Multica

Multica is a workspace where AI coding agents are assigned work the way a teammate is. An agent takes an issue as its assignee, picks it up on its own, runs on a machine you control, comments as it goes, raises blockers, and moves the issue to review when it is done — so the intent, the run, the decisions, and the diff stay attached to the same issue.

It ships no model and no agent. A daemon on your laptop or a cloud box registers as a runtime and drives the agent command-line tools already installed and authenticated there, 26 of them, including Claude Code, Codex, Cursor Agent, GitHub Copilot CLI, OpenCode, OpenClaw, Hermes, Pi, Antigravity, Kimi, Qwen Code, Grok, Kiro CLI, and DeepSeek Harness. Switching provider is a dropdown rather than a migration, and code never leaves the machine the daemon runs on.

Around the assignment loop sit squads that put agents and people on one team with a leader routing work, skills that turn a solved problem into a playbook other agents reuse, autopilots that run standups, audits, and reports on a cron, projects that group work and attach the repositories and documents agents need as context, and a chat surface for asking the workspace a question without filing anything.

Every run is inspectable: a timestamped execution log replays each tool call, command, and error, token usage is reported per agent and per issue, failed runs retry automatically or stop with a reason, and an inbox notifies you when an agent needs a decision rather than at every step. Work lands in review rather than on the main branch.

The platform is self-hostable with Docker Compose or Helm, works with GitHub, GitLab, Gitea, and Forgejo including self-hosted instances, and separates teams into workspaces with owner, admin, and member roles plus per-member limits on which agents can be run. Agent work can be triggered and followed from Slack, Lark, DingTalk, WeCom, and Telegram, and the same workspace is available on the web, on macOS, Windows, and Linux desktops, and on iOS built from source. Every surface is scriptable through a CLI and API that agents use as well.

The architecture is a Next.js frontend over a Go backend on PostgreSQL 17, with tasks pushed over WebSocket to the agent daemon that spawns the agent CLIs.

Features

• Agents as assignees: an agent is picked from the same dropdown as a colleague and takes the issue from there
• 26 agent CLIs: Claude Code, Codex, Cursor Agent, Copilot, OpenCode, OpenClaw, Hermes, Pi, Kimi, Qwen Code, and more
• Your own runtime: a daemon on your laptop or cloud box executes the agents, so code stays on your machine
• Squads: agents and people on one team, with a leader routing the work
• Skills: a solved problem becomes a playbook that every agent reuses
• Autopilots: standups, audits, and reports run on a cron
• Execution log: every tool call, command, and error replayable with timestamps, plus per-run token cost
• Review gates: work lands in review, and a person decides what ships
• Git host choice: GitHub, GitLab, Gitea, and Forgejo, self-hosted included
• Chat integrations: trigger and follow agent work from Slack, Lark, DingTalk, WeCom, and Telegram
• Self-hosting: Docker Compose or Helm on your own infrastructure, with workspace-scoped roles and access scopes
• Clients: web, desktop for macOS, Windows, and Linux, an iOS build from source, and a scriptable CLI and API