feat: add streaming support to LlamaCppClient

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

src/models/local/llamacpp.test.ts

@@ -1,5 +1,6 @@
 import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
 import { LlamaCppClient } from './llamacpp.js';
+import type { ChatStreamEvent } from '../types.js';
 
 describe('LlamaCppClient', () => {
   const mockFetch = vi.fn();
@@ -33,4 +34,51 @@ describe('LlamaCppClient', () => {
     expect(response.usage.inputTokens).toBe(10);
     expect(response.usage.outputTokens).toBe(5);
   });
+
+  it('streams responses via SSE', async () => {
+    const chunks = [
+      'data: {"choices":[{"delta":{"content":"Hello"}}]}\n\n',
+      'data: {"choices":[{"delta":{"content":" world"}}]}\n\n',
+      'data: {"choices":[{}],"usage":{"prompt_tokens":5,"completion_tokens":2}}\n\n',
+      'data: [DONE]\n\n',
+    ];
+
+    const encoder = new TextEncoder();
+    let chunkIndex = 0;
+
+    const mockStream = new ReadableStream({
+      pull(controller) {
+        if (chunkIndex < chunks.length) {
+          controller.enqueue(encoder.encode(chunks[chunkIndex]));
+          chunkIndex++;
+        } else {
+          controller.close();
+        }
+      },
+    });
+
+    mockFetch.mockResolvedValue({
+      ok: true,
+      body: mockStream,
+    });
+
+    const client = new LlamaCppClient({
+      endpoint: 'http://localhost:8080',
+    });
+
+    const events: ChatStreamEvent[] = [];
+    for await (const event of client.chatStream({
+      messages: [{ role: 'user', content: 'Hi' }],
+    })) {
+      events.push(event);
+    }
+
+    expect(events).toHaveLength(3);
+    expect(events[0]).toEqual({ type: 'content', content: 'Hello' });
+    expect(events[1]).toEqual({ type: 'content', content: ' world' });
+    expect(events[2]).toEqual({
+      type: 'done',
+      usage: { inputTokens: 5, outputTokens: 2 },
+    });
+  });
 });