Add JSON API endpoint (/api/ask, /api/remaining) for custom chat UI

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

app.py

@@ -459,7 +459,91 @@ def build_app() -> gr.Blocks:
     return demo
 
 
+def ask_question(question: str) -> dict:
+    """Synchronous API for the custom chat UI."""
+    if not _check_and_increment():
+        return {
+            "answer": None,
+            "error": "daily_limit",
+            "remaining": 0,
+            "trace_html": "",
+            "stats": "",
+        }
+
+    if not MODEL:
+        return {"answer": None, "error": "LH_MODEL not set", "remaining": _remaining()}
+
+    scratch_dir = Path(tempfile.mkdtemp(prefix="lh-scratch-"))
+    system_prompt = build_system_prompt(base_prompt="", workspace=WORKSPACE_DIR)
+    messages: list[Message] = [
+        {"role": "system", "content": system_prompt},
+        {"role": "user", "content": question},
+    ]
+
+    start = time.monotonic()
+    agent_run = run_agent_loop(
+        model=MODEL,
+        messages=messages,
+        tools=TOOL_DEFINITIONS,
+        completion=litellm.completion,
+        workspace=WORKSPACE_DIR,
+        scratch_dir=scratch_dir,
+        sandbox_fn=e2b_run_python,
+    )
+
+    try:
+        for event in agent_run:
+            pass
+    except Exception as exc:
+        return {"answer": None, "error": str(exc), "remaining": _remaining()}
+
+    trace = agent_run.trace
+    trace.wall_time_s = round(time.monotonic() - start, 2)
+    result = {
+        "question": question,
+        "source": SOURCE,
+        "passed": True,
+        "assertions": {},
+        "trace": asdict(trace),
+    }
+    upload_trace(result)
+    trace_html = render_trace(result, max_chars=2000)
+
+    return {
+        "answer": trace.answer or "(no answer)",
+        "stats": format_stats(trace),
+        "trace_html": trace_html,
+        "remaining": _remaining(),
+        "error": None,
+    }
+
+
 if __name__ == "__main__":
+    import fastapi
+    from fastapi.middleware.cors import CORSMiddleware
+
     WORKSPACE_DIR = load_workspace()
-    app = build_app()
-    app.launch(ssr_mode=False)
+    demo = build_app()
+
+    # Mount the JSON API on the same FastAPI app
+    fastapi_app = fastapi.FastAPI()
+    fastapi_app.add_middleware(
+        CORSMiddleware,
+        allow_origins=["*"],
+        allow_methods=["POST"],
+        allow_headers=["*"],
+    )
+
+    @fastapi_app.post("/api/ask")
+    async def api_ask(request: fastapi.Request):
+        body = await request.json()
+        question = body.get("question", "")
+        if not question:
+            return {"error": "No question provided"}
+        return ask_question(question)
+
+    @fastapi_app.get("/api/remaining")
+    async def api_remaining():
+        return {"remaining": _remaining(), "limit": DAILY_LIMIT}
+
+    demo.launch(ssr_mode=False, app=fastapi_app)