07/Integrations
Connecting Clients
Base URLs and auth to connect IDEs and tools to the gateway.
Most clients connect via a "custom OpenAI base URL". Some tools (Claude, Cursor, Codex, Grafana) use their own compatible paths. All authenticate with a Bearer token.
#Base URLs
| Path | Used by |
|---|---|
| /v1 | OpenAI-compatible clients (VS Code, Kilo, Codex CLI, Cursor override) |
| /claude | Claude Code, Claude Desktop (Anthropic-compatible) |
| /openclaw | OpenClaw |
| /codex | Codex Desktop App (Responses API conversion) |
| /cursor | Cursor IDE |
| /grafana/assistant | Grafana Assistant |
| /api/* | Ollama-native clients |
Model naming
The gateway forwards the model id the client sends; it does not guess unknown names. Use a mapped model name or a model group name. Using a group name makes dynamic model switching easy.
#Provider node types
| Provider | Endpoints | Highlights |
|---|---|---|
| Ollama | /api/chat, /api/generate, /api/pull | Local inference, model pull, warmup, health checks |
| vLLM | /v1/chat/completions, /v1/embeddings | OpenAI-compatible, SSE streaming, Bearer forwarding |
| Cursor AI | /v1/chat/completions | Auto base URL, model discovery |
| Antigravity | /v1internal | Google OAuth 2.0, Gemini & Claude, auto token refresh |
| AWS Bedrock | /bedrock | Converse API, image input, streaming, STS session tokens |