Multica: A 46K‑Star Open‑Source Hub for Human‑AI Collaborative Development

What Multica Is

Multica is an open‑source collaborative workspace launched on GitHub in January, now with 46 000 stars. It aggregates local AI coding tools and human teammates into a single Kanban‑style board, similar to Linear or GitHub Projects, but with agents displayed as "people".

Core Features

Adding Agents

Multica does not provide its own models; it drives any installed, logged‑in CLI AI tool. Over twenty agents are supported, including Claude Code, Codex, Cursor, Copilot, Kimi CLI, Qwen Code, DeepSeek, Trae, CodeBuddy, and MiniMax Code. Users select agents via the UI without migration effort.

Agent Representation

Each agent can be named, assigned a service provider, and bound to a runtime machine, appearing in the board’s member list where it can be assigned tasks, @‑mentioned, and commented on just like a human colleague.

Squads

Multiple agents and humans can be grouped into a squad with a designated leader who automatically distributes tasks, reducing manual assignment when workload spikes.

Skills

Resolved problems can be saved as reusable "skills" that any agent in the team can apply, avoiding repeated instruction.

Runtime Control

Agents execute on a user‑specified runtime—either a local notebook or a cloud machine—by running a daemon; code never leaves that machine.

Task Management

Assigning Work

Creating an issue and setting an agent as the assignee dispatches the task; the agent claims it, processes a brief natural‑language description, and delivers a pull request.

Autopilot

Repeating tasks such as meeting minutes, audits, or weekly reports can be scheduled as timed jobs that run automatically.

Chat Interaction

The workspace includes a chat interface where users can ask questions or issue a one‑line command to start work without creating an issue.

Projects

Related work can be grouped into a project that bundles repositories and documentation, giving agents immediate access to necessary context.

Observability

Execution Logs

Every tool call, command, and error is timestamped, allowing full replay to pinpoint failures.

Token Accounting

Token usage and cost are tracked per agent and per task, highlighting the most expensive agents at a glance.

Human Review

Completed work lands in a "Ready for Review" column; humans decide whether and when to merge.

Selective Notifications

Agents send messages to the inbox only when human approval is required; failed jobs automatically retry and, if still failing, stop with a clear reason.

Deployment and Integration

Self‑Hosting

Multica can be run on private infrastructure via Docker Compose or Helm; all data, code, and agent runtimes remain under user control.

Git and IM Connectors

Supports GitHub, GitLab, Gitea, Forgejo, and self‑hosted Git services. IM integrations include Slack, Feishu, DingTalk, and WeChat Work, enabling task triggers and progress tracking directly from chat groups.

Permissions and Multi‑Workspace

Workspaces are scoped by team with owner, admin, and member roles, and fine‑grained control over which agents each member may use.

Clients

Web, macOS, Windows, Linux, and iPhone clients are available, plus a CLI and API for scripting any UI action.

Installation Experience

Using the hosted version at multica.ai requires only registration. The desktop client registers the current machine as a runtime and auto‑detects installed agents. For self‑hosting, two commands suffice:

curl -fsSL https://raw.githubusercontent.com/multica-ai/multica/main/scripts/install.sh | bash -s -- --with-server
multica setup self-host

The scripts assume Docker is present and pull official images from GHCR; Windows users can run the PowerShell variant.

Who Should Use Multica

Individual developers juggling multiple AI tools who want to avoid repeatedly re‑explaining context.

Small teams with repetitive chores (release notes, document translation, feedback aggregation, feature‑flag cleanup) that can be delegated to agents.

Organizations handling sensitive data that require self‑hosted execution, full logs, and strict permission controls.

Teams already using chat platforms like Feishu or DingTalk, enabling task dispatch directly from those channels.

Author’s Perspective

The author finds Multica’s positioning compelling: while many projects focus on improving an agent’s coding ability, Multica addresses the orchestration problem of managing several agents, tracking progress, and ensuring human oversight. It reuses proven software‑team workflows (board, claim, report, review, merge) with agents as participants, minimizing learning curve. Self‑hosting and on‑premise execution are highlighted as crucial for corporate adoption. The extensive Chinese documentation is praised as matching the project’s star count.

https://github.com/multica-ai/multica