OpenAI Compatibility

“OpenAI-compatible” in FeltnerAI refers to the upstream provider you connect — the model service FeltnerAI calls on your behalf. FeltnerAI speaks the OpenAI Chat Completions protocol to that provider so you can bring almost any compatible service.

FeltnerAI is not an OpenAI gateway

FeltnerAI does not expose a user-facing OpenAI-compatible endpoint and does not issue personal API keys. Clients talk to FeltnerAI’s own versioned API under /api/v1 (see the API Reference). OpenAI compatibility is strictly about the provider it connects to.

What a compatible provider must offer

A provider works with FeltnerAI if it implements the OpenAI-style API surface:

• Base URL — the root of the provider’s OpenAI-compatible API.
• Chat completions — a streaming chat/completions endpoint.
• Model discovery — a models endpoint FeltnerAI calls to list available models.
• Authentication — a bearer API key, and/or custom secret headers the provider requires.

This covers the OpenAI API itself as well as the many self-hosted servers and hosted gateways that implement the same contract.

How FeltnerAI uses the provider

1. You add the provider during setup or from Admin → Providers.
2. FeltnerAI validates the credentials with a live connection test before saving.
3. On save, FeltnerAI discovers models from the provider’s model endpoint.
4. When a user sends a message, FeltnerAI calls the provider’s streaming chat endpoint and normalizes the response into its own SSE events (started, delta, completed, error) for the client.

Provider API keys and custom secret headers are encrypted at rest with the server’s local encryption key.

Why this design

Normalizing at the server keeps clients simple and consistent, keeps your provider credentials on the server (never shipped to clients), and lets you switch providers without changing how clients behave.

See Configure Providers for the step-by-step setup.