The real terminal
for the AI Native era.
Run every AI agent side by side, each session in its own worktree. See which one needs you, and share context between them.
See it run, in under a minute
A short tour: several agents in one window, each saying what it is doing, and what happens when one of them needs you.
A terminal made for running agents
Knows your agents
It can tell whether an agent is working, finished, or waiting for you.
Runs them side by side
Workspaces, tabs and split panes, so several agents fit in one window.
Orchestrates a team
Write one goal and a planned team of agents runs it in waves, then reviews the result.
Shows what's going on
Processes, usage, git and subagents each get their own screen.
Small and quick
Under a 30MB download, written in Rust and drawn on the GPU. It opens right away.
Reaches your phone
Open your workspaces in a browser on another device, over your own Tailscale network.
One goal. A whole team of agents.
Write what you want done. An orchestrator plans a team of agents, shows you the plan, runs it in waves and has the result reviewed, and you watch every step live.
Agentty orchestration turns one goal into a team of AI coding agents that plan, run in waves and review the result.
Plans the team
Claude Code, Codex, Gemini CLI and Antigravity CLI, each on the part it fits.
You approve first
View the plan, ask for changes, then Start.
Live graph
Every agent's state, model and step, wave by wave.
Resumes after restarts
Resume runs only what is left. Permission checks stay on.
Your agent's terminal, as a chat
Claude Code, Codex, Antigravity and Gemini CLI keep running as the real CLI. Show the pane as a desktop-style chat, and flip back to the terminal whenever you like.
- Switch any agent pane with ⇧⌘J or the speech bubble; the session never restarts
- Formatted replies with syntax-colored code, tool steps as cards that open, to-do lists as checklists
- Answer approvals right in the chat, drag files in, and Stop a turn without losing your message
- Select text to reply with it or start a side chat; new sessions open on a start screen with your earlier ones
Every AI coding agent, in one terminal app
Claude Code, Codex, Gemini CLI, Antigravity, Amp, OpenCode and more run side by side, with no limit on panes. Each session can get its own worktree, so none of them touch the same files.
Edit code with your agents
Open a file, connect Claude Code or Codex, and watch its changes land at its caret.
- Go to definition, completions, quick fixes and rename in nine languages
- Select lines and press ⌘K to Ask AI about them
- Every change stays marked until you Keep or Undo it
It knows what your agents are doing
Agentty notices when an agent finishes, asks for permission or waits for you, and tells you with a border, a badge and a notification.
- Working, done, waiting and permission states per agent
- The highlight stays until you look
- Notifications show the reply’s first line
- Model, context, usage limit and branch on every pane
- Prevent sleep keeps the computer awake while an agent works, or for 1–72 hours
macOS only Clicking a notification to jump to its pane is macOS only for now.
- workingClaude Codeapi · refactor
- workingCodexapi · tests
- waitingGemini CLIdocs
- doneAntigravityweb · ui
Your terminals, on any device
Open them in a browser, wherever you are and whatever you are holding, and pick the work up where you left it.
- Every workspace, tab and split pane, each with the state it is in
- Type into the terminal, or answer a question with one tap
- Start something new: pick or make a folder in ~/AgenttyRemote — the one folder the page can reach — and open a terminal, Claude Code or Codex in it
- Browser notifications when a session finishes or asks you something
- Slack, Discord and Telegram notices can link to that exact terminal
- Speak a prompt instead of typing it, transcribed on your own machine by a model you install once
- The page gets itself back after sleep or a network change — Agentty checks every 30 seconds and restarts it
A mobile app for your terminals is in the works, so you can reach them without opening a browser.
Reachable through Tailscale, and only for the account that owns the Mac — not by IP, not from the same Wi‑Fi. Agentty refuses to run if Funnel is on for its port.
Kept as a hash in the system Keychain. Five wrong tries lock the page, doubling up to an hour, and the sign‑in is an HttpOnly cookie with cross‑site requests refused.
Disconnect, sign out or quit Tailscale and the page goes down within seconds. It comes back when you reconnect.
From an idea to a live URL
Describe the app you want. Agentty sets up the project and you watch the agent build it in the in‑app browser. When it is done, Launch puts it on the web.
Describe an app or drop in a spec or files, and Claude Code or Codex builds it with a live preview.
Creates a GitHub repo and a Vercel deployment and hands you a public URL. If a step fails, the error goes to the agent to fix.
- Project created · habit-streaks
- Claude Code is building…
- Live preview · localhost:3000
- GitHub repository
- Vercel production deploy
- Supabase migrations
Under 30MB to download, and it opens right away
No Electron. Written in Rust and drawn by the GPU, so the CPU and memory stay with your agents.
Download size · macOS
Agentty beside the tools it is usually compared with
Four terminals for running coding agents, and the honest version of where each one stops. A dash means the project does not document that feature — including one row where the dash is ours.
These are reference results, compared against each project's open-source code; features may have changed since the comparison was made.
Everything Agentty does, on one page
Orchestration
Write a goal: an orchestrator plans a team of Claude Code, Codex, Gemini CLI and Antigravity CLI agents, runs it in waves and has the result reviewed.
Build my idea
Describe an app, paste a plan or drop files, and an agent builds it with a live preview.
Launch
A GitHub repo, a Vercel production deploy and a public URL, with Supabase when you need it.
Agent status
Working, done, waiting or needs permission, with highlighted panes and notifications.
Unlimited splits
As many workspaces, tabs, panes and windows as you need. Agents open in a new tab or a split.
A working tree per session
A new AI session in a busy project gets a fresh git worktree and branch, or joins an existing one — your choice.
Session manager
Search, resume and move sessions between Claude Code and Codex.
Chat workspaces
Talk to one lead agent. It plans the work and starts workers, each with its own terminal and its own git worktree — four at a time, reviewers included.
Remote access
Your workspaces in a browser on another device, over your own Tailscale network — watch, type and send prompts.
Voice prompts
On the remote page, tap the microphone and speak. The clip is transcribed by a local model on the machine running Agentty, and no audio leaves it.
In‑app browser
See results instantly while agents click, type and test — each terminal in a browser of its own.
AgentGit
Built‑in Git for the repo of the session you're looking at. ⌘‑click the branch chip to open it on your Git host.
Your conversations stay on your computer
Session lists and usage come from files on your machine, and Agentty never sends your conversations anywhere. The source is Apache 2.0.
- Conversations stay local
- Secrets in Keychain
- Owner‑only file access
- Apache 2.0 licensed
Free. The whole thing.
No account, no subscription, no seats, no paid tier. Every release is the same build for everyone.
- Every feature on this page, on macOS, Windows and Linux
- Unlimited workspaces, tabs, panes and agents
- Plugins from the Marketplace, a folder or your own Git
- Updates for as long as there are releases
- The source itself — read it, fork it, build it
You pay your AI provider directly, the way you already do. Agentty never sits between the agent CLIs and their providers, and takes no cut of what you spend there.
Frequently asked questions
What is Agentty?
Agentty is a native terminal app that orchestrates AI coding agents such as Claude Code, Codex, Gemini CLI and Antigravity CLI. Write one goal and it plans a team of agents, runs it in waves and has the result reviewed; or run as many agents as you want side by side in split panes, see what each one is doing, and edit code together with them.
What is Agentty orchestration?
Agentty orchestration turns one goal into a team of AI coding agents that plan, run in waves and review the result. An orchestrator agent writes the plan and shows it to you; after you press Start it runs the agents wave by wave, a reviewer checks the result against the goal, and the Orchestration page (⇧⌘F) shows every agent's state, model and current step live.
Which AI agents does Agentty work with?
Claude Code and Codex have the deepest integration, including live status, session management and context sharing. Gemini CLI, Antigravity CLI, Amp, OpenCode, Copilot CLI, Cursor Agent, Qwen Code, Droid, Goose, Crush, Aider and local Ollama models are detected and can be launched from the app. In an orchestration, Claude Code, Codex, Gemini CLI and Antigravity CLI can work on one team; Claude Code leads when it is installed, Codex otherwise.
Does orchestration run agents with their permission checks off?
No. Agents in an orchestration never run with their permission checks off, and the plan can wait for you to press Start before anything runs.
Can an orchestration resume after a crash or a restart?
Yes. Orchestrations are kept across restarts, and Resume runs only what is left: the orchestrator checks what is already done and carries on from there. Stop all halts the orchestrator and every agent still running, and Resume picks it up later.
How is Agentty different from tmux?
tmux splits a terminal into panes but doesn't know what runs inside them. Agentty also gives you unlimited splits, and on top of that it knows when an agent is working or waiting for you, manages Claude Code and Codex sessions, passes context between them, and includes an in‑app browser, Git and usage views.
How is Agentty different from cmux?
Both are native terminals for AI coding agents with split panes, attention notifications and an in‑app browser, but cmux is macOS‑only while Agentty also runs on Windows and Linux. cmux is built on Ghostty and focuses on vertical tabs and a scriptable CLI and socket API. Agentty goes further into the sessions themselves: it links Claude Code and Codex sessions so context flows between them live or both ways, lets you search, resume and move sessions between the two, and adds an AI process monitor, a usage dashboard, a built‑in Git page and MCP server management.
Is Agentty free?
Yes, Agentty is free to download. The source code is public under the Apache License 2.0.
Does Agentty upload my conversations?
No. Session lists, usage numbers and handoffs are built from files on your computer.
Can Agentty build and deploy an app for me?
Yes. Build my idea takes a description, a plan or files, creates a project and starts Claude Code or Codex to build it with a live preview in the in‑app browser. The built‑in Launch plugin then creates a GitHub repository and a Vercel production deploy with a public URL, and can connect Supabase.
Which platforms does Agentty support?
macOS 13 or later on Apple silicon, Windows 10 version 1809 or later on x64, and Linux on x86_64 as a .deb (Debian 12+, Ubuntu 22.04+) or an .rpm (RHEL 9+, Fedora). Intel Macs are not supported. The in‑app browser, the browser tools agents and plugins drive, remote access, the menu bar item, mini mode and click‑to‑focus notifications are macOS‑only for now; everything else works on all three.
Why does Windows warn me when I install Agentty?
The Windows installer is not code‑signed yet, so SmartScreen may say the publisher is unknown. Choose More info → Run anyway. It installs for your user only and needs no administrator rights. If your browser refuses to download the .exe at all — Chrome blocks unsigned executables — take the .zip instead; it holds the same installer.
Can I use Agentty from my phone?
Remote access puts your workspaces in a browser on another device — a phone, a tablet or another computer — over your own Tailscale network: every workspace, tab and split pane with its status, the terminal live, typing into it, quick buttons to answer an agent, a prompt box, and browser notifications when a session finishes or asks something. It is never on the public internet: only your own tailnet reaches it, only the Tailscale account that owns the Mac is let in, and then a web password you set yourself. It is macOS‑only for now, and it arrived in v0.2.11. Since v0.2.14 you can also speak a prompt instead of typing it: tap the microphone and the clip is transcribed on your own machine by a local model you install once from Settings → Remote access, so no audio is sent anywhere. A chat workspace opens on the page as its chat, too. Since v0.2.19 the page can start new work as well: + New workspace browses ~/AgenttyRemote on the Mac, makes a folder there if you want one, and opens a terminal, Claude Code or Codex inside it — and nothing outside that folder is reachable from the page.
Smarter AI development
starts here.
Free to download, and open source under the Apache License 2.0.