LLM gateway · traces · evals
Glimrel is a core AI/ML control plane: route LLM calls, nest traces for tools and retries, and grade live spans with the same evals you already run in CI. Point the SDK you have at one host. Glimrel is not an agent product — it sits under the agents you already ship.
AI features
Four named jobs on LLM traffic. Each card is a feature: what it does, who it is for, and the route. Glimrel routes, traces, and grades model calls. It does not train those models, and it does not ship agents.
AI feature · Multi-model routing
For platform teams wiring OpenAI, Anthropic, Gemini, Llama, Nova, or Groq
Keep the OpenAI-shaped client. Change the base URL and the key. Fallbacks, caches, and spend caps attach to that key. Planned access paths: Bedrock, Anthropic, OpenAI direct, Groq, Hugging Face.
SDK → Glimrel gateway → provider fallback → POST /v1/chat/completions
AI feature · Nested LLM traces
For engineers debugging agent and RAG traffic
Every completion, tool call, and retry is a span you can search. Customer, prompt version, and cost sit on the tree. This is observability for LLM traffic — not a chatbot.
Request → nested spans → GET /v1/traces/{id}
AI feature · LLM-as-judge evals
For quality leads who refuse silent drift
LLM judge, code check, or human. The same scorer that already sits in CI runs on a sample of production spans. Prompt versions that drop faithfulness get held.
Sample prod → same grader as CI → POST /v1/evals/runs
AI feature · Agent traffic plane
For teams that already ship agents
Glimrel is the plane under agent frameworks (LangChain, OpenAI Assistants, homebrew). It does not choose tools or act as an agent. Spend caps halt traffic at the limit you set.
Agent SDK → one Glimrel key → route, trace, grade
Model families on the route sheet: Claude 3.7 Sonnet, GPT-4o / GPT-4o mini, Gemini 2.5, Llama 3 70B, Amazon Nova, Groq. Glimrel does not train those models. MVP stage — no invented customers or uptime claims on this page.
Method
Swap the host. Watch the current. Sample production with the grader you already trust.
Keep your OpenAI client. Change the host and the key. Routing, fallbacks, and caches attach to that key.
Requests, tokens, cache hits, and spend draw on one dashboard. Search nested LLM spans across millions of calls.
Sample production, score it with the LLM judge or code check you already run in CI, and hold the prompt version that drops below your line.
Stack
An LLM control plane. Gateway, traces, and evals — not a chat widget and not a trained foundation model.
One base URL, fallbacks, caches, and budgets per key, customer, or org.
Nested spans for tools, retrieval, and retries. Search millions of LLM calls.
LLM judge, code check, or human. Same scorer on a fixture and on 5% of prod.
Error rate, cost, and latency trip Slack the minute they cross your line.
Start
Write team@glimrel.store. You get a test token, a sample gateway project, and a 30-minute walkthrough on the LLM traffic you want to watch.
Write team@glimrel.store. This form stores the request on the page so you can copy it into that mail.
Thanks. We saved this request on the page. Mail team@glimrel.store if you want a live reply.