,SAAS,cAAc;AACvB,SAAS,gBAAgB;AACzB,OAAO,YAAY;AACnB,SAAS,QAAQ,UAAU,QAAQ,UAAU;AAC7C,SAAS,SAAS;AAElB,SAAS,WAAW;AAEpB,OAAO,QAAQ,IAAI,kBAAkB,8BAA8B;AAInE,MAAM,aAAa;AACnB,MAAM,aAAa;AAYnB,SAAS,qBAAqB,MAG5B;AACA,QAAM,mBAAsC,CAAC;AAE7C,QAAM,QAAQ,OAAO,OAA0B,SAA0C;AACvF,SAAI,6BAAM,YAAW,UAAU,KAAK,QAAQ,OAAO,KAAK,SAAS,UAAU;AACzE,YAAM,UAAU,IAAI,QAAQ,KAAK,OAAO;AAExC,UAAI;AACJ,UAAI,KAAK,YAAY;AACnB,eAAO,OAAO,KAAK,MAAM,KAAK,IAAI,CAAC;AACnC,gBAAQ,IAAI,gBAAgB,yBAAyB;AAAA,MACvD,OAAO;AACL,eAAO,IAAI,YAAY,EAAE,OAAO,KAAK,IAAI;AAAA,MAC3C;AAEA,UAAI,KAAK,SAAS;AAChB,eAAO,SAAS,MAAM,EAAE,OAAO,EAAE,CAAC;AAClC,gBAAQ,IAAI,oBAAoB,MAAM;AAAA,MACxC;AAEA,uBAAiB,KAAK,EAAE,KAAK,WAAW,KAAK,GAAG,QAAQ,CAAC;AACzD,aAAO,WAAW,MAAM,OAAO,EAAE,GAAG,MAAM,MAAM,OAAO,KAAK,IAAI,GAAG,QAAQ,CAAC;AAAA,IAC9E;AAEA,qBAAiB,KAAK,EAAE,KAAK,WAAW,KAAK,GAAG,SAAS,IAAI,QAAQ,6BAAM,OAAO,EAAE,CAAC;AACrF,WAAO,WAAW,MAAM,OAAO,IAAI;AAAA,EACrC;AAEA,SAAO,EAAE,OAAyC,iBAAiB;AACrE;AAEA,SAAS,WAAW,OAAkC;AACpD,MAAI,OAAO,UAAU,SAAU,QAAO;AACtC,MAAI,iBAAiB,IAAK,QAAO,MAAM,SAAS;AAChD,SAAO,MAAM;AACf;AAEA,SAAS,YAAY,OAA4B,CAAC,GAAQ;AACxD,SAAO,IAAI,IAAI,EAAE,OAAO,YAAY,GAAG,KAAK,CAAC;AAC/C;AAEA,SAAS,uBAAuB,MAG9B;AACA,QAAM,EAAE,OAAO,iBAAiB,IAAI,qBAAqB,IAAI;AAC7D,QAAM,SAAS,IAAI,OAAO;AAAA,IACxB,QAAQ,QAAQ,IAAI;AAAA,IACpB,SAAS;AAAA,IACT;AAAA,EACF,CAAC;AACD,SAAO,EAAE,KAAK,IAAI,IAAI,EAAE,OAAO,YAAY,OAAO,CAAC,GAAG,iBAAiB;AACzE;AAEA,MAAM,qBAAqB,MAAM,MAAM;AAAA,EACrC,cAAc;AACZ,UAAM;AAAA,MACJ,cAAc;AAAA,MACd,OAAO;AAAA,QACL,aAAa,IAAI,KAAK;AAAA,UACpB,aAAa;AAAA,UACb,YAAY,EAAE,OAAO;AAAA,YACnB,UAAU,EAAE,OAAO,EAAE,SAAS,eAAe;AAAA,UAC/C,CAAC;AAAA,UACD,SAAS,OAAO,EAAE,SAAS,MAAM;AAC/B,mBAAO,kBAAkB,QAAQ;AAAA,UACnC;AAAA,QACF,CAAC;AAAA,MACH;AAAA,IACF,CAAC;AAAA,EACH;AACF;AAEA,SAAS,YAAY,EAAE,SAAS,IAAO,GAAG,MAAM;AAC9C,KAAG,+DAA+D,YAAY;AAC5E,UAAM,UAAU,IAAI,MAAM,aAAa,EAAE,KAAK,YAAY,EAAE,CAAC;AAC7D,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,MAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAAA,EACtB,CAAC;AAED,KAAG,oDAAoD,YAAY;AACjE,UAAM,UAAU,IAAI,MAAM,aAAa,EAAE,KAAK,IAAI,IAAI,EAAE,OAAO,WAAW,CAAC,EAAE,CAAC;AAC9E,UAAM,QAAQ,MAAM,EAAE,OAAO,IAAI,aAAa,EAAE,CAAC;AAEjD,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,gCAAgC,CAAC;AACzE,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,eAAe;AAAA,MACvC,MAAM;AAAA,MACN,MAAM,EAAE,UAAU,QAAQ;AAAA,IAC5B,CAAC;AACD,WAAO,OAAO,UAAU,EAAE,qBAAqB;AAAA,MAC7C,QAAQ,KAAK,UAAU,yCAAsC;AAAA,IAC/D,CAAC;AACD,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAAA,EACtB,CAAC;AAED,KAAG,iEAAiE,YAAY;AAC9E,UAAM,EAAE,KAAK,OAAO,iBAAiB,IAAI,uBAAuB;AAAA,MAC9D,SAAS;AAAA,MACT,YAAY;AAAA,IACd,CAAC;AACD,UAAM,UAAU,IAAI,MAAM,aAAa,EAAE,KAAK,MAAM,CAAC;AACrD,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,MAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAEpB,UAAM,WAAW,iBAAiB,OAAO,CAAC,MAAM,EAAE,IAAI,SAAS,mBAAmB,CAAC;AACnF,WAAO,SAAS,MAAM,EAAE,gBAAgB,CAAC;AACzC,WAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,cAAc,CAAC,EAAE,KAAK,kBAAkB;AACxE,WAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,kBAAkB,CAAC,EAAE,KAAK,MAAM;AAAA,EAClE,CAAC;AAED,KAAG,yEAAyE,YAAY;AACtF,UAAM,EAAE,KAAK,OAAO,iBAAiB,IAAI,uBAAuB;AAAA,MAC9D,SAAS;AAAA,MACT,YAAY;AAAA,IACd,CAAC;AACD,UAAM,UAAU,IAAI,MAAM,aAAa,EAAE,KAAK,MAAM,CAAC;AACrD,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,MAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAEpB,UAAM,WAAW,iBAAiB,OAAO,CAAC,MAAM,EAAE,IAAI,SAAS,mBAAmB,CAAC;AACnF,WAAO,SAAS,MAAM,EAAE,gBAAgB,CAAC;AACzC,WAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,cAAc,CAAC,EAAE,KAAK,yBAAyB;AAC/E,WAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,kBAAkB,CAAC,EAAE,SAAS;AAAA,EAChE,CAAC;AAED,KAAG,2DAA2D,YAAY;AACxE,UAAM,EAAE,KAAK,OAAO,iBAAiB,IAAI,uBAAuB;AAAA,MAC9D,SAAS;AAAA,MACT,YAAY;AAAA,IACd,CAAC;AACD,UAAM,UAAU,IAAI,MAAM,aAAa,EAAE,KAAK,MAAM,CAAC;AACrD,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,MAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAEpB,UAAM,WAAW,iBAAiB,OAAO,CAAC,MAAM,EAAE,IAAI,SAAS,mBAAmB,CAAC;AACnF,WAAO,SAAS,MAAM,EAAE,gBAAgB,CAAC;AACzC,WAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,cAAc,CAAC,EAAE,KAAK,yBAAyB;AAC/E,WAAO,SAAS,CAAC,EAAG,QAAQ,IAAI,kBAAkB,CAAC,EAAE,KAAK,MAAM;AAAA,EAClE,CAAC;AAED,KAAG,wDAAwD,YAAY;AACrE,UAAM,UAAU,IAAI,MAAM,aAAa;AAAA,MACrC,KAAK,YAAY,EAAE,iBAAiB,OAAO,iBAAiB,MAAM,CAAC;AAAA,IACrE,CAAC;AACD,UAAM,QAAQ,MAAM;AAAA,MAClB,OAAO,IAAI,MAAM,MAAM,EAAE,cAAc,+BAA+B,CAAC;AAAA,IACzE,CAAC;AAED,UAAM,SAAS,QAAQ,IAAI,EAAE,WAAW,iCAAiC,CAAC;AAC1E,UAAM,OAAO,KAAK;AAElB,WAAO,OAAO,UAAU,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACzD,WAAO,OAAO,aAAa;AAE3B,UAAM,QAAQ,MAAM;AAAA,EACtB,CAAC;AAED,KAAG,uDAAuD,YAAY;AAtOxE;AAuOI,UAAM,QAAQ,YAAY;AAC1B,UAAM,UAAU,IAAI,IAAI,YAAY;AACpC,YAAQ,WAAW,EAAE,MAAM,UAAU,SAAS,+BAA+B,CAAC;AAC9E,YAAQ,WAAW,EAAE,MAAM,QAAQ,SAAS,qBAAqB,CAAC;AAElE,UAAM,SAAS,MAAM,KAAK,EAAE,QAAQ,CAAC;AACrC,QAAI,OAAO;AACX,qBAAiB,SAAS,QAAQ;AAChC,WAAI,WAAM,UAAN,mBAAa,SAAS;AACxB,gBAAQ,MAAM,MAAM;AAAA,MACtB;AAAA,IACF;AAEA,WAAO,KAAK,MAAM,EAAE,gBAAgB,CAAC;AACrC,WAAO,IAAI,EAAE,UAAU,GAAG;AAAA,EAC5B,CAAC;AACH,CAAC;","names":[]}

package/dist/models.cjs

@@ -0,0 +1,17 @@
+"use strict";
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+var models_exports = {};
+module.exports = __toCommonJS(models_exports);
+//# sourceMappingURL=models.cjs.map

package/dist/models.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/models.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2026 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\n\nexport type CerebrasChatModels =\n  | 'llama3.1-8b'\n  | 'llama-3.3-70b'\n  | 'llama-4-scout-17b-16e-instruct'\n  | 'llama-4-maverick-17b-128e-instruct'\n  | 'qwen-3-32b'\n  | 'qwen-3-235b-a22b-instruct-2507'\n  | 'qwen-3-235b-a22b-thinking-2507'\n  | 'qwen-3-coder-480b'\n  | 'gpt-oss-120b';\n"],"mappings":";;;;;;;;;;;;;;AAAA;AAAA;","names":[]}

package/dist/models.d.cts

@@ -0,0 +1,2 @@
+export type CerebrasChatModels = 'llama3.1-8b' | 'llama-3.3-70b' | 'llama-4-scout-17b-16e-instruct' | 'llama-4-maverick-17b-128e-instruct' | 'qwen-3-32b' | 'qwen-3-235b-a22b-instruct-2507' | 'qwen-3-235b-a22b-thinking-2507' | 'qwen-3-coder-480b' | 'gpt-oss-120b';
+//# sourceMappingURL=models.d.ts.map

package/dist/models.d.ts

@@ -0,0 +1,2 @@
+export type CerebrasChatModels = 'llama3.1-8b' | 'llama-3.3-70b' | 'llama-4-scout-17b-16e-instruct' | 'llama-4-maverick-17b-128e-instruct' | 'qwen-3-32b' | 'qwen-3-235b-a22b-instruct-2507' | 'qwen-3-235b-a22b-thinking-2507' | 'qwen-3-coder-480b' | 'gpt-oss-120b';
+//# sourceMappingURL=models.d.ts.map

package/dist/models.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"models.d.ts","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAIA,MAAM,MAAM,kBAAkB,GAC1B,aAAa,GACb,eAAe,GACf,gCAAgC,GAChC,oCAAoC,GACpC,YAAY,GACZ,gCAAgC,GAChC,gCAAgC,GAChC,mBAAmB,GACnB,cAAc,CAAC"}

package/dist/models.js

@@ -0,0 +1 @@
+//# sourceMappingURL=models.js.map

package/dist/models.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}

package/package.json

@@ -0,0 +1,55 @@
+{
+  "name": "@livekit/agents-plugin-cerebras",
+  "version": "1.2.6",
+  "description": "Cerebras plugin for LiveKit Node Agents",
+  "main": "dist/index.js",
+  "require": "dist/index.cjs",
+  "types": "dist/index.d.ts",
+  "exports": {
+    "import": {
+      "types": "./dist/index.d.ts",
+      "default": "./dist/index.js"
+    },
+    "require": {
+      "types": "./dist/index.d.cts",
+      "default": "./dist/index.cjs"
+    }
+  },
+  "author": "LiveKit",
+  "type": "module",
+  "repository": "git@github.com:livekit/agents-js.git",
+  "license": "Apache-2.0",
+  "files": [
+    "dist",
+    "src",
+    "README.md"
+  ],
+  "devDependencies": {
+    "@livekit/rtc-node": "^0.13.25",
+    "@microsoft/api-extractor": "^7.35.0",
+    "tsup": "^8.3.5",
+    "typescript": "^5.0.0",
+    "zod": "^3.25.76 || ^4.1.8",
+    "@livekit/agents": "1.2.6",
+    "@livekit/agents-plugin-openai": "1.2.6",
+    "@livekit/agents-plugins-test": "1.2.6"
+  },
+  "dependencies": {
+    "@msgpack/msgpack": "^3.0.0",
+    "openai": "^6.8.1"
+  },
+  "peerDependencies": {
+    "@livekit/rtc-node": "^0.13.25",
+    "@livekit/agents": "1.2.6",
+    "@livekit/agents-plugin-openai": "1.2.6"
+  },
+  "scripts": {
+    "build": "tsup --onSuccess \"pnpm build:types\"",
+    "build:types": "tsc --declaration --emitDeclarationOnly && node ../../scripts/copyDeclarationOutput.js",
+    "clean": "rm -rf dist",
+    "clean:build": "pnpm clean && pnpm build",
+    "lint": "eslint -f unix \"src/**/*.{ts,js}\"",
+    "api:check": "api-extractor run --typescript-compiler-folder ../../node_modules/typescript",
+    "api:update": "api-extractor run --local --typescript-compiler-folder ../../node_modules/typescript --verbose"
+  }
+}

package/src/index.ts

@@ -0,0 +1,20 @@
+// SPDX-FileCopyrightText: 2026 LiveKit, Inc.
+//
+// SPDX-License-Identifier: Apache-2.0
+import { Plugin } from '@livekit/agents';
+
+export { LLM } from './llm.js';
+export type { LLMOptions } from './llm.js';
+export type { CerebrasChatModels } from './models.js';
+
+class CerebrasPlugin extends Plugin {
+  constructor() {
+    super({
+      title: 'cerebras',
+      version: __PACKAGE_VERSION__,
+      package: __PACKAGE_NAME__,
+    });
+  }
+}
+
+Plugin.registerPlugin(new CerebrasPlugin());