Key Points
- What Multica is: An open-source management platform (Apache 2.0) for AI coding agents, treating agents as first-class teammates rather than personal CLI tools. Repository: https://github.com/multica-ai/multica
- Positioning: A coordination layer for human + AI hybrid teams, not a coding agent itself. Output quality still depends on the underlying CLI (Claude Code, Codex, OpenClaw, etc.).
- Core features: agents as teammates with profiles and assignments, autonomous task lifecycle (enqueue → claim → start → complete/fail) with WebSocket streaming, reusable Skills system, unified runtimes (local daemon + cloud), and multi-workspace isolation.
- Vendor-neutral agent support (11 CLIs): Claude Code, Codex, GitHub Copilot CLI, OpenClaw, OpenCode, Hermes, Gemini, Pi, Cursor Agent, Kimi, Kiro CLI.
- Tech stack: Next.js 16 (App Router) frontend, Go backend (Chi router, sqlc, gorilla/websocket), PostgreSQL 17 with pgvector, local agent daemon, Electron desktop client.
- Agent daemon mechanics: registers installed CLIs per workspace, polls every 3s (configurable), spawns agents in isolated directories, streams results, sends 15s heartbeats, and supports three garbage-collection modes (full cleanup after TTL, orphan cleanup >72h, artifact cleanup for open issues idle >12h).
- Deployment:
multica setupfor cloud; self-host via install script ormake selfhost(Docker Compose). Default ports: frontend :3000, API :8080. Install via Homebrew, curl/PowerShell scripts, or source. - MCP ecosystem: Community server (https://github.com/Korkyzer/multica-mcp) exposes 27 tools covering issues, agents, projects, runtimes, and workspaces, enabling AI-native orchestration from Claude Desktop or Codex Desktop.
- Competitive landscape: Compared with Devin (closed-source, $500/mo Teams), OpenHands, SWE-agent, and Paperclip. Multica differentiates on multi-agent team coordination, project management, skill reuse, and self-hosting.
- Anti-hype audit: "Your next 10 hires won't be human" is rhetorical; Multica does not raise individual agent intelligence. Real risks include pgvector hosting constraints, CLI API breakage, Electron maintenance overhead, and potential competition from Anthropic's Managed Agents.
- Best fit: 2–5 person teams already using multiple coding agents, with needs for skill reuse, self-hosting (compliance/privacy), and AI workflow automation.
- Not a fit: solo developers (CLI is simpler), teams with zero AI experience, SWE-bench score chasers, and users expecting one-click project generation.
- Significance: Signals a rising "agent management layer" category above individual coding agents; embodies a human-AI collaboration model where agents appear on kanban boards, post updates, and accumulate organizational knowledge.
- Growth metrics (as of 2026-04-17): 15,400+ GitHub stars, top-3,300 global rank, GitHub Trending #1 (TypeScript) on 2026-04-12, peak 1,680 stars/day.