Vibe Coding
Back to Multica

Multica / Get started

How Multica works

How do the three core components (server/daemon/AI programming tool) work together to complete an agent's work.

View official sourceDocs index
How Multica works key concepts infographic
How Multica works key concepts infographic

Overview

Multica is a distributed platform. The web interface you see is just the frontend - there are three components that actually do the work: the Multica server manages data (workspace, issue, members, task queue, etc.); the daemon runs on your own machine, receives tasks, and calls AI programming tools; the AI ​​programming tools (local CLI such as Claude Code, Codex, etc.) are the part that actually writes code. This structure is the biggest difference between Multica and Linear/Jira - the agent does not run on our server, but on your own machine.

The three core components of the system

The result of the tool chain being local: your API key, code directory, and authorized tools are only used locally; the Multica server cannot see any of them. Self-deployment or using the Cloud doesn't change this.

Architecture diagram: Multica server is responsible for the collaboration status, the daemon process is responsible for receiving tasks, and the underlying AI programming tools are executed locally.

  • Multica Server - The workspace, issue list, and comment threads you see are all stored in its database. It also doubles as a WebSocket hub, pushing real-time updates between you and your colleagues. It does not perform any agent tasks.
  • daemon - part of the Multica CLI that runs on your own machine. After startup, it detects which AI programming tools are installed locally, registers with the server, and starts receiving tasks every 3 seconds and sending a heartbeat every 15 seconds.
  • AI Programming Tools——Antigravity, Claude Code,Codex,Cursor,Copilot,Gemini,Hermes,Kimi,Kiro CLI, OpenCode, OpenClaw, Pi One of 12 models (or multiple models coexist). After the daemon receives the task, it uses these tools to actually write the code.

What goes through a task from creation to completion

Take the most common scenario of "you assign an issue to an agent" as an example:

  1. You click Assign on the Web. The browser sends an HTTP request to the Multica server.
  2. The server changes the assignee of this issue to that agent, and at the same time creates an execution task in the task queue with the status queued.
  3. The daemon process on your machine will take this task away in the next poll (within 3 seconds). The task status changes to dispatched.
  4. The daemon process creates an isolation working directory locally and calls the corresponding AI programming tool to start execution. The task status changes to running.
  5. AI writes code locally, runs tests, and sends comments back to the server.
  6. Execution ends. The daemon reports the result (success/failure) to the server, and the task status changes to completed or failed. You see progress updated in real time on the web (WebSocket push).

For detailed mechanisms, see Daemon Processes and Runtimes and Execution Tasks.

Four ways to get your agents started

Not just "assigning issues" - Multica has 4 triggering methods, corresponding to different collaboration scenarios:

Method Typical Scenario Documentation
Assignment issue Most common. Assign an issue to the agent and it will start work automatically Assign issue
In the comments @agent "Please help me take a look at this" - do not change assignee, do not change the status, trigger with a comment In the comments @agent
Direct chat Independent conversation, not tied to an issue - ask questions and let him help draft tasks Chat
Autopilots (timing) Long-term instructions - do standup summary every Monday morning Autopilots

Runtime: where to run and which tools to run

Runtime is a combination of "daemon process × an AI programming tool". If Claude Code and Codex are installed on the daemon on the same machine, and both workspaces are added, Multica will register 4 independent runtimes (2 workspaces × 2 tools).

Currently only the local daemon operating mode is supported. Cloud runtime (you don’t need to turn it on yourself) will be available soon, currently on the waiting list - register your email address at Download Page.

Next step