Remote API Gateway

Biyan can expose configured cloud providers through a local OpenAI-compatible endpoint. The server runs on your device, but inference remains remote: requests are forwarded to the provider selected by the model mapping.

⚠️

This is not a local inference server. Biyan never starts llama.cpp, MLX, Foundation Models, or another local model backend.

Quick start

Configure a provider

Add credentials under Settings → Model Providers and verify a normal chat works.

Start the gateway

Open Settings → API Gateway, choose a loopback host and port, set an API key, and start the gateway. Keep the default 127.0.0.1 binding unless you understand the network exposure.

Send a request


curl http://127.0.0.1:1337/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_LOCAL_GATEWAY_KEY" \
  -d '{
    "model": "YOUR_CONFIGURED_REMOTE_MODEL",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Security

Use a different local gateway key from every provider API key.
Prefer 127.0.0.1; binding to 0.0.0.0 exposes the service to the network.
Provider credentials stay in Biyan and are not returned to gateway clients.
Requests, attachments, and tool results are still processed by the selected cloud provider.

Retired local-model routes return HTTP 410 with code LOCAL_RUNTIME_REMOVED. They do not download a model or start a backend.

MCP Servers App Settings