app.py

"""HF Spaces 单进程入口:后台线程起 FastAPI,前台起 Gradio(指回本机 API)。

Space 用 OpenAI 兼容后端(LLM_BACKEND=openai / EMBED_BACKEND=openai),端点与模型名经
Space secrets 注入:OPENAI_BASE_URL、OPENAI_API_KEY、MODEL_BASE、MODEL_FT
(可选 OPENAI_EMBED_BASE_URL、FINSIGHT_EMBED_MODEL)。本地 `python app.py` 不设这些即走 Ollama。
"""
import os
import threading
import time

import requests
import uvicorn

API_PORT = int(os.getenv("API_PORT", "8000"))
os.environ.setdefault("FINSIGHT_API", f"http://127.0.0.1:{API_PORT}")


def _serve_api():
    uvicorn.run("api.main:app", host="127.0.0.1", port=API_PORT, log_level="warning")


def _wait_api(timeout: int = 30) -> bool:
    for _ in range(timeout * 2):
        try:
            requests.get(f"http://127.0.0.1:{API_PORT}/health", timeout=2)
            return True
        except Exception:
            time.sleep(0.5)
    return False


def main():
    # 知识库为空时用种子语料初始化(Space 首次启动 rag/.chroma 不在仓库里)
    try:
        from rag import KnowledgeBase, build_seed_kb
        if KnowledgeBase().count() == 0:
            build_seed_kb()
            print("seed KB built.")
    except Exception as e:
        print("seed KB init skipped:", e)

    threading.Thread(target=_serve_api, daemon=True).start()
    if not _wait_api():
        print("⚠️ API 未就绪,前端仍会启动(各 Tab 会显示后端错误)。")

    from frontend.app import build
    build().launch(server_name="0.0.0.0", server_port=int(os.getenv("PORT", "7860")))


if __name__ == "__main__":
    main()