openai-chatkit-backend-python

Design a self-hosted ChatKit backend by creating FastAPI endpoints (/chatkit/api and optional /chatkit/api/upload), integrating the OpenAI Agents SDK, configuring a provider factory to switch between OpenAI and Gemini, and implementing Server-Sent Events for response streaming while enforcing store contracts and protecting API keys server-side.

Yes. ChatKit runs on custom Python infrastructure via FastAPI, Django, or Flask backends that expose the required /chatkit/api endpoint, manage agent orchestration through the OpenAI Agents SDK, and handle authentication and persistence through store and FileStore contracts.

Point the ChatKit frontend widget to your backend's /chatkit/api endpoint. Configure environment variables (LLM_PROVIDER, OPENAI_API_KEY, GEMINI_API_KEY) server-side, implement streaming via Server-Sent Events, and maintain clean frontend/backend separation with proper authentication and no API key exposure.

Use the Agents SDK within your Python backend to orchestrate reasoning and tool execution. Implement a provider factory (create_model) to instantiate agents, route chat requests through /chatkit/api, stream agent responses via SSE, and optionally support Gemini through an OpenAI-compatible endpoint.

Yes. Self-hosted backends stream chat responses via Server-Sent Events through /chatkit/api and handle file uploads via optional POST /chatkit/api/upload, with uploads managed through the FileStore contract and proper server-side authentication.

Gemini requires an OpenAI-compatible endpoint to work alongside OpenAI in ChatKit backends. The provider factory must correctly instantiate models based on configuration, and both providers must adhere to the Agents SDK interface for consistent reasoning and tool execution behavior.