fastapi-ai-chat 0.1.1

Minimal FastAPI backend for streaming AI chat over SSE (token/done/error), compatible with ai-chat-kit

fastapi-ai-chat

Minimal FastAPI backend for streaming AI chat over Server‑Sent Events (SSE).

It’s designed to be wire‑compatible with ai-chat-kit’s React FastAPI adapter:

• POST /chat/stream returns text/event-stream
• emits SSE events: token, done, error

Features

• Token streaming over SSE: incremental token events + final done event (or error).
• Provider routing by model name:
  • OpenAI for most models (default: gpt-4.1)
  • Anthropic for models starting with claude
  • Gemini for models starting with gemini
• Mock streaming fallback: deterministic mock stream when no provider keys are configured (great for UI/dev).
• Conversation persistence (optional): SQLite chat history + basic conversation CRUD endpoints.
• CORS enabled: defaults to * for local dev (configurable).

Install

From PyPI:

python -m pip install "fastapi-ai-chat[server]"

Quickstart (run the example server)

cd Backend/fastapi-ai-chat
python -m uvicorn examples.server:app --reload --port 8000

Configuration (environment variables)

This library intentionally does not load .env files for you; it reads process env vars (or you can pass keys directly to create_app(...)).

• Provider keys
  • OPENAI_API_KEY or FASTAPI_AI_CHAT_OPENAI_API_KEY
  • ANTHROPIC_API_KEY or FASTAPI_AI_CHAT_ANTHROPIC_API_KEY
  • GEMINI_API_KEY or FASTAPI_AI_CHAT_GEMINI_API_KEY
• Persistence
  • FASTAPI_AI_CHAT_SQLITE_PATH (default: ./chat_history.sqlite3)
• Optional provider base URL overrides
  • OPENAI_BASE_URL / FASTAPI_AI_CHAT_OPENAI_BASE_URL
  • ANTHROPIC_BASE_URL / FASTAPI_AI_CHAT_ANTHROPIC_BASE_URL
  • GEMINI_BASE_URL / FASTAPI_AI_CHAT_GEMINI_BASE_URL

Example use

1) Stream tokens (curl)

curl -N -X POST "http://localhost:8000/chat/stream" \
  -H "Content-Type: application/json" \
  -d '{
    "userId":"u1",
    "messages":[{"role":"user","content":"Hello streaming"}]
  }'

2) Pick a real model/provider

Send params.model in the request body:

• OpenAI: gpt-4.1 (default if omitted)
• Anthropic: claude-3-5-sonnet (or any model starting with claude)
• Gemini: gemini-1.5-pro (or any model starting with gemini)

Example:

curl -N -X POST "http://localhost:8000/chat/stream" \
  -H "Content-Type: application/json" \
  -d '{
    "userId":"u1",
    "messages":[{"role":"user","content":"Write a haiku about SSE."}],
    "params":{"model":"gpt-4.1"}
  }'

If you don’t configure provider keys, the backend falls back to a deterministic mock stream unless a real model was explicitly requested via params.model.

3) Embed in your own FastAPI app

from fastapi_ai_chat import create_app

app = create_app(
    openai_api_key="...",  # or None / "" to disable OpenAI
    anthropic_api_key="...",
    gemini_api_key="...",
    sqlite_path="./chat_history.sqlite3",
    cors_allow_origins=["http://localhost:5173"],
)

API

Streaming protocol (SSE)

The server emits frames separated by a blank line (\n\n):

• event: token with JSON: {"type":"token","token":"..."} (many times)
• event: done with JSON:
  • {"type":"done","message":{"role":"assistant","content":"..."}, "provider":"...", "usage":{...}, "conversationId":"..."}
• event: error with JSON: {"type":"error","error":{"message":"..."}}

Endpoints

• POST /chat/stream: stream assistant tokens (SSE)
• GET /chat/history?userId=...: get message history for a user
• POST /chat/clear: clear a user’s history
• GET /chat/conversations?userId=...: list conversations (SQLite mode)
• GET /chat/conversation?userId=...&conversationId=...: fetch one conversation (SQLite mode)
• POST /chat/conversation/create: create/ensure a conversation (SQLite mode)
• POST /chat/conversation/delete: delete a conversation (SQLite mode)

Publish to PyPI (maintainers)

Build:

python -m pip install -U build twine
python -m build Backend/fastapi-ai-chat

Upload:

• Recommended: PyPI Trusted Publishing
• Or token-based:

python -m twine upload Backend/fastapi-ai-chat/dist/*

Author

• Name: Shrikant Jagtap
• Email: [e-mail]