LunaRoute Docs

LunaRoute is an inference router for coding agents. We curate a roster of frontier models and serve them from our own managed US fleet behind one OpenAI-, Anthropic-, or Responses-compatible endpoint — billed flat by lane concurrency, with zero data retention.

Routing through your own provider keys is supported, but it is the side door: the curated roster is what we run, certify, and stand behind.

What you get

• A curated model roster, served by us. GLM, Kimi, Qwen, MiniMax, DeepSeek and more, each run and certified by LunaRoute. Switch models by name; new ones dock without an endpoint change. See Models.
• The agent toolbelt, on the same key. Over MCP:
  • web_search across Brave, Exa, and Kagi — bring your own search-provider key if you have one.
  • Document conversion and OCR for Word, PowerPoint, Excel, PDF, and scanned documents → Markdown.
  • Image generation and editing across multiple image models.
• Plans priced by the lane, not the token. No usage cap, no overage bill. Interactive work gets priority; background work uses a -background model variant.
• Bring your own keys — optional. Already hold OpenAI, Anthropic, or other provider keys? You can route through them as a secondary path, or use ours. Either way it is one endpoint.
• Zero data retention. Request and response bodies are processed in memory and never stored. See Zero data retention.

Start here

1. Get an API key and send your first request.
2. Read Lanes & scheduling before you size a worker pool.
3. Point your coding agent at LunaRoute with the CLI or the hosted MCP server.
4. Wire up the API — chat completions, Messages, Responses, embeddings, rerank, and images.
5. Running a personal agent? Connect OpenClaw, Hermes, or NanoClaw.

Planning for a team? Teams adds SSO, audit logs, and per-user/per-key lane assignment; Enterprise adds private, single-tenant inference.