# Loom > Loom is a free, open-source (MIT), local-first workspace for coding agents. A daemon on your computer runs Claude Code, Codex, OpenCode, Grok, Antigravity, Kiro, ACP agents and plain model endpoints side by side, with one shared memory ("the brain"), parallel Orchestra runs across fleets of agents on different models, a desktop app, a terminal UI, a web UI and an Android/iPhone remote. No account, no keys of Loom's own. Install: `npm i -g @loompad/cli` (Node 22.5+), or the desktop app from https://loompad.tech/download. Try it with no agents or keys: `loom try`. ## Docs - [README](https://github.com/nickthelegend/loom/blob/main/README.md): everything Loom does, with commands - [Changelog](https://github.com/nickthelegend/loom/blob/main/CHANGELOG.md): what changed in each release - [Architecture](https://github.com/nickthelegend/loom/blob/main/ARCHITECTURE.md): daemon, adapters, brain, baton - [MCP server](https://github.com/nickthelegend/loom/blob/main/docs/mcp-server.md): Loom's own MCP tools - [Teams](https://github.com/nickthelegend/loom/blob/main/docs/teams.md): runners, hub, invites ## Skills for agents - [loom-skills](https://github.com/nickthelegend/loom-skills): eight Agent Skills (loom, loom-orchestrate, loom-chain, loom-agents, loom-automation, loom-collaborate, loom-ops, loom-mcp). Install with `loom skills install loom`. ## Core commands - `loom up` / `loom down` / `loom status`: start, stop and inspect the daemon - `loom new --agent claude-code@claude-opus-5-5x2 --agent codex@gpt-5.6x2`, `loom clone `, `loom init`: projects - `loom agents:add --model --count `: several sessions of one agent on one model - `loom send "" [--chat c] [--model m] [--effort e]`, `loom steer ""`, `loom btw ""` - `loom orchestra "" -w claude-code@claude-opus-5-5x2,codex@gpt-5.6x2`: plan, run in parallel worktrees, review, merge - `loom compare "" -w agent@model,...`: the same prompt to several agents/models, side by side - `loom scoreboard [--all]`: which model wins, at what cost - `loom intake`: start assigned GitHub/Linear issues as chats or goals - `loom brain:search `, `loom decision ""`: the shared memory - `loom mcp-server`, `loom connect `: Loom as an MCP server for your agents - `loom pair`: pair the phone app ## New in 2.1 - [Release notes](https://github.com/nickthelegend/loom/blob/main/docs/release-notes/v2.1.0.md) - `loom orchestra "" --plan-only`: show the plan and wait for approval before workers start - `loom recipes run `, `loom orchestra:graph `, `loom watch --run `: saved goals, task graphs, live runs - `loom send --wait --json` prints one JSON object; `loom send --stream` streams NDJSON (accepted, delta, event, outcome) - `loom models --health`, `loom models:effort`, `loom scoreboard --csv`: model health, default effort, standings - `loom brain:export --md `: the brain as Markdown - `loom pr:describe`, `loom git:stats`: PR descriptions from a chat, lines per agent - `loom audit --risky` / `--csv`: risky tool calls, safety denials and approvals - `loom backup --every 1d --keep 7`, `loom du`: scheduled backups, disk use per project - `loom alias set -- `, `loom export --format html`: your own commands, standalone chat exports ## Optional - [Download page](https://loompad.tech/download) - [Phone app](https://loompad.tech/mobile) - [Releases](https://github.com/nickthelegend/loom/releases)