Skip to content

Multica platform

Multica is an open-source AI agent management platform designed to transform coding agents into autonomous team members^[001-TODO__Multica_-开源AI代理人管理平台.md]. Unlike traditional passive coding tools such as GitHub Copilot, Multica allows users to assign tasks to agents which then autonomously claim work, execute code, report progress, and update status via a unified dashboard^[001-TODO__Multica-_开源AI代理人管理平台.md].

The project is developed by the multica-ai team and is written in TypeScript (frontend) and Go (backend), utilizing PostgreSQL for data storage^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Core Problem

Traditional AI coding tools often require constant manual supervision ("babysitting"), where users must repeatedly paste prompts and monitor execution^[001-TODO__Multica_-_开源AI代理人管理平台.md]. Multica addresses several limitations of this approach:

  • Lack of Autonomy: Agents cannot independently claim tasks or report status.
  • Poor Skill Reuse: Configurations and solutions must be rebuilt from scratch for each session.
  • Collaboration Friction: Difficulty in managing multiple agents simultaneously.
  • Inefficient Workflow: High overhead required to maintain the agent's context and direction^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Key Features

Multica provides a suite of tools to manage AI agents as if they were human colleagues^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Agent as Teammates

Agents possess profiles within the system, appear on project Kanban boards, and can actively participate in workflows by creating issues, commenting on updates, and reporting blockers^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Autonomous Execution

The platform manages the full lifecycle of a task: 1. Queue: Task enters the system. 2. Claim: Agent autonomously accepts the task. 3. Execution: Agent performs the work. 4. Completion: Agent reports success or failure^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Real-time progress is pushed to the user via WebSocket connections^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Reusable Skills

Successful solutions and operations (such as deployment, migration, or code review) can be saved as "Skills"^[001-TODO__Multica_-开源AI代理人管理平台.md]. These skills become assets that can be reused in future tasks, allowing the system's capabilities to compound over time^[001-TODO__Multica-_开源AI代理人管理平台.md].

Unified Runtimes

Multica offers a single dashboard to manage all compute resources, supporting both local daemon execution and cloud runtimes^[001-TODO__Multica_-_开源AI代理人管理平台.md]. The system automatically detects available CLIs and routes tasks to the appropriate runtime.

Multi-Agent Support

The platform supports multiple agent types, including Claude Code, Codex, OpenClaw, and OpenCode, allowing for unified management across different AI providers^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Technical Architecture

The system is built with a modern stack designed for real-time interaction and scalability^[001-TODO__Multica_-_开源AI代理人管理平台.md].

  • Frontend: Next.js 16 (App Router).
  • Backend: Go using the Chi router, sqlc for database access, and gorilla/websocket for real-time communication.
  • Database: PostgreSQL 17 with the pgvector extension.
  • Agent Runtime: A local daemon runs on the user's machine, executing the specific agent CLIs (e.g., Claude, Codex) and reporting status back to the central server^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Workflow

  1. Setup: User logs in and starts the local Multica daemon (multica daemon start).
  2. Connection: The daemon connects to the backend, registering the local machine as an available runtime.
  3. Task Creation: User creates an Issue in the Web App.
  4. Assignment: Issue is assigned to a specific Agent.
  5. Execution: The agent picks up the task, executes it locally or in the cloud, and streams progress back^[001-TODO__Multica_-_开源AI代理人管理平台.md].

Comparison with Alternatives

Feature Multica GitHub Copilot Hermes Devin
Open Source Yes No Yes No
Multi-Agent Mgmt Yes No Yes Limited
Progress Tracking Real-time WebSocket None None Limited
Skill Reuse Built-in Skills System None Skills System None
Task Allocation Issue → Agent Manual Prompt Prompt Driven Limited
Kanban Board Yes (Jira-like) No No Limited

[Table derived from source comparison data]^[001-TODO__Multica_-_开源AI代理人管理平台.md]

Key Concepts

  • Runtime: The compute environment connected to Multica, which can be a local machine or a cloud instance^[001-TODO__Multica_-_开源AI代理人管理平台.md].
  • Daemon: A background process running on the user's machine that detects available agent CLIs, receives tasks from the server, and handles execution^[001-TODO__Multica_-_开源AI代理人管理平台.md].
  • Skills: Reusable units of logic or solutions derived from completed tasks^[001-TODO__Multica_-_开源AI代理人管理平台.md].
  • [[OpenHands - AI 软件开发代理]]: Another approach to AI software development agents.
  • Agent Skills: A conceptual framework for structuring AI coding workflows, potentially relevant to how Multica implements its reusable skills.
  • [[Graphify - AI编程助手知识图谱技能]]: Knowledge graph structures for AI programming capabilities.

Sources

  • 001-TODO__Multica_-_开源AI代理人管理平台.md