Guides
Your key, your Conductor.
The fleet runs on your Claude login and never asks for an API key. The Conductor, the model that plans and supervises everything, runs on whichever key you bring.
Adding a key takes one paste
Open Settings, pick a provider, paste the key. That is the whole ceremony. Loom speaks to OpenAI, Anthropic, Google, xAI, Groq, Cerebras, DeepSeek, Mistral, OpenRouter, any OpenAI-compatible endpoint, and local servers through LM Studio, MLX, or Ollama.
- The key goes straight into your operating system's keychain. It is never written to a config file on disk and never leaves your machine except to call the provider you chose.
- Every outbound call travels through Loom's SSRF-guarded Rust HTTP proxy, so a Conductor request can only go where a Conductor request belongs.
- Switching Conductors is the same motion in reverse. Pick a different provider and model in Settings and the next plan comes from the new brain. The fleet does not change at all.
Why bring your own
Your provider, your prices, your call.
No markup
Loom is free and Apache-2.0. You pay your model provider directly, at the provider's own prices, with no reseller margin and no credits system in between.
Keychain only
Keys live in the OS keychain, full stop. Loom ships no telemetry and requires no account beyond your Claude login, so there is nowhere else for a key to go.
Swap anytime
Try a frontier model for a hard refactor, a cheaper one for routine sweeps, a local one for private work. Changing Conductors is a Settings change, not a migration.
Common questions
Keys, answered.
Does the fleet need an API key?
No. The fleet is six real Claude Code CLI sessions signed in with your existing Claude login. Only the Conductor takes a key. The full story is in The Claude login.
Can I run the Conductor with no key at all?
Yes. Point it at LM Studio, MLX, or Ollama and planning runs on your own hardware, with no key and no bill. See Local Conductors.
I want one key that covers many models. What then?
That is exactly what OpenRouter is for: a single key, models from many labs, switchable whenever you like.
Hand it the work.
Walk away.
macOS, Linux, and Windows. Around 13 MB. Free and open source.