multica-ai/multica: Make humans and AI agents work as one team — open-source and self-hostable.

Make humans and AI agents work as one team — open-source and self-hostable. - multica-ai/multica

Summary

Multica is a self-hostable workspace that puts AI coding agents on the same board as your human teammates. It drives existing agent CLIs (Claude Code, Codex, Cursor, Kimi, and others) through a local daemon, so agents pick up assigned work, report progress, and hand results back for human review.

What it is useful for

Teams already running one or more agent CLIs and wanting a single place to coordinate them. Multica turns each agent into an assignable teammate, ties the intent, run log, and diff to one issue, and keeps code on a runtime you control. It fits orchestrating multiple heterogeneous agents through one queue, enforcing review gates before code reaches the default branch, replaying what an agent actually ran, scheduling recurring work via "Autopilots", and routing notifications into Slack, Feishu/Lark, DingTalk, WeCom, or Telegram. Self-hosting fits teams whose source must not leave their infrastructure; GitHub, GitLab, Gitea, and Forgejo are all supported.

How engineers can use it

Three documented paths:

• Cloud — sign up at multica.ai.
• Desktop — download Multica Desktop for macOS, Windows, or Linux. The machine it runs on auto-registers as a runtime and detects installed agent CLIs.
• Self-host — curl -fsSL https://raw.githubusercontent.com/multica-ai/multica/main/scripts/install.sh | bash -s -- --with-server then multica setup self-host. Requires Docker; on Windows use the PowerShell installer with $env:MULTICA_MODE="with-server". Fall back to make selfhost-build from a checkout if the GHCR tag is not yet published.

The prerequisite is at least one supported agent CLI installed and signed in on the runtime machine — Multica drives them, it does not ship them. The basic flow: connect a runtime, create an agent, assign it an issue. For local development the README lists Node.js 22, pnpm 10.28.2, Go 1.26.6, and Docker, with make dev bringing up the full stack.

Documented caveats: the iOS app builds from source and is not yet on the App Store; Feishu/Lark is open to mainland-China Feishu only; DingTalk, WeCom, and Telegram are community-maintained; self-hosted installs send one anonymous deployment snapshot per day unless you set DO_NOT_TRACK=1 on the API server.