# LunaRoute

> LLM gateway: send OpenAI-, Anthropic-, or Responses-shaped requests to one
> endpoint and they route to a model on the LunaRoute GPU fleet, or to a
> provider key you bring. No SDK to install — plain HTTP.

## Configure a coding agent (recommended)

One command via the CLI — no manual config editing:

    npx @lunaroute/cli setup <harness>

harness: pi | claude-code | claude-desktop | codex | opencode | openclaw | hermes | copilot-cli | dsh | generic.
Non-interactive: add `--key lr_... --yes`. For pi and opencode, setup offers to
install an extension; once installed, models sync automatically.

## Generic API configuration (any OpenAI/Anthropic-compatible client)

- SDK base URL: https://gw.lunaroute.com/v1 — clients append their own path
  (e.g. chat/completions). Raw HTTP requests go to
  https://gw.lunaroute.com/v1/<endpoint> (paths below).
- API key: create one in the dashboard (https://app.lunaroute.com); `lr_` prefix;
  authenticates one organization (optionally one project).
- Auth header: `LUNAROUTE-API-KEY: <key>` (variants per dialect on the
  Authentication page).
- Endpoints: /v1/chat/completions (OpenAI), /v1/messages (Anthropic),
  /v1/responses (Responses), /v1/embeddings, /v1/rerank, /v1/images/*.
- Models: live catalog at https://gw.lunaroute.com/v1/models — never hardcode a
  model list; read it at setup time.

## Docs

Every link below is the raw markdown twin of the page; drop the `.md` for the
rendered HTML, or read `/llms-full.txt` for all of it in one file.

- CLI: https://docs.lunaroute.com/cli/index.md: per-harness setup details, `lunaroute run`,
  catalog/pricing/usage commands, project memory.
- Getting started: https://docs.lunaroute.com/getting-started/index.md: key + first request.
- Claude Code: https://docs.lunaroute.com/getting-started/claude-code.md: per-session `lunaroute run claude-code`, or the permanent `ANTHROPIC_*` exports.
- Codex: https://docs.lunaroute.com/getting-started/codex.md: per-session `lunaroute run codex`, or the `codex --profile lunaroute` profile file.
- Authentication: https://docs.lunaroute.com/getting-started/authentication.md: header variants per dialect.
- API reference: https://docs.lunaroute.com/api/index.md: full endpoint list, error envelopes, feature-gated 503s.
- Model catalog: https://docs.lunaroute.com/models/index.md: curated roster, quantization variants.
- MCP server: https://docs.lunaroute.com/mcp/index.md: hosted MCP at mcp.lunaroute.com.
- Personal assistants: https://docs.lunaroute.com/assistants/index.md: connect OpenClaw, Hermes, NanoClaw, Mira.
- Concurrency & overflow: https://docs.lunaroute.com/concepts/lanes.md: priority scheduling, automatic overflow, queues and member caps.
- Usage & billing: https://docs.lunaroute.com/concepts/usage-and-billing.md: flat-rate inference — no token caps or inference overages; some MCP tools (e.g. web_search) metered separately; usage ledger in the dashboard (https://app.lunaroute.com).
- Full text: https://docs.lunaroute.com/llms-full.txt: every docs page concatenated into a single file for a one-fetch read.

## Optional

- First request examples: https://docs.lunaroute.com/getting-started/first-request.md: curl examples for OpenAI, Anthropic, streaming.
- Zero data retention: https://docs.lunaroute.com/concepts/zero-data-retention.md: privacy posture.
- Projects & orgs: https://docs.lunaroute.com/concepts/projects-and-orgs.md: key scoping.