Connect the gateway to editors / IDEs
This service exposes OpenAI-compatible POST /v1/chat/completions (and /v1/models, etc.). Most editors or plugins that support OpenAI API / custom OpenAI base URL only need Base URL at …/v1 and API Key set to the tenant sk-… for completions, chat, etc.
Product UIs change with versions; this page lists wiring parameters and troubleshooting angles.
Three essentials
| Item | Value |
|---|---|
| API Base URL | /v1 root, e.g. https://gw.example.com/v1 |
| API Key | Tenant key used in Authorization: Bearer |
| Model | Exact code from GET /v1/models (the model field in the body) |
Verify /v1/models with curl or the browser before IDE setup to separate network/TLS from config issues.
Common setups
- Continue / Cline / some JetBrains extensions / Zed: OpenAI-compatible or custom: base + key + model.
- Cursor / Windsurf: if custom OpenAI base URL is supported, point here; if only official OpenAI is allowed, you need a compatible proxy.
- GitHub Copilot itself: usually cannot change base; use Continue-style extensions toward the same endpoint.
CLI debugging: cURL examples, SDK quickstart.
Network and SSE
Intranet must reach the host; HTTPS needs trusted certs. If streaming completions stall, check reverse-proxy proxy_buffering, timeouts, HTTP/2 vs SSE (see Troubleshooting).
HTTP protocol surface
- Multimodal:
input_filepick one of three options—Chat Completions. - tools / tool_calls: Tool calls and streaming.