Multica – Open-Source AI Agent Team Collaboration Platform

1. What Is Multica

Multica is an open-source AI Agent team collaboration platform founded by former TikTok engineer Jiayuan. It turns coding agents such as Claude Code, Codex, and Kimi into first-class teammates on a kanban board. With a Linear-style interface, multi-workspace support, and local/cloud runtimes, it handles task assignment, autonomous execution, real-time progress reporting, and reusable skill accumulation. Multica addresses the industry pain point that "people's AIs can't collaborate together," enabling human–agent mixed teams to operate like traditional teams. Its vendor-neutral architecture avoids lock-in to a single model or agent.

Technical positioning and domain: Multica belongs to the AI Agent collaboration platform space, integrating multiple coding agents into unified kanban management for task flow and mixed-team coordination. Its distinctive stance treats agents as peers with human members—not mere tools or chat assistants.

Development background: Built independently by Jiayuan, drawing on large-company team and project management experience. Motivation: most agent tools stay in single-user chat mode, lacking multi-agent parallelism and team-level management.

Core value: Multica solves the lack of a unified collaboration framework for parallel multi-agent development. Kanban, assignment, and skill accumulation systematize and manage agent output. It imports project-management best practices—boards, issues, status flow—into agent collaboration for human–agent parity.

Technical characteristics: Vendor-neutral support for Claude Code, Codex, GitHub Copilot CLI, OpenClaw, OpenCode, Hermes, Gemini, Pi, Cursor Agent, Kimi, Kiro CLI, and more. WebSocket pushes real-time progress; reusable Skills let team capability compound with use instead of starting from zero each time.

2. Key Features

• Agents as formal teammates: Claude Code, Codex, Kimi, and others appear on the board with profiles and status. Assign issues like colleagues; agents sit alongside humans for equal footing.

• Autonomous execution and lifecycle: Full flow from queued → claimed → in progress → done/failed. Agents auto-claim assignments; WebSocket streams progress; humans can intervene anytime.

• Reusable Skills: Each run can capture solutions into team Skills for deploy, migration, code review, etc.—capability compounds; no rewriting configs from scratch.

• Unified runtime management: One dashboard for local daemons and cloud runtimes; auto-detects installed agent CLIs—no manual engine setup.

• Multi-workspace isolation: Separate spaces per team with own agents, issues, and permissions—parallel orgs/projects without collision.

• Low-interruption design: Issue/board/status language instead of IM-style chat that blows context—agents update status and report blockers; humans review at key gates—suited to many parallel agents.

• Vendor-neutral architecture: No single-model lock-in; pick the best agent per task from 10+ supported options.

3. How to Use

1. Requirements: macOS (recommended), Linux, or Windows. Install at least one coding agent CLI with network access. Homebrew or system package managers suggested.

2. Install CLI: macOS: brew install multica-ai/tap/multica. Other platforms: GitHub binaries or install scripts. Verify with multica --version.

3. One-shot setup: Run multica setup for config, browser auth, and local daemon start. Guides login and detects installed agent CLIs.

4. Verify runtime: Open Multica Web → Settings → Runtimes; confirm this machine is online. Add daemon address/port manually if needed.

5. Create agents: Settings → Agents → New Agent—pick runtime and provider (Claude Code, Kimi, etc.), name and tag. Dedicated agents for review, deploy, etc. recommended.

6. Assign and monitor: Create issues on the Board with description, priority, deadline; assign to an agent from members. Track progress on the board in real time.

Best practices: Start with simple tasks per agent; review Skills regularly; standardize frequent ops; monitor completion and block rates and adjust assignment.

4. Pros and Cons

5. Comparison with Similar Tools

Selection advice: For multi-person teams coordinating many coding agents, Multica is the best fit—vendor-neutral and Linear-like UX. Small teams or solo devs may prefer lighter Vibe Kanban. Fully autonomous "AI-native companies" may want Paperclip's org/budget/approvals at higher complexity. SWE-Kit suits deep agent customization without a ready-made collaboration UI.

6. Editor's Take

Multica's forward-looking idea treats agents as formal team members, not tools—shifting from "using tools" to "working with agents." Kanban, skill compounding, and vendor neutrality form a coherent human–agent framework—still rare in the industry.

It solves parallel multi-agent chaos: separate chat windows, no tracking, no knowledge capture. Boards and Skills make agent output manageable—ideal for AI-native teams running several coding agents.

Best for teams already on Claude Code, Codex, etc. who want agents in real PM workflows. Overkill for solo devs; strong for 3–10 person teams.

Openness and neutrality set up ecosystem growth. Enterprise governance (fine permissions, approvals, audit) could widen adoption.

Strong innovation and utility; enterprise features, docs, and community still maturing. Among the most worth-trying human–agent platforms for AI-native dev teams.

7. Use Cases

• Multi-agent parallel development: Claude Code, Codex, Kimi on daily PM and releases—agents claim, code, submit; humans review and merge.

• Agent capacity optimization: Monitor idle time and task state on the board; cut Agent Idle rate; tune assignment from completion, blocks, skill reuse.

• Human–agent mixed teams: One space to assign, track, and review humans and agents—humans architect/review; agents implement and test.

• Cloud-hosted runtimes: Distributed teams run agents in Multica cloud—browser-only access, no local CLI setup.

• Team skill accumulation: Turn review, deploy, migration into reusable Skills—new members and agents reuse them.

8. FAQ

Q: Which coding agents does Multica support?
A: Claude Code, Codex, GitHub Copilot CLI, OpenClaw, OpenCode, Hermes, Gemini, Pi, Cursor Agent, Kimi, Kiro CLI, and more. Vendor-neutral config can add new types without core code changes.

Q: Is Multica free?
A: Open source under MIT—self-host at no license cost. Cloud hosting (cloud-first) has its own pricing—see the website.

Q: How do I add a custom agent?
A: Via config and API in Settings → Agents → New Agent. Non-listed agents need adapter code.

Q: How do Skills work?
A: Reusable solutions from completed runs—steps, tools, outputs as templates. Later tasks invoke Skills; agents adapt to new context.

Q: Self-hosted deployment?
A: Yes—clone from GitHub and follow docs. Data and execution logs stay on your servers.

Q: Failures and blockers?
A: WebSocket updates in real time. Agents comment blockers; humans adjust or reassign. Board shows Blocked/Failed states.

9. Project Links

• Website: https://multica.ai/
• GitHub: https://github.com/multica-ai/multica