Glimrel team@glimrel.store Get a gateway key

Docs

Gateway URL, LLM routes, traces, evals.

Base URL https://gateway.glimrel.store. Bearer tokens start with glm_live_. OpenAPI lives at /v1/openapi.json.

Quickstart

Create a project in the console, copy a test token, and hit the health route. Ujjwal Kumar Singh issues the first live token after a short call.

curl -s https://gateway.glimrel.store/v1/health \
  -H "Authorization: Bearer glm_live_test"

Auth

Send Authorization: Bearer on every call. Rotate keys from the console. Test keys cannot touch production objects.

AI routes

Three jobs on LLM traffic. JSON in, JSON out. Errors use type, message, and request_id.

  • Route — OpenAI-shaped completions through the gateway, with fallbacks and caches. POST /v1/chat/completions
  • Trace — nested spans for completions, tools, retrieval, and retries. GET /v1/traces/{id}
  • Eval — LLM-as-judge, code check, or human on a production sample. Same grader as CI. POST /v1/evals/runs
curl -s https://gateway.glimrel.store/v1/chat/completions \
  -H "Authorization: Bearer glm_live_test" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o","messages":[{"role":"user","content":"ping"}]}'
POST /v1/chat/completions
GET  /v1/traces/{id}
POST /v1/evals/runs

Planned families on the route sheet: Claude 3.7 Sonnet, GPT-4o / mini, Gemini 2.5, Llama 3 70B, Amazon Nova, Groq. Access paths: Bedrock, Anthropic, OpenAI direct, Groq, Hugging Face. Glimrel does not train those models and does not run agents.

Limits

Default 30 requests per second per key on River. Burst tokens refill each minute. File a note to team@glimrel.store if a launch needs a higher ceiling.

Status

Incident notes go to team@glimrel.store. Webhooks retry for 24 hours with exponential backoff.