Personal assistants

LunaRoute is inference for the personal agent you already run — not another assistant app. Personal agents connect like any OpenAI-compatible endpoint: point them at https://gw.lunaroute.com/v1 with your lr_ API key. Tool-using agents can also add the hosted MCP server for web search, document conversion, and images.

Personal-agent traffic gets its own lane. See the Personal Agent Inference plan — a flat $19.99/mo lane (GLM-5.2) with zero data retention.

OpenClaw

lunaroute setup openclaw

This merges a lunaroute provider into ~/.openclaw/openclaw.json (api: openai-completions, baseUrl: https://gw.lunaroute.com/v1, key read from ${LUNAROUTE_API_KEY}). Your existing config is preserved. Then restart the gateway and confirm with openclaw models list / openclaw models set lunaroute/<model>.

Hermes

lunaroute setup hermes

Hermes owns its own config, so setup drives its CLI instead of writing files:

hermes config set providers.lunaroute.api https://gw.lunaroute.com/v1
hermes config set providers.lunaroute.key_env LUNAROUTE_API_KEY
hermes config set providers.lunaroute.transport chat_completions
hermes config set LUNAROUTE_API_KEY lr_...

Then start Hermes and pick a model:

/model custom:lunaroute:<model>

Models auto-discover from /v1/models. Your default model is never changed.

NanoClaw and other OpenAI-compatible agents

NanoClaw has no dedicated adapter yet — configure it as a generic OpenAI-compatible provider:

lunaroute setup generic     # prints the base URL and the LUNAROUTE_API_KEY export

Use base URL https://gw.lunaroute.com/v1, your lr_ key, and any model id from Models.

Other harnesses

See CLI for lunaroute setup on coding agents, and Harness setup for wiring the MCP toolbelt into your client.