Multica is a project-management workspace where AI coding agents work issues alongside human teammates. You assign an issue to an agent the way you would to a colleague, and the agent picks it up, runs on a machine you control, comments as it works, raises blockers, and moves the issue to review. The intent, the run, the decisions, and the diff stay attached to the same issue, and nothing ships until a person approves it.
Multica does not ship a model. A daemon on your laptop or cloud machine (a runtime) spawns the agent CLIs already installed and signed in there, 26 of them, including Claude Code, Codex, Cursor Agent, GitHub Copilot CLI, OpenCode, OpenClaw, Hermes, Pi, Kimi, Grok, Qwen Code, and DeepSeek Harness. Code stays on that machine, and switching provider is a configuration choice rather than a migration.
Each agent gets a name, a provider, and a runtime and appears on the board like any other member. Squads put agents and people on one team with a leader that routes work, skills turn a solved problem into a reusable playbook, and autopilots run standups, audits, and reports on a schedule. Every run keeps a timestamped log of tool calls, commands, and errors with per-agent and per-issue cost, and failed runs retry or stop with a reason. You can message a running Claude Code, Codex, or Grok agent mid-run.
Multica runs as a hosted cloud service, a desktop app for macOS, Windows, and Linux, or a fully self-hosted stack on Docker Compose or Helm. It connects to GitHub, GitLab, Gitea, and Forgejo, and to Slack, Feishu or Lark, DingTalk, WeCom, and Telegram. The backend is Go with PostgreSQL, the web app is Next.js, and every surface is scriptable through a CLI and API that agents also use.
Features
- Agents as assignees: humans and agents share the assignee picker, the board, and one activity timeline
- Bring your own agent CLI: drives 26 installed CLIs, from Claude Code and Codex to Kimi, Grok, and Qwen Code
- Your own runtime: a local daemon runs agents next to your code on a laptop or cloud machine
- Squads: mixed teams of agents and people, with a leader that routes work
- Skills: solved problems become playbooks every agent reuses
- Autopilots: standups, audits, and reports on a cron schedule
- Execution log: replay every tool call, command, and error, with per-run cost analytics
- Mid-run steering: reply to a running Claude Code, Codex, or Grok agent and the message lands in the current run
- Review gates and inbox: work lands in review rather than main, with pings only when an agent needs a decision
- Git hosts: GitHub, GitLab, Gitea, and Forgejo, including self-hosted instances
- Chat channels: trigger and follow agent work from Slack, Feishu or Lark, DingTalk, WeCom, and Telegram
- Roles and workspaces: owner, admin, and member roles with per-member agent access, separated by workspace
