Multica — Open-Source Managed Agents Platform

Introduction

Multica is an open-source platform that lets teams deploy and manage AI coding agents as persistent collaborators rather than one-shot tools. It provides a unified interface for assigning tasks, tracking progress, and reviewing agent output across multiple concurrent agents working on the same codebase.

What Multica Does

• Orchestrates multiple AI coding agents working on a shared repository
• Provides a dashboard for assigning, monitoring, and reviewing agent work
• Handles context sharing and conflict resolution between agents
• Supports role-based agent configurations (reviewer, implementer, tester)
• Integrates with GitHub for pull request workflows and CI feedback

Architecture Overview

Multica runs a coordination server that maintains a shared project state and dispatches tasks to agent workers. Each agent worker runs in an isolated environment with its own context window and tool access. The coordinator manages task queues, merges partial results, and surfaces conflicts through a web dashboard built with Next.js and TypeScript.

Self-Hosting & Configuration

• Requires Node.js 20+ and a PostgreSQL database for state persistence
• Configure agent provider credentials (Anthropic, OpenAI) via environment variables
• Set repository access tokens for GitHub integration
• Adjust concurrency limits and context budgets in the config file
• Deploy with Docker Compose for a production-ready setup

Key Features

• Multi-agent collaboration on a single codebase with conflict detection
• Persistent agent sessions that survive restarts and context resets
• Built-in code review pipeline where agents review each other's output
• Real-time dashboard showing agent activity, token usage, and task status
• Plugin system for custom agent behaviors and tool integrations

Comparison with Similar Tools

• CrewAI — framework-level multi-agent orchestration; Multica provides a full managed platform with UI
• AutoGen — research-focused conversation framework; Multica targets production team workflows
• Claude Code subagents — built-in but ephemeral; Multica adds persistence and coordination
• Devin — proprietary AI engineer; Multica is open-source and self-hostable

FAQ

Q: Can I use different LLM providers for different agents? A: Yes. Each agent role can be configured with its own model and provider.

Q: Does it support private repositories? A: Yes. Configure repository access via SSH keys or personal access tokens.

Q: How does it handle merge conflicts between agents? A: The coordinator detects overlapping file changes and either serializes the tasks or flags them for human review.

Q: What is the minimum hardware requirement? A: A machine with 4 GB RAM is sufficient for the coordinator; agents run API calls to external LLM providers.

Sources

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