diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 3f105fa..0112244 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -30,8 +30,6 @@ jobs:
   e2e:
     name: E2E Tests
     runs-on: ubuntu-latest
-    # Only run e2e on main branch (has secrets)
-    if: github.ref == 'refs/heads/main' || github.event_name == 'push'
     steps:
       - uses: actions/checkout@v4
       
@@ -47,6 +45,8 @@ jobs:
         run: npm run build
       
       - name: Run e2e tests
+        # Tests requiring secrets (bot.e2e, models.e2e) skip gracefully via describe.skipIf.
+        # OpenAI SDK compat tests always run (no secrets needed, uses mock gateway).
         run: npm run test:e2e
         env:
           LETTA_API_KEY: ${{ secrets.LETTA_API_KEY }}
diff --git a/e2e/openai-compat.e2e.test.ts b/e2e/openai-compat.e2e.test.ts
new file mode 100644
index 0000000..ed59984
--- /dev/null
+++ b/e2e/openai-compat.e2e.test.ts
@@ -0,0 +1,282 @@
+/**
+ * E2E Tests for OpenAI-compatible API endpoint
+ *
+ * Uses the real `openai` npm SDK as the client to prove full compatibility.
+ * No Letta API secrets needed -- uses a mock AgentRouter so this runs in CI.
+ *
+ * Run with: npm run test:e2e
+ */
+
+import { describe, it, expect, beforeAll, afterAll, vi } from 'vitest';
+import * as http from 'http';
+import OpenAI from 'openai';
+import { createApiServer } from '../src/api/server.js';
+import type { AgentRouter } from '../src/core/interfaces.js';
+
+const TEST_API_KEY = 'e2e-test-key-openai-compat';
+
+function createMockRouter(overrides: Partial<AgentRouter> = {}): AgentRouter {
+  return {
+    deliverToChannel: vi.fn().mockResolvedValue('msg-1'),
+    sendToAgent: vi.fn().mockResolvedValue('Hello from lettabot! I can help you with that.'),
+    streamToAgent: vi.fn().mockReturnValue((async function* () {
+      yield { type: 'reasoning', content: 'Let me think about this...' };
+      yield { type: 'assistant', content: 'Hello' };
+      yield { type: 'assistant', content: ' from' };
+      yield { type: 'assistant', content: ' lettabot!' };
+      yield { type: 'tool_call', toolCallId: 'call_abc123', toolName: 'web_search', toolInput: { query: 'lettabot docs' } };
+      yield { type: 'tool_result', content: 'Search results...' };
+      yield { type: 'assistant', content: ' I found' };
+      yield { type: 'assistant', content: ' the answer.' };
+      yield { type: 'result', success: true };
+    })()),
+    getAgentNames: vi.fn().mockReturnValue(['lettabot', 'helper-bot']),
+    ...overrides,
+  };
+}
+
+function getPort(server: http.Server): number {
+  const addr = server.address();
+  if (typeof addr === 'object' && addr) return addr.port;
+  throw new Error('Server not listening');
+}
+
+describe('e2e: OpenAI SDK compatibility', () => {
+  let server: http.Server;
+  let port: number;
+  let router: AgentRouter;
+  let client: OpenAI;
+
+  beforeAll(async () => {
+    router = createMockRouter();
+    server = createApiServer(router, {
+      port: 0, // OS-assigned port
+      apiKey: TEST_API_KEY,
+      host: '127.0.0.1',
+    });
+    await new Promise<void>((resolve) => {
+      if (server.listening) { resolve(); return; }
+      server.once('listening', resolve);
+    });
+    port = getPort(server);
+
+    // Create OpenAI SDK client pointing at our server
+    client = new OpenAI({
+      apiKey: TEST_API_KEY,
+      baseURL: `http://127.0.0.1:${port}/v1`,
+    });
+  });
+
+  afterAll(async () => {
+    await new Promise<void>((resolve) => server.close(() => resolve()));
+  });
+
+  // ---------------------------------------------------------------------------
+  // Models
+  // ---------------------------------------------------------------------------
+
+  it('lists models via OpenAI SDK', async () => {
+    const models = await client.models.list();
+
+    // The SDK returns a page object; iterate to get all
+    const modelList: OpenAI.Models.Model[] = [];
+    for await (const model of models) {
+      modelList.push(model);
+    }
+
+    expect(modelList).toHaveLength(2);
+    expect(modelList[0].id).toBe('lettabot');
+    expect(modelList[1].id).toBe('helper-bot');
+    expect(modelList[0].owned_by).toBe('lettabot');
+  });
+
+  // ---------------------------------------------------------------------------
+  // Non-streaming (sync)
+  // ---------------------------------------------------------------------------
+
+  it('sends a sync chat completion via OpenAI SDK', async () => {
+    const completion = await client.chat.completions.create({
+      model: 'lettabot',
+      messages: [{ role: 'user', content: 'Hello!' }],
+    });
+
+    // Validate the SDK parsed it correctly
+    expect(completion.id).toMatch(/^chatcmpl-/);
+    expect(completion.object).toBe('chat.completion');
+    expect(completion.model).toBe('lettabot');
+    expect(completion.choices).toHaveLength(1);
+    expect(completion.choices[0].message.role).toBe('assistant');
+    expect(completion.choices[0].message.content).toBe('Hello from lettabot! I can help you with that.');
+    expect(completion.choices[0].finish_reason).toBe('stop');
+
+    // Verify the router received the right call
+    expect(router.sendToAgent).toHaveBeenCalledWith(
+      'lettabot',
+      'Hello!',
+      expect.objectContaining({ type: 'webhook' }),
+    );
+  });
+
+  it('defaults to first model when model field is omitted', async () => {
+    // The OpenAI SDK requires model, but we can test with the first agent name
+    const completion = await client.chat.completions.create({
+      model: 'lettabot',
+      messages: [
+        { role: 'system', content: 'You are helpful.' },
+        { role: 'user', content: 'First question' },
+        { role: 'assistant', content: 'First answer' },
+        { role: 'user', content: 'Follow-up question' },
+      ],
+    });
+
+    // Should extract last user message
+    expect(router.sendToAgent).toHaveBeenCalledWith(
+      'lettabot',
+      'Follow-up question',
+      expect.any(Object),
+    );
+    expect(completion.choices[0].message.content).toBeTruthy();
+  });
+
+  it('throws on unknown model', async () => {
+    await expect(
+      client.chat.completions.create({
+        model: 'nonexistent-model',
+        messages: [{ role: 'user', content: 'Hi' }],
+      })
+    ).rejects.toThrow(); // SDK throws on 404
+  });
+
+  // ---------------------------------------------------------------------------
+  // Streaming
+  // ---------------------------------------------------------------------------
+
+  it('streams a chat completion via OpenAI SDK', async () => {
+    // Fresh mock for streaming (generators are consumed once)
+    (router as any).streamToAgent = vi.fn().mockReturnValue((async function* () {
+      yield { type: 'reasoning', content: 'thinking...' };
+      yield { type: 'assistant', content: 'Hello' };
+      yield { type: 'assistant', content: ' world' };
+      yield { type: 'result', success: true };
+    })());
+
+    const stream = await client.chat.completions.create({
+      model: 'lettabot',
+      messages: [{ role: 'user', content: 'Stream test' }],
+      stream: true,
+    });
+
+    const chunks: OpenAI.Chat.Completions.ChatCompletionChunk[] = [];
+    for await (const chunk of stream) {
+      chunks.push(chunk);
+    }
+
+    // Should have role announcement + content deltas + stop
+    expect(chunks.length).toBeGreaterThanOrEqual(3);
+
+    // First chunk should announce the role
+    expect(chunks[0].choices[0].delta.role).toBe('assistant');
+
+    // Collect all content
+    const content = chunks
+      .map(c => c.choices[0].delta.content)
+      .filter(Boolean)
+      .join('');
+    expect(content).toBe('Hello world');
+
+    // Last chunk should have finish_reason
+    const lastChunk = chunks[chunks.length - 1];
+    expect(lastChunk.choices[0].finish_reason).toBe('stop');
+
+    // All chunks should share the same ID
+    const ids = new Set(chunks.map(c => c.id));
+    expect(ids.size).toBe(1);
+    expect(chunks[0].id).toMatch(/^chatcmpl-/);
+  });
+
+  it('streams tool calls in OpenAI format', async () => {
+    (router as any).streamToAgent = vi.fn().mockReturnValue((async function* () {
+      yield { type: 'assistant', content: 'Let me search.' };
+      yield { type: 'tool_call', toolCallId: 'call_xyz', toolName: 'web_search', toolInput: { query: 'test' } };
+      yield { type: 'tool_result', content: 'results' };
+      yield { type: 'assistant', content: ' Found it!' };
+      yield { type: 'result', success: true };
+    })());
+
+    const stream = await client.chat.completions.create({
+      model: 'lettabot',
+      messages: [{ role: 'user', content: 'Search for something' }],
+      stream: true,
+    });
+
+    const chunks: OpenAI.Chat.Completions.ChatCompletionChunk[] = [];
+    for await (const chunk of stream) {
+      chunks.push(chunk);
+    }
+
+    // Find tool call chunks
+    const toolCallChunks = chunks.filter(c => c.choices[0].delta.tool_calls);
+    expect(toolCallChunks).toHaveLength(1);
+
+    const toolCall = toolCallChunks[0].choices[0].delta.tool_calls![0];
+    expect(toolCall.function?.name).toBe('web_search');
+    expect(toolCall.function?.arguments).toContain('test');
+    expect(toolCall.id).toBe('call_xyz');
+
+    // Content should not include reasoning or tool results
+    const content = chunks
+      .map(c => c.choices[0].delta.content)
+      .filter(Boolean)
+      .join('');
+    expect(content).toBe('Let me search. Found it!');
+    expect(content).not.toContain('thinking');
+    expect(content).not.toContain('results');
+  });
+
+  it('filters reasoning from streamed output', async () => {
+    (router as any).streamToAgent = vi.fn().mockReturnValue((async function* () {
+      yield { type: 'reasoning', content: 'Deep reasoning about the problem...' };
+      yield { type: 'reasoning', content: 'More thinking happening here...' };
+      yield { type: 'assistant', content: 'Here is my answer.' };
+      yield { type: 'result', success: true };
+    })());
+
+    const stream = await client.chat.completions.create({
+      model: 'lettabot',
+      messages: [{ role: 'user', content: 'Think hard' }],
+      stream: true,
+    });
+
+    const allContent: string[] = [];
+    for await (const chunk of stream) {
+      if (chunk.choices[0].delta.content) {
+        allContent.push(chunk.choices[0].delta.content);
+      }
+    }
+
+    const fullText = allContent.join('');
+    expect(fullText).toBe('Here is my answer.');
+    expect(fullText).not.toContain('Deep reasoning');
+    expect(fullText).not.toContain('More thinking');
+  });
+
+  // ---------------------------------------------------------------------------
+  // Auth
+  // ---------------------------------------------------------------------------
+
+  it('authenticates with Bearer token (OpenAI SDK default)', async () => {
+    // The OpenAI SDK sends Authorization: Bearer <key> by default
+    // If we got this far, auth is working. But let's also verify a wrong key fails.
+    const badClient = new OpenAI({
+      apiKey: 'wrong-key',
+      baseURL: `http://127.0.0.1:${port}/v1`,
+    });
+
+    await expect(
+      badClient.chat.completions.create({
+        model: 'lettabot',
+        messages: [{ role: 'user', content: 'Hi' }],
+      })
+    ).rejects.toThrow();
+  });
+});
diff --git a/src/api/auth.ts b/src/api/auth.ts
index 08279d7..4eedad7 100644
--- a/src/api/auth.ts
+++ b/src/api/auth.ts
@@ -99,12 +99,45 @@ export function saveApiKey(key: string): void {
 }
 
 /**
- * Validate API key from request headers
+ * Extract API key from request headers.
+ * Checks X-Api-Key first (lettabot convention), then Authorization: Bearer <key> (OpenAI convention).
+ * 
+ * Note: When using Authorization header, ensure CORS includes 'Authorization' in Access-Control-Allow-Headers.
+ * 
+ * @param headers - HTTP request headers
+ * @returns The extracted API key, or null if not found
+ */
+export function extractApiKey(headers: IncomingHttpHeaders): string | null {
+  // 1. X-Api-Key header (lettabot convention)
+  const xApiKey = headers['x-api-key'];
+  if (xApiKey && typeof xApiKey === 'string') {
+    return xApiKey;
+  }
+
+  // 2. Authorization: Bearer <key> (OpenAI convention)
+  const auth = headers['authorization'];
+  if (auth && typeof auth === 'string') {
+    const match = auth.match(/^Bearer\s+(.+)$/i);
+    if (match) {
+      return match[1];
+    }
+  }
+
+  return null;
+}
+
+/**
+ * Validate API key from request headers.
+ * Supports both X-Api-Key and Authorization: Bearer <key> formats.
+ * 
+ * @param headers - HTTP request headers
+ * @param expectedKey - The expected API key to validate against
+ * @returns true if the provided key matches the expected key, false otherwise
  */
 export function validateApiKey(headers: IncomingHttpHeaders, expectedKey: string): boolean {
-  const providedKey = headers['x-api-key'];
+  const providedKey = extractApiKey(headers);
 
-  if (!providedKey || typeof providedKey !== 'string') {
+  if (!providedKey) {
     return false;
   }
 
diff --git a/src/api/openai-compat.test.ts b/src/api/openai-compat.test.ts
new file mode 100644
index 0000000..655b3c4
--- /dev/null
+++ b/src/api/openai-compat.test.ts
@@ -0,0 +1,834 @@
+import { describe, it, expect, vi, beforeAll, afterAll } from 'vitest';
+import * as http from 'http';
+import { createApiServer } from './server.js';
+import type { AgentRouter } from '../core/interfaces.js';
+import {
+  generateCompletionId,
+  extractLastUserMessage,
+  buildCompletion,
+  buildChunk,
+  buildToolCallChunk,
+  formatSSE,
+  SSE_DONE,
+  buildErrorResponse,
+  buildModelList,
+  validateChatRequest,
+} from './openai-compat.js';
+import type { OpenAIChatMessage } from './openai-compat.js';
+
+const TEST_API_KEY = 'test-key-12345';
+const TEST_PORT = 0;
+
+function createMockRouter(overrides: Partial<AgentRouter> = {}): AgentRouter {
+  return {
+    deliverToChannel: vi.fn().mockResolvedValue('msg-1'),
+    sendToAgent: vi.fn().mockResolvedValue('Agent says hello'),
+    streamToAgent: vi.fn().mockReturnValue(
+      (async function* () {
+        yield { type: 'reasoning', content: 'thinking...' };
+        yield { type: 'assistant', content: 'Hello ' };
+        yield { type: 'assistant', content: 'world' };
+        yield { type: 'result', success: true };
+      })(),
+    ),
+    getAgentNames: vi.fn().mockReturnValue(['LettaBot']),
+    ...overrides,
+  };
+}
+
+function getPort(server: http.Server): number {
+  const addr = server.address();
+  if (typeof addr === 'object' && addr) return addr.port;
+  throw new Error('Server not listening');
+}
+
+async function request(
+  port: number,
+  method: string,
+  path: string,
+  body?: string,
+  headers: Record<string, string> = {},
+): Promise<{ status: number; headers: http.IncomingHttpHeaders; body: string }> {
+  return new Promise((resolve, reject) => {
+    const req = http.request(
+      { hostname: '127.0.0.1', port, method, path, headers },
+      (res) => {
+        let data = '';
+        res.on('data', (chunk) => {
+          data += chunk;
+        });
+        res.on('end', () =>
+          resolve({ status: res.statusCode!, headers: res.headers, body: data }),
+        );
+      },
+    );
+    req.on('error', reject);
+    if (body) req.write(body);
+    req.end();
+  });
+}
+
+// ============================================================================
+// UNIT TESTS FOR UTILITY FUNCTIONS
+// ============================================================================
+
+describe('openai-compat utilities', () => {
+  describe('generateCompletionId', () => {
+    it('returns a string starting with chatcmpl-', () => {
+      const id = generateCompletionId();
+      expect(id).toMatch(/^chatcmpl-/);
+    });
+
+    it('returns unique IDs', () => {
+      const id1 = generateCompletionId();
+      const id2 = generateCompletionId();
+      expect(id1).not.toBe(id2);
+    });
+
+    it('returns IDs with reasonable length', () => {
+      const id = generateCompletionId();
+      expect(id.length).toBeGreaterThan(10);
+    });
+  });
+
+  describe('extractLastUserMessage', () => {
+    it('returns the last user message content', () => {
+      const messages: OpenAIChatMessage[] = [
+        { role: 'user', content: 'First message' },
+        { role: 'assistant', content: 'Response' },
+        { role: 'user', content: 'Last message' },
+      ];
+      expect(extractLastUserMessage(messages)).toBe('Last message');
+    });
+
+    it('returns null when no user messages', () => {
+      const messages: OpenAIChatMessage[] = [
+        { role: 'system', content: 'System prompt' },
+        { role: 'assistant', content: 'Hello' },
+      ];
+      expect(extractLastUserMessage(messages)).toBeNull();
+    });
+
+    it('skips system and assistant messages', () => {
+      const messages: OpenAIChatMessage[] = [
+        { role: 'system', content: 'System' },
+        { role: 'assistant', content: 'Assistant' },
+        { role: 'user', content: 'User message' },
+        { role: 'assistant', content: 'Another assistant' },
+      ];
+      expect(extractLastUserMessage(messages)).toBe('User message');
+    });
+
+    it('returns null for empty array', () => {
+      expect(extractLastUserMessage([])).toBeNull();
+    });
+
+    it('handles only one user message', () => {
+      const messages: OpenAIChatMessage[] = [{ role: 'user', content: 'Only message' }];
+      expect(extractLastUserMessage(messages)).toBe('Only message');
+    });
+  });
+
+  describe('buildCompletion', () => {
+    it('builds a valid completion response', () => {
+      const result = buildCompletion('chatcmpl-1', 'LettaBot', 'Hello!');
+      expect(result.object).toBe('chat.completion');
+      expect(result.id).toBe('chatcmpl-1');
+      expect(result.model).toBe('LettaBot');
+      expect(result.choices).toHaveLength(1);
+      expect(result.choices[0].index).toBe(0);
+      expect(result.choices[0].message.role).toBe('assistant');
+      expect(result.choices[0].message.content).toBe('Hello!');
+      expect(result.choices[0].finish_reason).toBe('stop');
+      expect(result.usage).toBeNull();
+      expect(result.created).toBeGreaterThan(0);
+    });
+
+    it('respects custom finish_reason', () => {
+      const result = buildCompletion('chatcmpl-2', 'bot', 'text', 'tool_calls');
+      expect(result.choices[0].finish_reason).toBe('tool_calls');
+    });
+
+    it('sets created timestamp', () => {
+      const before = Math.floor(Date.now() / 1000);
+      const result = buildCompletion('chatcmpl-3', 'bot', 'text');
+      const after = Math.floor(Date.now() / 1000);
+      expect(result.created).toBeGreaterThanOrEqual(before);
+      expect(result.created).toBeLessThanOrEqual(after);
+    });
+  });
+
+  describe('buildChunk', () => {
+    it('builds a valid streaming chunk', () => {
+      const chunk = buildChunk('chatcmpl-1', 'LettaBot', { content: 'Hello' });
+      expect(chunk.object).toBe('chat.completion.chunk');
+      expect(chunk.id).toBe('chatcmpl-1');
+      expect(chunk.model).toBe('LettaBot');
+      expect(chunk.choices).toHaveLength(1);
+      expect(chunk.choices[0].index).toBe(0);
+      expect(chunk.choices[0].delta).toEqual({ content: 'Hello' });
+      expect(chunk.choices[0].finish_reason).toBeNull();
+      expect(chunk.created).toBeGreaterThan(0);
+    });
+
+    it('includes finish_reason when provided', () => {
+      const chunk = buildChunk('chatcmpl-2', 'bot', { content: 'Done' }, 'stop');
+      expect(chunk.choices[0].finish_reason).toBe('stop');
+    });
+
+    it('handles role delta', () => {
+      const chunk = buildChunk('chatcmpl-3', 'bot', { role: 'assistant' });
+      expect(chunk.choices[0].delta).toEqual({ role: 'assistant' });
+    });
+  });
+
+  describe('buildToolCallChunk', () => {
+    it('builds a tool call chunk with correct structure', () => {
+      const chunk = buildToolCallChunk(
+        'chatcmpl-1',
+        'LettaBot',
+        0,
+        'call_123',
+        'web_search',
+        '{"query":"test"}',
+      );
+      expect(chunk.object).toBe('chat.completion.chunk');
+      expect(chunk.choices[0].delta.tool_calls).toHaveLength(1);
+      expect(chunk.choices[0].delta.tool_calls![0]).toEqual({
+        index: 0,
+        id: 'call_123',
+        type: 'function',
+        function: {
+          name: 'web_search',
+          arguments: '{"query":"test"}',
+        },
+      });
+    });
+
+    it('handles different tool indices', () => {
+      const chunk = buildToolCallChunk(
+        'chatcmpl-2',
+        'bot',
+        2,
+        'call_456',
+        'calculator',
+        '{}',
+      );
+      expect(chunk.choices[0].delta.tool_calls![0].index).toBe(2);
+    });
+  });
+
+  describe('formatSSE', () => {
+    it('formats data as SSE line', () => {
+      expect(formatSSE({ test: 1 })).toBe('data: {"test":1}\n\n');
+    });
+
+    it('handles complex objects', () => {
+      const data = { nested: { value: 'test' }, array: [1, 2, 3] };
+      const result = formatSSE(data);
+      expect(result).toMatch(/^data: /);
+      expect(result).toMatch(/\n\n$/);
+      expect(JSON.parse(result.replace('data: ', '').trim())).toEqual(data);
+    });
+
+    it('SSE_DONE constant is correct', () => {
+      expect(SSE_DONE).toBe('data: [DONE]\n\n');
+    });
+  });
+
+  describe('buildErrorResponse', () => {
+    it('builds error with default values', () => {
+      const result = buildErrorResponse('Something went wrong');
+      expect(result.status).toBe(400);
+      expect(result.body.error.message).toBe('Something went wrong');
+      expect(result.body.error.type).toBe('invalid_request_error');
+    });
+
+    it('respects custom status and type', () => {
+      const result = buildErrorResponse('Not found', 'model_not_found', 404);
+      expect(result.status).toBe(404);
+      expect(result.body.error.type).toBe('model_not_found');
+      expect(result.body.error.message).toBe('Not found');
+    });
+
+    it('includes null code field', () => {
+      const result = buildErrorResponse('Test');
+      expect(result.body.error.code).toBeNull();
+    });
+  });
+
+  describe('buildModelList', () => {
+    it('builds model list from agent names', () => {
+      const list = buildModelList(['bot1', 'bot2']);
+      expect(list.object).toBe('list');
+      expect(list.data).toHaveLength(2);
+      expect(list.data[0].id).toBe('bot1');
+      expect(list.data[0].object).toBe('model');
+      expect(list.data[0].owned_by).toBe('lettabot');
+      expect(list.data[1].id).toBe('bot2');
+    });
+
+    it('handles empty agent list', () => {
+      const list = buildModelList([]);
+      expect(list.object).toBe('list');
+      expect(list.data).toHaveLength(0);
+    });
+
+    it('sets created timestamps', () => {
+      const list = buildModelList(['bot1']);
+      expect(list.data[0].created).toBeGreaterThan(0);
+    });
+  });
+
+  describe('validateChatRequest', () => {
+    it('returns null for valid request', () => {
+      expect(
+        validateChatRequest({ messages: [{ role: 'user', content: 'hi' }] }),
+      ).toBeNull();
+    });
+
+    it('returns error for missing messages', () => {
+      const err = validateChatRequest({});
+      expect(err).not.toBeNull();
+      expect(err!.status).toBe(400);
+      expect(err!.body.error.message).toContain('messages');
+    });
+
+    it('returns error for empty messages', () => {
+      const err = validateChatRequest({ messages: [] });
+      expect(err).not.toBeNull();
+      expect(err!.status).toBe(400);
+      expect(err!.body.error.message).toContain('non-empty array');
+    });
+
+    it('returns error for non-object body', () => {
+      const err1 = validateChatRequest(null);
+      expect(err1).not.toBeNull();
+      expect(err1!.status).toBe(400);
+
+      const err2 = validateChatRequest('string');
+      expect(err2).not.toBeNull();
+      expect(err2!.status).toBe(400);
+    });
+
+    it('returns error for non-array messages', () => {
+      const err = validateChatRequest({ messages: 'not an array' });
+      expect(err).not.toBeNull();
+      expect(err!.status).toBe(400);
+    });
+
+    it('accepts request with multiple messages', () => {
+      const result = validateChatRequest({
+        messages: [
+          { role: 'system', content: 'You are helpful' },
+          { role: 'user', content: 'Hello' },
+          { role: 'assistant', content: 'Hi!' },
+          { role: 'user', content: 'How are you?' },
+        ],
+      });
+      expect(result).toBeNull();
+    });
+  });
+});
+
+// ============================================================================
+// SERVER ROUTE TESTS: GET /v1/models
+// ============================================================================
+
+describe('GET /v1/models', () => {
+  let server: http.Server;
+  let port: number;
+  let router: AgentRouter;
+
+  beforeAll(async () => {
+    router = createMockRouter();
+    server = createApiServer(router, {
+      port: TEST_PORT,
+      apiKey: TEST_API_KEY,
+      host: '127.0.0.1',
+    });
+    await new Promise<void>((resolve) => {
+      if (server.listening) {
+        resolve();
+        return;
+      }
+      server.once('listening', resolve);
+    });
+    port = getPort(server);
+  });
+
+  afterAll(async () => {
+    await new Promise<void>((resolve) => server.close(() => resolve()));
+  });
+
+  it('returns 401 without auth', async () => {
+    const res = await request(port, 'GET', '/v1/models');
+    expect(res.status).toBe(401);
+    const parsed = JSON.parse(res.body);
+    expect(parsed.error.type).toBe('invalid_request_error');
+  });
+
+  it('accepts Bearer token auth', async () => {
+    const res = await request(port, 'GET', '/v1/models', undefined, {
+      authorization: `Bearer ${TEST_API_KEY}`,
+    });
+    expect(res.status).toBe(200);
+    const parsed = JSON.parse(res.body);
+    expect(parsed.object).toBe('list');
+    expect(parsed.data).toHaveLength(1);
+    expect(parsed.data[0].id).toBe('LettaBot');
+  });
+
+  it('accepts X-Api-Key header', async () => {
+    const res = await request(port, 'GET', '/v1/models', undefined, {
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(200);
+    const parsed = JSON.parse(res.body);
+    expect(parsed.object).toBe('list');
+  });
+
+  it('returns model list with correct structure', async () => {
+    const res = await request(port, 'GET', '/v1/models', undefined, {
+      authorization: `Bearer ${TEST_API_KEY}`,
+    });
+    expect(res.status).toBe(200);
+    const parsed = JSON.parse(res.body);
+    expect(parsed.object).toBe('list');
+    expect(parsed.data).toHaveLength(1);
+    expect(parsed.data[0]).toMatchObject({
+      id: 'LettaBot',
+      object: 'model',
+      owned_by: 'lettabot',
+    });
+    expect(parsed.data[0].created).toBeGreaterThan(0);
+  });
+
+  it('returns 401 with wrong API key', async () => {
+    const res = await request(port, 'GET', '/v1/models', undefined, {
+      authorization: 'Bearer wrong-key',
+    });
+    expect(res.status).toBe(401);
+  });
+});
+
+// ============================================================================
+// SERVER ROUTE TESTS: POST /v1/chat/completions
+// ============================================================================
+
+describe('POST /v1/chat/completions', () => {
+  let server: http.Server;
+  let port: number;
+  let router: AgentRouter;
+
+  beforeAll(async () => {
+    router = createMockRouter();
+    server = createApiServer(router, {
+      port: TEST_PORT,
+      apiKey: TEST_API_KEY,
+      host: '127.0.0.1',
+    });
+    await new Promise<void>((resolve) => {
+      if (server.listening) {
+        resolve();
+        return;
+      }
+      server.once('listening', resolve);
+    });
+    port = getPort(server);
+  });
+
+  afterAll(async () => {
+    await new Promise<void>((resolve) => server.close(() => resolve()));
+  });
+
+  it('returns 401 without auth', async () => {
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Hello' }],
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+    });
+    expect(res.status).toBe(401);
+    const parsed = JSON.parse(res.body);
+    expect(parsed.error.type).toBe('invalid_request_error');
+  });
+
+  it('accepts Bearer auth', async () => {
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Hello' }],
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      authorization: `Bearer ${TEST_API_KEY}`,
+    });
+    expect(res.status).toBe(200);
+  });
+
+  it('accepts X-Api-Key header', async () => {
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Hello' }],
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(200);
+  });
+
+  it('returns sync completion by default', async () => {
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Hello' }],
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(200);
+    const parsed = JSON.parse(res.body);
+    expect(parsed.object).toBe('chat.completion');
+    expect(parsed.id).toMatch(/^chatcmpl-/);
+    expect(parsed.model).toBe('LettaBot');
+    expect(parsed.choices).toHaveLength(1);
+    expect(parsed.choices[0].index).toBe(0);
+    expect(parsed.choices[0].message.role).toBe('assistant');
+    expect(parsed.choices[0].message.content).toBe('Agent says hello');
+    expect(parsed.choices[0].finish_reason).toBe('stop');
+    expect(parsed.usage).toBeNull();
+    expect(parsed.created).toBeGreaterThan(0);
+  });
+
+  it('returns 404 for unknown model', async () => {
+    const body = JSON.stringify({
+      model: 'UnknownBot',
+      messages: [{ role: 'user', content: 'hi' }],
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(404);
+    const parsed = JSON.parse(res.body);
+    expect(parsed.error.type).toBe('model_not_found');
+    expect(parsed.error.message).toContain('UnknownBot');
+  });
+
+  it('returns 400 for missing messages', async () => {
+    const body = JSON.stringify({ model: 'LettaBot' });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(400);
+    const parsed = JSON.parse(res.body);
+    expect(parsed.error.type).toBe('invalid_request_error');
+    expect(parsed.error.message).toContain('messages');
+  });
+
+  it('returns 400 for empty messages array', async () => {
+    const body = JSON.stringify({ model: 'LettaBot', messages: [] });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(400);
+  });
+
+  it('returns 400 for invalid JSON', async () => {
+    const res = await request(
+      port,
+      'POST',
+      '/v1/chat/completions',
+      'not valid json',
+      {
+        'content-type': 'application/json',
+        'x-api-key': TEST_API_KEY,
+      },
+    );
+    expect(res.status).toBe(400);
+  });
+
+  it('extracts last user message from messages array', async () => {
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [
+        { role: 'system', content: 'You are helpful' },
+        { role: 'user', content: 'First message' },
+        { role: 'assistant', content: 'I see' },
+        { role: 'user', content: 'Second message' },
+      ],
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(200);
+    expect(router.sendToAgent).toHaveBeenCalledWith(
+      'LettaBot',
+      'Second message',
+      expect.any(Object),
+    );
+  });
+
+  it('returns SSE stream when stream: true', async () => {
+    // Reset mock for streaming
+    (router as any).streamToAgent = vi.fn().mockReturnValue(
+      (async function* () {
+        yield { type: 'reasoning', content: 'thinking...' };
+        yield { type: 'assistant', content: 'Hello ' };
+        yield { type: 'assistant', content: 'world' };
+        yield {
+          type: 'tool_call',
+          toolCallId: 'call_1',
+          toolName: 'web_search',
+          toolInput: { query: 'test' },
+        };
+        yield { type: 'tool_result', content: 'result data' };
+        yield { type: 'assistant', content: '!' };
+        yield { type: 'result', success: true };
+      })(),
+    );
+
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Stream' }],
+      stream: true,
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(200);
+    expect(res.headers['content-type']).toBe('text/event-stream');
+    expect(res.headers['cache-control']).toBe('no-cache');
+    expect(res.headers['connection']).toBe('keep-alive');
+
+    // Parse SSE events
+    const events = res.body
+      .split('\n\n')
+      .filter((line) => line.startsWith('data: '))
+      .map((line) => line.replace('data: ', ''))
+      .filter((line) => line !== '[DONE]')
+      .map((line) => JSON.parse(line));
+
+    // Should have: role announcement, content chunks, tool_call, final chunk
+    expect(events.length).toBeGreaterThanOrEqual(5);
+
+    // First chunk: role
+    expect(events[0].object).toBe('chat.completion.chunk');
+    expect(events[0].choices[0].delta.role).toBe('assistant');
+    expect(events[0].choices[0].finish_reason).toBeNull();
+
+    // Content deltas (reasoning should be skipped, tool_result should be skipped)
+    const contentChunks = events.filter(
+      (e: any) => e.choices[0].delta.content !== undefined,
+    );
+    const contentParts = contentChunks.map((e: any) => e.choices[0].delta.content);
+    expect(contentParts).toContain('Hello ');
+    expect(contentParts).toContain('world');
+    expect(contentParts).toContain('!');
+    expect(contentParts).not.toContain('thinking...'); // reasoning filtered
+
+    // Tool call chunk
+    const toolChunks = events.filter((e: any) => e.choices[0].delta.tool_calls);
+    expect(toolChunks).toHaveLength(1);
+    expect(toolChunks[0].choices[0].delta.tool_calls[0].function.name).toBe(
+      'web_search',
+    );
+    expect(toolChunks[0].choices[0].delta.tool_calls[0].id).toBe('call_1');
+    expect(
+      JSON.parse(toolChunks[0].choices[0].delta.tool_calls[0].function.arguments),
+    ).toEqual({ query: 'test' });
+
+    // Final chunk has finish_reason
+    const lastEvent = events[events.length - 1];
+    expect(lastEvent.choices[0].finish_reason).toBe('stop');
+
+    // data: [DONE] should be present
+    expect(res.body).toContain('data: [DONE]');
+  });
+
+  it('handles stream with only assistant content', async () => {
+    (router as any).streamToAgent = vi.fn().mockReturnValue(
+      (async function* () {
+        yield { type: 'assistant', content: 'Simple ' };
+        yield { type: 'assistant', content: 'response' };
+        yield { type: 'result', success: true };
+      })(),
+    );
+
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Hi' }],
+      stream: true,
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+    expect(res.status).toBe(200);
+
+    const events = res.body
+      .split('\n\n')
+      .filter((line) => line.startsWith('data: '))
+      .map((line) => line.replace('data: ', ''))
+      .filter((line) => line !== '[DONE]')
+      .map((line) => JSON.parse(line));
+
+    // Role + 2 content chunks + final chunk
+    expect(events.length).toBe(4);
+    expect(events[0].choices[0].delta.role).toBe('assistant');
+    expect(events[1].choices[0].delta.content).toBe('Simple ');
+    expect(events[2].choices[0].delta.content).toBe('response');
+    expect(events[3].choices[0].finish_reason).toBe('stop');
+  });
+
+  it('calls streamToAgent with correct parameters', async () => {
+    (router as any).streamToAgent = vi.fn().mockReturnValue(
+      (async function* () {
+        yield { type: 'assistant', content: 'test' };
+        yield { type: 'result', success: true };
+      })(),
+    );
+
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Test message' }],
+      stream: true,
+    });
+    await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+
+    expect(router.streamToAgent).toHaveBeenCalledWith(
+      'LettaBot',
+      'Test message',
+      expect.any(Object),
+    );
+  });
+
+  it('filters out reasoning events in stream', async () => {
+    (router as any).streamToAgent = vi.fn().mockReturnValue(
+      (async function* () {
+        yield { type: 'reasoning', content: 'This should not appear' };
+        yield { type: 'reasoning', content: 'Neither should this' };
+        yield { type: 'assistant', content: 'But this should' };
+        yield { type: 'result', success: true };
+      })(),
+    );
+
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Test' }],
+      stream: true,
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+
+    const events = res.body
+      .split('\n\n')
+      .filter((line) => line.startsWith('data: '))
+      .map((line) => line.replace('data: ', ''))
+      .filter((line) => line !== '[DONE]')
+      .map((line) => JSON.parse(line));
+
+    // Should only have role, content, and final chunk (no reasoning)
+    const allContent = events
+      .map((e: any) => e.choices[0].delta.content)
+      .filter(Boolean)
+      .join('');
+    expect(allContent).not.toContain('This should not appear');
+    expect(allContent).not.toContain('Neither should this');
+    expect(allContent).toBe('But this should');
+  });
+
+  it('filters out tool_result events in stream', async () => {
+    (router as any).streamToAgent = vi.fn().mockReturnValue(
+      (async function* () {
+        yield {
+          type: 'tool_call',
+          toolCallId: 'call_1',
+          toolName: 'test',
+          toolInput: {},
+        };
+        yield { type: 'tool_result', content: 'This should be hidden' };
+        yield { type: 'assistant', content: 'Final answer' };
+        yield { type: 'result', success: true };
+      })(),
+    );
+
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Test' }],
+      stream: true,
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+
+    const events = res.body
+      .split('\n\n')
+      .filter((line) => line.startsWith('data: '))
+      .map((line) => line.replace('data: ', ''))
+      .filter((line) => line !== '[DONE]')
+      .map((line) => JSON.parse(line));
+
+    const allContent = events
+      .map((e: any) => e.choices[0].delta.content)
+      .filter(Boolean)
+      .join('');
+    expect(allContent).not.toContain('This should be hidden');
+    expect(allContent).toBe('Final answer');
+  });
+
+  it('handles multiple tool calls in stream', async () => {
+    (router as any).streamToAgent = vi.fn().mockReturnValue(
+      (async function* () {
+        yield {
+          type: 'tool_call',
+          toolCallId: 'call_1',
+          toolName: 'tool1',
+          toolInput: { arg: 1 },
+        };
+        yield {
+          type: 'tool_call',
+          toolCallId: 'call_2',
+          toolName: 'tool2',
+          toolInput: { arg: 2 },
+        };
+        yield { type: 'assistant', content: 'Done' };
+        yield { type: 'result', success: true };
+      })(),
+    );
+
+    const body = JSON.stringify({
+      model: 'LettaBot',
+      messages: [{ role: 'user', content: 'Test' }],
+      stream: true,
+    });
+    const res = await request(port, 'POST', '/v1/chat/completions', body, {
+      'content-type': 'application/json',
+      'x-api-key': TEST_API_KEY,
+    });
+
+    const events = res.body
+      .split('\n\n')
+      .filter((line) => line.startsWith('data: '))
+      .map((line) => line.replace('data: ', ''))
+      .filter((line) => line !== '[DONE]')
+      .map((line) => JSON.parse(line));
+
+    const toolChunks = events.filter((e: any) => e.choices[0].delta.tool_calls);
+    expect(toolChunks).toHaveLength(2);
+    expect(toolChunks[0].choices[0].delta.tool_calls[0].function.name).toBe('tool1');
+    expect(toolChunks[1].choices[0].delta.tool_calls[0].function.name).toBe('tool2');
+  });
+});
diff --git a/src/api/openai-compat.ts b/src/api/openai-compat.ts
new file mode 100644
index 0000000..9c06db1
--- /dev/null
+++ b/src/api/openai-compat.ts
@@ -0,0 +1,316 @@
+import { randomUUID } from 'crypto';
+
+// ============================================================================
+// Request types
+// ============================================================================
+
+/**
+ * OpenAI Chat Completions request body.
+ */
+export interface OpenAIChatRequest {
+  model: string;
+  messages: OpenAIChatMessage[];
+  stream?: boolean;
+  // We ignore other OpenAI params (temperature, max_tokens, tools, etc.)
+}
+
+/**
+ * A single message in the OpenAI messages array.
+ */
+export interface OpenAIChatMessage {
+  role: 'system' | 'user' | 'assistant' | 'tool';
+  content: string | null;
+  tool_calls?: OpenAIToolCall[];
+}
+
+// ============================================================================
+// Response types (non-streaming)
+// ============================================================================
+
+/**
+ * OpenAI Chat Completion response (non-streaming).
+ */
+export interface OpenAIChatCompletion {
+  id: string;
+  object: 'chat.completion';
+  created: number;
+  model: string;
+  choices: OpenAIChatChoice[];
+  usage: null;
+}
+
+/**
+ * A single choice in a non-streaming completion response.
+ */
+export interface OpenAIChatChoice {
+  index: number;
+  message: {
+    role: 'assistant';
+    content: string | null;
+    tool_calls?: OpenAIToolCall[];
+  };
+  finish_reason: 'stop' | 'tool_calls' | 'length' | null;
+}
+
+// ============================================================================
+// Response types (streaming)
+// ============================================================================
+
+/**
+ * OpenAI Chat Completion chunk (streaming).
+ */
+export interface OpenAIChatChunk {
+  id: string;
+  object: 'chat.completion.chunk';
+  created: number;
+  model: string;
+  choices: OpenAIChatChunkChoice[];
+}
+
+/**
+ * A single choice in a streaming chunk.
+ */
+export interface OpenAIChatChunkChoice {
+  index: number;
+  delta: {
+    role?: 'assistant';
+    content?: string | null;
+    tool_calls?: OpenAIToolCallDelta[];
+  };
+  finish_reason: 'stop' | 'tool_calls' | 'length' | null;
+}
+
+// ============================================================================
+// Tool call types
+// ============================================================================
+
+/**
+ * OpenAI tool call (non-streaming).
+ */
+export interface OpenAIToolCall {
+  id: string;
+  type: 'function';
+  function: {
+    name: string;
+    arguments: string;
+  };
+}
+
+/**
+ * OpenAI tool call delta (streaming).
+ */
+export interface OpenAIToolCallDelta {
+  index: number;
+  id?: string;
+  type?: 'function';
+  function?: {
+    name?: string;
+    arguments?: string;
+  };
+}
+
+// ============================================================================
+// Models endpoint
+// ============================================================================
+
+/**
+ * OpenAI models list response.
+ */
+export interface OpenAIModelList {
+  object: 'list';
+  data: OpenAIModel[];
+}
+
+/**
+ * A single model in the models list.
+ */
+export interface OpenAIModel {
+  id: string;
+  object: 'model';
+  created: number;
+  owned_by: string;
+}
+
+// ============================================================================
+// Error response
+// ============================================================================
+
+/**
+ * OpenAI error response.
+ */
+export interface OpenAIErrorResponse {
+  error: {
+    message: string;
+    type: string;
+    param: string | null;
+    code: string | null;
+  };
+}
+
+// ============================================================================
+// Helper functions
+// ============================================================================
+
+/**
+ * Generate a unique chat completion ID.
+ */
+export function generateCompletionId(): string {
+  return `chatcmpl-${randomUUID()}`;
+}
+
+/**
+ * Extract the last user message from an OpenAI messages array.
+ * Returns the content string, or null if none found.
+ */
+export function extractLastUserMessage(messages: OpenAIChatMessage[]): string | null {
+  for (let i = messages.length - 1; i >= 0; i--) {
+    if (messages[i].role === 'user' && messages[i].content) {
+      return messages[i].content as string;
+    }
+  }
+  return null;
+}
+
+/**
+ * Build a sync (non-streaming) completion response.
+ */
+export function buildCompletion(
+  id: string,
+  model: string,
+  content: string,
+  finishReason: 'stop' | 'tool_calls' = 'stop',
+): OpenAIChatCompletion {
+  return {
+    id,
+    object: 'chat.completion',
+    created: Math.floor(Date.now() / 1000),
+    model,
+    choices: [{
+      index: 0,
+      message: { role: 'assistant', content },
+      finish_reason: finishReason,
+    }],
+    usage: null,
+  };
+}
+
+/**
+ * Build a streaming chunk.
+ */
+export function buildChunk(
+  id: string,
+  model: string,
+  delta: OpenAIChatChunkChoice['delta'],
+  finishReason: 'stop' | 'tool_calls' | null = null,
+): OpenAIChatChunk {
+  return {
+    id,
+    object: 'chat.completion.chunk',
+    created: Math.floor(Date.now() / 1000),
+    model,
+    choices: [{
+      index: 0,
+      delta,
+      finish_reason: finishReason,
+    }],
+  };
+}
+
+/**
+ * Build a tool call streaming chunk.
+ */
+export function buildToolCallChunk(
+  id: string,
+  model: string,
+  toolIndex: number,
+  toolCallId: string,
+  functionName: string,
+  args: string,
+): OpenAIChatChunk {
+  return buildChunk(id, model, {
+    tool_calls: [{
+      index: toolIndex,
+      id: toolCallId,
+      type: 'function',
+      function: { name: functionName, arguments: args },
+    }],
+  });
+}
+
+/**
+ * Format an SSE data line. Returns "data: <json>\n\n".
+ */
+export function formatSSE(data: unknown): string {
+  return `data: ${JSON.stringify(data)}\n\n`;
+}
+
+/**
+ * The SSE terminator.
+ */
+export const SSE_DONE = 'data: [DONE]\n\n';
+
+/**
+ * Build an OpenAI-format error response.
+ */
+export function buildErrorResponse(
+  message: string,
+  type: string = 'invalid_request_error',
+  status: number = 400,
+): { status: number; body: OpenAIErrorResponse } {
+  return {
+    status,
+    body: {
+      error: {
+        message,
+        type,
+        param: null,
+        code: null,
+      },
+    },
+  };
+}
+
+/**
+ * Build the models list from agent names.
+ */
+export function buildModelList(agentNames: string[]): OpenAIModelList {
+  const now = Math.floor(Date.now() / 1000);
+  return {
+    object: 'list',
+    data: agentNames.map(name => ({
+      id: name,
+      object: 'model' as const,
+      created: now,
+      owned_by: 'lettabot',
+    })),
+  };
+}
+
+/**
+ * Validate an OpenAI chat completion request.
+ * Returns null if valid, or an error response object.
+ */
+export function validateChatRequest(body: unknown): { status: number; body: OpenAIErrorResponse } | null {
+  if (!body || typeof body !== 'object') {
+    return buildErrorResponse('Invalid request body', 'invalid_request_error', 400);
+  }
+
+  const req = body as Record<string, unknown>;
+
+  if (!Array.isArray(req.messages) || req.messages.length === 0) {
+    return buildErrorResponse('messages is required and must be a non-empty array', 'invalid_request_error', 400);
+  }
+
+  // Validate each message has role and content
+  for (const msg of req.messages) {
+    if (!msg || typeof msg !== 'object') {
+      return buildErrorResponse('Each message must be an object', 'invalid_request_error', 400);
+    }
+    const m = msg as Record<string, unknown>;
+    if (!m.role || typeof m.role !== 'string') {
+      return buildErrorResponse('Each message must have a role', 'invalid_request_error', 400);
+    }
+  }
+
+  return null;
+}
diff --git a/src/api/server.ts b/src/api/server.ts
index 9ce3f36..66f5a0c 100644
--- a/src/api/server.ts
+++ b/src/api/server.ts
@@ -11,6 +11,12 @@ import { listPairingRequests, approvePairingCode } from '../pairing/store.js';
 import { parseMultipart } from './multipart.js';
 import type { AgentRouter } from '../core/interfaces.js';
 import type { ChannelId } from '../core/types.js';
+import {
+  generateCompletionId, extractLastUserMessage, buildCompletion,
+  buildChunk, buildToolCallChunk, formatSSE, SSE_DONE,
+  buildErrorResponse, buildModelList, validateChatRequest,
+} from './openai-compat.js';
+import type { OpenAIChatRequest } from './openai-compat.js';
 
 import { createLogger } from '../logger.js';
 
@@ -36,7 +42,7 @@ export function createApiServer(deliverer: AgentRouter, options: ServerOptions):
     const corsOrigin = options.corsOrigin || req.headers.origin || 'null';
     res.setHeader('Access-Control-Allow-Origin', corsOrigin);
     res.setHeader('Access-Control-Allow-Methods', 'GET, POST, OPTIONS');
-    res.setHeader('Access-Control-Allow-Headers', 'Content-Type, X-Api-Key');
+    res.setHeader('Access-Control-Allow-Headers', 'Content-Type, X-Api-Key, Authorization');
 
     // Handle OPTIONS preflight
     if (req.method === 'OPTIONS') {
@@ -306,6 +312,173 @@ export function createApiServer(deliverer: AgentRouter, options: ServerOptions):
       return;
     }
 
+    // Route: GET /v1/models (OpenAI-compatible)
+    if (req.url === '/v1/models' && req.method === 'GET') {
+      try {
+        if (!validateApiKey(req.headers, options.apiKey)) {
+          const err = buildErrorResponse('Invalid API key', 'invalid_request_error', 401);
+          res.writeHead(err.status, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(err.body));
+          return;
+        }
+
+        const models = buildModelList(deliverer.getAgentNames());
+        res.writeHead(200, { 'Content-Type': 'application/json' });
+        res.end(JSON.stringify(models));
+      } catch (error: any) {
+        console.error('[API] Models error:', error);
+        const err = buildErrorResponse(error.message || 'Internal server error', 'server_error', 500);
+        res.writeHead(err.status, { 'Content-Type': 'application/json' });
+        res.end(JSON.stringify(err.body));
+      }
+      return;
+    }
+
+    // Route: POST /v1/chat/completions (OpenAI-compatible)
+    if (req.url === '/v1/chat/completions' && req.method === 'POST') {
+      try {
+        if (!validateApiKey(req.headers, options.apiKey)) {
+          const err = buildErrorResponse('Invalid API key', 'invalid_request_error', 401);
+          res.writeHead(err.status, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(err.body));
+          return;
+        }
+
+        const contentType = req.headers['content-type'] || '';
+        if (!contentType.includes('application/json')) {
+          const err = buildErrorResponse('Content-Type must be application/json', 'invalid_request_error', 400);
+          res.writeHead(err.status, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(err.body));
+          return;
+        }
+
+        const body = await readBody(req, MAX_BODY_SIZE);
+        let parsed: unknown;
+        try {
+          parsed = JSON.parse(body);
+        } catch {
+          const err = buildErrorResponse('Invalid JSON body', 'invalid_request_error', 400);
+          res.writeHead(err.status, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(err.body));
+          return;
+        }
+
+        // Validate OpenAI request shape
+        const validationError = validateChatRequest(parsed);
+        if (validationError) {
+          res.writeHead(validationError.status, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(validationError.body));
+          return;
+        }
+
+        const chatReq = parsed as OpenAIChatRequest;
+
+        // Extract the last user message
+        const userMessage = extractLastUserMessage(chatReq.messages);
+        if (!userMessage) {
+          const err = buildErrorResponse('No user message found in messages array', 'invalid_request_error', 400);
+          res.writeHead(err.status, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(err.body));
+          return;
+        }
+
+        if (userMessage.length > MAX_TEXT_LENGTH) {
+          const err = buildErrorResponse(`Message too long (max ${MAX_TEXT_LENGTH} chars)`, 'invalid_request_error', 400);
+          res.writeHead(err.status, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(err.body));
+          return;
+        }
+
+        // Resolve agent from model field
+        const agentNames = deliverer.getAgentNames();
+        const modelName = chatReq.model || agentNames[0];
+        const agentName = agentNames.includes(modelName) ? modelName : undefined;
+
+        // If an explicit model was requested but doesn't match any agent, error
+        if (chatReq.model && !agentNames.includes(chatReq.model)) {
+          const err = buildErrorResponse(
+            `Model not found: ${chatReq.model}. Available: ${agentNames.join(', ')}`,
+            'model_not_found',
+            404,
+          );
+          res.writeHead(err.status, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(err.body));
+          return;
+        }
+
+        const completionId = generateCompletionId();
+        const context = { type: 'webhook' as const, outputMode: 'silent' as const };
+
+        console.log(`[API] OpenAI chat: model="${modelName}", stream=${!!chatReq.stream}, msg="${userMessage.slice(0, 100)}..."`);
+
+        if (chatReq.stream) {
+          // ---- Streaming response ----
+          res.writeHead(200, {
+            'Content-Type': 'text/event-stream',
+            'Cache-Control': 'no-cache',
+            'Connection': 'keep-alive',
+          });
+
+          let clientDisconnected = false;
+          req.on('close', () => { clientDisconnected = true; });
+
+          // First chunk: role announcement
+          res.write(formatSSE(buildChunk(completionId, modelName, { role: 'assistant' })));
+
+          try {
+            let toolIndex = 0;
+
+            for await (const msg of deliverer.streamToAgent(agentName, userMessage, context)) {
+              if (clientDisconnected) break;
+
+              if (msg.type === 'assistant' && msg.content) {
+                // Text content delta
+                res.write(formatSSE(buildChunk(completionId, modelName, { content: msg.content })));
+              } else if (msg.type === 'tool_call') {
+                // Tool call delta (emit name + args in one chunk)
+                const toolCallId = msg.toolCallId || `call_${msg.uuid || 'unknown'}`;
+                const toolName = msg.toolName || 'unknown';
+                const args = msg.toolInput ? JSON.stringify(msg.toolInput) : '{}';
+                res.write(formatSSE(buildToolCallChunk(
+                  completionId, modelName, toolIndex++, toolCallId, toolName, args,
+                )));
+              } else if (msg.type === 'result') {
+                // Final chunk
+                break;
+              }
+              // Skip 'reasoning', 'tool_result', and other internal types
+            }
+          } catch (streamError: any) {
+            if (!clientDisconnected) {
+              // Emit error as a content delta so clients see it
+              res.write(formatSSE(buildChunk(completionId, modelName, {
+                content: `\n\n[Error: ${streamError.message}]`,
+              })));
+            }
+          }
+
+          // Finish chunk + done sentinel
+          if (!clientDisconnected) {
+            res.write(formatSSE(buildChunk(completionId, modelName, {}, 'stop')));
+            res.write(SSE_DONE);
+          }
+          res.end();
+        } else {
+          // ---- Sync response ----
+          const response = await deliverer.sendToAgent(agentName, userMessage, context);
+          const completion = buildCompletion(completionId, modelName, response);
+          res.writeHead(200, { 'Content-Type': 'application/json' });
+          res.end(JSON.stringify(completion));
+        }
+      } catch (error: any) {
+        console.error('[API] OpenAI chat error:', error);
+        const err = buildErrorResponse(error.message || 'Internal server error', 'server_error', 500);
+        res.writeHead(err.status, { 'Content-Type': 'application/json' });
+        res.end(JSON.stringify(err.body));
+      }
+      return;
+    }
+
     // Route: 404 Not Found
     sendError(res, 404, 'Not found');
   });