Open source, self-hosted, cloud infrastructure

Run parallel AI agents in one repository.

AgentSwarm enables multiple agents to work on different tasks in parallel within the same repository. It runs in your own cloud infrastructure and supports both Codex and Claude in isolated Docker containers.

Get AgentSwarm View on GitHub 
# clone and start the stack
git clone https://github.com/coretracker/agentswarm.git
cd agentswarm
cp .env.example .env
./agentswarm.sh init
./agentswarm.sh start
Repository Shared codebase
Execution Parallel tasks
Runtime Docker
Infrastructure Own cloud

What it does

A control plane for parallel repository work.

The app is built around parallel task execution: select a repository, create separate work items, let multiple agents run in isolated containers, and decide which changes move forward.

Real repositories

Run multiple tasks against the same repository while preserving branch state, file context, and review boundaries.

Isolated execution

Codex and Claude run in isolated Docker containers so agent work stays separated from the infrastructure that coordinates it.

Reviewable output

Follow each task's logs, checkpoints, diffs, Git operations, and terminal output from the browser.

Tasks

Split repository work across agents.

AgentSwarm is designed for teams that want multiple agents working independently on different tasks inside the same repository.

build

Ask an agent to modify the repository in its own isolated workspace.

ask

Ask questions about the repository without writing files.

snippet

Reuse prompt templates with variables for recurring repository tasks.

automate

Use repository rules to trigger repeatable task creation automatically.

Runtime model

Agents run without owning your Git operations.

Agent containers do the coding work. The server coordinates parallel tasks, owns credentials, computes diffs, and exposes review controls for applying, reverting, pushing, or merging changes.

  1. 01

    Connect

    Add repositories, credentials, agent defaults, and snippets.

  2. 02

    Create

    Start build or ask work from prompts, snippets, or automation rules.

  3. 03

    Observe

    Watch logs, terminal output, checkpoints, and pending diffs for each agent.

  4. 04

    Decide

    Apply, reject, revert, push, or merge each task result from the review UI.

Get started

Run the stack in your own cloud.

Deploy AgentSwarm into your infrastructure, connect a repository, and create parallel tasks for agents to work through independently.

Clone from GitHub Read overview 
git clone https://github.com/coretracker/agentswarm.git
cd agentswarm
cp .env.example .env
./agentswarm.sh init
./agentswarm.sh start

open https://agentswarm.example.com/login