Call the HTTP gateway
Authenticate and send OpenAI, Responses, or Anthropic-compatible requests through RouteKit.
Call the HTTP gateway
Use the HTTP gateway for applications and tools that can target an OpenAI-compatible or Anthropic-compatible endpoint.
Get the local endpoint and token
routekit status
ROUTEKIT_TOKEN="$(routekit daemon auth show)"
ROUTEKIT_URL="http://127.0.0.1:8080"routekit daemon auth show reveals the private owner token. Do not put it in
logs, source control, issue reports, or shared shell history. Issue a named
token with routekit token issue <label> for another client or user.
Discover available models
curl -sS "$ROUTEKIT_URL/v1/models" \
-H "Authorization: Bearer $ROUTEKIT_TOKEN"Use an advertised provider/model ID exactly as returned.
Send a request
curl -sS "$ROUTEKIT_URL/v1/chat/completions" \
-H "Authorization: Bearer $ROUTEKIT_TOKEN" \
-H 'Content-Type: application/json' \
-d '{
"model": "openai/gpt-5.5",
"messages": [{"role": "user", "content": "Say hello."}]
}'The gateway accepts Authorization: Bearer <token> or x-api-key. Model,
usage, and catalog routes require authentication. /health is intentionally
unauthenticated for readiness checks.
Endpoint families
| Family | Paths |
|---|---|
| Health and usage | GET /health, GET /usage |
| Models | GET /v1/models, GET /models, GET /v1/models/{id} |
| OpenAI | POST /v1/chat/completions, POST /chat/completions, POST /v1/embeddings |
| Anthropic | POST /v1/messages, POST /v1/messages/count_tokens |
| Responses and Codex | POST /v1/responses, GET /backend-api/codex/models, POST /backend-api/codex/responses |
Chat Completions, Messages, and Responses support streaming and tools. The
retained /v1/cursor adapter is an internal compatibility surface, not a
public client-support claim. See
client compatibility for the supported
client boundary.
Inspect a result or failure
Model-call responses include x-routekit-model-call-id. Inspect one with:
routekit calls inspect <call-id>Request bodies are limited to 16 MiB. Unknown models return 400; an empty
catalog returns 503; normalized provider rate limits return 429 with
Retry-After when a reset is known; other upstream failures return 502.
Client disconnects cancel upstream work.