Open Beta: Pro unlocks automatically when you sign upDownload Free
CodeAgentSwarm Logo
CODEAGENTSWARM

One workspace · Your coding agents

Parallel coding agents.
One visual command center.

Download free

Pro included during beta · macOS 12+ · Windows 10/11 · Linux

Works with
Claude CodeCodex CLIAntigravity CLIOpenCodeKimi CodeGrok BuildCursor AgentMuse CodePiDevin CLI

Now in 2.4.0: Muse Code, Pi and Devin CLI.

Parallel AI Agents

Multiple coding agents working side by side

Search Everything

Full conversation history, never lose context

Real-Time Visibility

Live diffs and change tracking

Agentic Task Board

Tasks your agents read, update, and complete

The Hidden Cost

Window switching is wasting your time

When your workflow depends on one AI agent at a time, the mental tax compounds. Switching windows, losing threads and repeating prompts quietly drains your time.

WithoutCODEAGENTSWARM

Window switching is wasting your time

Constant Alt+Tab between windows
Claude changes code without you seeing
Lost context, repeated prompts
auth-service: Visual Studio Code
// where was I?...
export async function refreshToken(req) {
  const session = await getSession(req)
  if (!session) throw new AuthError('expired')
  // TODO: did Claude already change this?
  return rotate(session.token)
}
localhost:3000/api/users

500 Internal Server Error

Something broke while you were in another window.

claude: ~/dev/mobile
✳ Claude Code · session expired 2h ago
claude: ~/dev/api
 ▐▛███▜▌
▝▜█████▛▘
  ▘▘ ▝▝Claude Code v2.1.211
Fable 5 · Claude Max
~/dev/api
⠧ Waiting for your approval… (14 min)
  ❯ 1. Yes   2. Yes, don’t ask again
codex: ~/dev/web
>_ OpenAI Codex (v0.48.0)
model: gpt-5.2-codex · ~/dev/web

> fix the login redirect…
■ session expired. context lost
⌘⇥ … again … and again23112
Claude finished 12 min agoYou never saw it.
Constant frustration
CodeAgentSwarm
WithCODEAGENTSWARM

All visible at once · Multiple agents, 1 view

Multiple agents visible at once
Real-time diffs of every change
Agents that create and complete tasks autonomously
Total control

Your agents got faster.
Now your workflow can too.

Every agent, every diff, every notification in one view. Nothing to hunt, nothing to miss.

Try it freeFree. No credit card. macOS, Windows and Linux.
Core Features

Everything You Need

One command center for all your AI agents. Run them in parallel, track their work, and never lose context again.

Multiple AI Agents in Parallel

Each agent has its own project, context and task. Work on the frontend while another agent handles the backend, without losing track of either.

View guide

Agent Task Board

Create tasks, assign them to agents and review the changes when they finish. Agents read the board via MCP and update their progress automatically.

Real-Time AI CLI Notifications

Stop babysitting agent windows. When an agent finishes, needs input or fails, CodeAgentSwarm alerts you so you can focus elsewhere.

View guide

Real-Time Visibility & Change Tracking

Live diffs while Claude edits. Dynamic titles showing what each agent is doing. Know exactly what changed without checking git afterwards.

View guide

Full Conversation History

Every decision, every explanation, every snippet - searchable. Recover context from 3 weeks ago in seconds. Never repeat a prompt because it 'got lost'.

View guide

Turbo Mode (Skip Permissions)

Skip confirmation prompts for routine operations but maintain guardrails for sensitive ones. Block pushes to protected branches, prevent destructive deletes, limit sensitive commands. You set the limits.

View guide
New in 2.4.0NEW

Your agents now run the team

Give one agent the goal. It opens the sessions, keeps the board moving and stays inside your budget.

01 · Coordinators

One agent plans. The rest get to work.

Start a coordinator for one project or for all of them. Tell it what you need. It splits the work, opens worker sessions in the background and reports back in the same Chat.

  • Plans across one project or every project
  • Opens each session with the agent that fits the job
  • Reports back when the work is done
View guide
Try a goal
Global coordinator
Opus 5.5BuildManual
02 · Kanban on autopilot

Your backlog moves while you don’t.

Queue a task with the lightning button and pick who does it: the agent, the model and the workspace. Auto starts each task on its own, as many at once as you allow, and the finished work waits for you in testing.

  • Each task can run with a different agent and model
  • Auto opens the session and hands over the task
  • Watch every task work live, then review it in testing
View guide
Your turn: queue task #4 with the lightning button
KANBANSearch tasks...Project:All ProjectsCreate ProjectAdd TaskRefresh
PENDING0
AUTO0GPT-5.5
IN PROGRESS0
IN TESTING0
GPT-5.5highAsk for approval
03 · A daily budget for every agent

Spend your quota by the day, not by lunch.

Set a daily limit for each provider, or let Smart pace spread what is left of your window across the days. Choose what happens at the limit. Need a little more? Add 5% for today.

  • A daily cap per provider, or Smart pace
  • At the limit: notify you, pause new work or stop
  • One click for an extra 5% today
View guide
Settings › Providers
Usage todaySMART PACE08:00

See each feature in action

Multi-agent control, multiple models, conversation history, kanban, Git and notifications.

Multi-Agent Control

Manage multiple AI agents simultaneously

Works with

Run a swarm of AI coding agents

One workspace for Claude Code, Codex CLI, Antigravity CLI, OpenCode, Kimi Code, Grok Build, Cursor Agent, Muse Code, Pi and Devin CLI. Run them in parallel, mix vendors, and watch all of them from one place.

See how an AI CLI agent swarm works
One-Click Integrations

Connect Everything in One Click

A catalog of integrations you enable per project. Your AI agent reads GitHub PRs, creates Jira tickets and updates Notion docs without leaving its workflow.

Click any integration to install

Database & Storage

Supabase, PostgreSQL, Google Drive - all your data sources in one place.

Browser Automation

Playwright and Puppeteer for testing, scraping, and automation.

Team Communication

Slack, Notion, GitHub - keep your team in sync automatically.

Bring your own sub

Works with the subscription you already have.

The agents you already pay for, running on your machine with your credentials. We just put them in one window.

Claude Code
claude
Codex CLI
codex
Antigravity CLI
agy
opencode
opencode
Kimi Code
kimi
Grok Build
grok
Cursor Agent
cursor-agent
Muse Code
muse
Pi
pi
Devin CLI
devin
Your plan's limits, not ours.
Mix vendors in the same swarm.
CodeAgentSwarm never receives or stores your code.

Open Beta

The complete product is free during beta.

€0for the entire open beta

Download the app and create an account. Pro activates automatically, with no CodeAgentSwarm subscription or credit card.

Download the Free Beta

Run

50 AI agents

Work in parallel

  • Grid + Tabs
  • Resizable Agent Panels
  • Live notifications and changes

Organize

Unlimited projects

Keep context separate

  • 6 Project Shortcuts
  • Task Labels
  • Conversation History
View all 17 included Pro features
  • 50 AI agents
  • Unlimited projects
  • Live Notifications
  • Real-time Agent Changes
  • Grid + Tabs
  • 6 Project Shortcuts
  • Unlimited Agent Shortcuts
  • Resizable Agent Panels
  • MCP Configuration
  • MCP Marketplace
  • Git Integration & AI Commits
  • Task Labels
  • Keyboard Shortcuts
  • Manage MCP Permissions
  • Manage AI CLI Permissions
  • Turbo Mode
  • Conversation History

Product Roadmap

Our vision for the future of AI-powered development

3 of 4 shipped·on track for 2026
Q12026
✓

Multi-Agent Integration

✓ Shipped

Expand beyond Claude to support multiple AI CLI agents.

Codex CLI integrationAntigravity CLI compatibilityopencode supportUnified agent orchestration
Q22026
✓

Windows Support

✓ Shipped

CodeAgentSwarm runs natively on Windows.

Full Windows 10/11 supportNative installer with auto-updatesWindows on ARM builds
Q32026
✓

Linux Support

✓ Shipped

Bringing CodeAgentSwarm to every developer's desktop.

Ubuntu, Debian & Fedora buildsAppImage & .deb packagesSame agents, Chat and Kanban as on macOS
Q42026

Autonomous Mode

Planned

You define the goal: the swarm plans, builds and ships it.

AI-powered task decompositionSmart task prioritizationAutonomous workflow executionCross-agent collaboration

Community

Join the people shaping CodeAgentSwarm.

See what's coming, report what's broken, vote on what's next, and trade real workflows with other builders.

  • Sneak peeks

    See new features while they're still taking shape.

  • Direct bug reports

    Put issues in front of the team, with the context we need to fix them.

  • Roadmap votes

    Upvote what matters and influence what we build next.

  • Real workflows

    See how other builders run their agents, then share what works for you.

Join the CodeAgentSwarm Discord

Free to join · Lurk, ask, share, or vote.

Got Questions?

Frequently Asked Questions

Open CodeAgentSwarm, choose a project and launch an agent for each task. Sessions run independently, so you can work on different tasks or projects in parallel while seeing their status in one place. Use separate Git worktrees when sessions need isolated working copies of the same repository.

CodeAgentSwarm keeps each project's agents, task board and conversation history together. You can switch between codebases without losing context or rebuilding your setup.

Yes. CodeAgentSwarm shows live file changes and session status, so you can follow what each agent is doing and review its changes. Review the diff and run the relevant checks before committing or merging.

Yes. CodeAgentSwarm sends a desktop notification when any agent finishes a task, needs your input or encounters an error.

CodeAgentSwarm saves the full conversation history of every supported CLI across sessions. Search any past chat, decision, or code snippet instantly. Session history is fully searchable, so you never lose context or have to repeat explanations.

Yes. Visual permission settings control the files and actions available to each agent. You can approve sensitive operations while Turbo Mode handles the safe routine ones within the limits you set.

CodeAgentSwarm includes an MCP marketplace for configuring integrations such as GitHub, Notion, Supabase and Playwright. Available tools and configuration options depend on the selected agent and MCP server.

Yes. CodeAgentSwarm keeps searchable conversation history across agent sessions, shows each agent's changes in real time and ties every session to its project. Past work stays easy to find and resume.

tmux is a good fit when you want terminal panes and are comfortable configuring your own workflow. CodeAgentSwarm adds project tasks, session status, notifications, searchable conversation history and file-change review in a desktop app. Choose it when you want those controls together or use agents from several providers.

Yes. Each agent session is an independent process, so Claude Code and Codex CLI can run at the same time. Pick the agent for each session and run as many as your plan allows.

Yes. CodeAgentSwarm runs Antigravity CLI (agy), Google's official successor to the retired Gemini CLI, alongside the other supported agents, with shared notifications, searchable history and live diffs for every agent.

Yes. A single CodeAgentSwarm workspace can run Claude Code, Codex CLI, Antigravity CLI, OpenCode, Kimi Code, Grok Build, Cursor Agent, Muse Code, Pi and Devin CLI side by side. Use the best agent for each task and supervise all of them from one place.

An AI CLI agent swarm is several independent AI coding CLI agents running in parallel in one place, with shared visibility, notifications and history. CodeAgentSwarm is the desktop workspace built to run and supervise that swarm.

Yes. CodeAgentSwarm runs OpenCode alongside the other supported agents. Add one or several OpenCode sessions to the same swarm.

Yes. Kimi Code CLI by Moonshot is a first-class agent in CodeAgentSwarm, alongside the other supported agents. Pick Kimi Code from the launcher like any other agent and it gets the same notifications, searchable conversation history and live diffs. You can run a full swarm of Kimi Code sessions or mix it with the other CLIs.

An Agentic Development Environment is a workspace built around delegating software tasks to AI coding agents, running them in isolated or parallel sessions and reviewing their results. CodeAgentSwarm is an ADE: it lets you direct your coding agents from one place while keeping tasks, notifications, history and permissions visible.

Yes. CodeAgentSwarm runs xAI's Grok Build CLI (the grok command) as a first-class agent, so you can supervise several Grok Build agents alongside Claude Code, Codex or any other supported CLI. You install and log into Grok Build yourself, and CodeAgentSwarm adds desktop notifications, live diffs for every agent, searchable history and per-project switching.

Yes. CodeAgentSwarm connects to the official cursor-agent CLI through ACP. It streams text and tool activity, exposes Agent, Plan and Ask modes, handles permissions and cancellation, respects .cursor/mcp.json and enables resume only when the installed Cursor version advertises loadSession support.

The desktop app runs on macOS 12+, Windows 10/11 and Linux (deb and AppImage, x64 and ARM64), with Pro included at no cost during the open beta. Create a CodeAgentSwarm account and sign in to the agents you want to use. Their subscriptions, API charges and usage limits remain separate.

Learn to Master CODEAGENTSWARM

Practical guides to get the most out of it

Ready to Transform
Your Development?

Run all your AI coding agents from one visual command center, with live status, notifications and searchable conversation history.

Already using Claude Code, Codex, Antigravity CLI, OpenCode, Kimi Code, Grok Build, Cursor Agent, Muse Code, Pi or Devin CLI? It just works. Install and open.

Loading releases...
hello@codeagentswarm.com

Available for

macOSmacOS
WindowsWindows

System Requirements

OS: macOS 12+ / Windows 10/11 / Linux
Memory: 4GB RAM minimum
Storage: 500MB available
Agent: Claude Code, Codex, Antigravity CLI, OpenCode, Kimi Code, Grok Build, Cursor Agent, Muse Code, Pi or Devin CLI

First time installing? You might need to right-click and select "Open" to bypass Gatekeeper.