The multiplexer your agents can drive

Claude Code, Codex and 20 more, on your Mac or any SSH host. Agents spawn agents, schedule runs, use your apps, and talk you through what they did.

Download for macOS
Other platforms
macOS 15+ · Linux betaRelease notes
Linux beta installation guide Windows · in progress All releases
Claude Code starts Codex on a remote host called forge in a split pane beside it, waits for it to fix a flaky test, and reads back its result.
Claude Code Codex Cursor Gemini Pi OpenCode Copilot Kimi +14 more

Features

A scheduled Claude Code session, marked with an indigo clock because diri woke the Mac for it, triages new GitHub issues and opens a pull request.

Scheduled runs that wake your Mac.

Every weekday at 7, or once tomorrow. diri wakes your Mac for the run.

Claude Code finishes and hands diri a summary to read aloud; dictation listens to the spoken answer, pastes it into the prompt and sends it.

Talk while you do something else.

diri reads you each summary. Answer out loud and it goes back to the agent.

Claude Code splits an onboarding check with Codex and Pi in three tabs. Each opens its own iPhone Simulator: Codex signs in on the SE and finds an overlap, Pi pages through the tour in dark mode, and Claude Code turns on notifications; the overlap is then fixed.

Computer use for every agent.

Any agent can use your Mac’s apps in the background. Your cursor stays yours.

And everything else

Any machine over SSH
Nothing to install by hand, no tmux, no relay.
A worktree per agent
Parallel work never collides.
Sessions that stay open
Close diri and agents keep working.
Review and ship
Diffs, commits and PR checks beside the session.
Notifications
Jump straight to the session waiting on you.
A browser per agent
Each session tests what it built in its own browser.
One-click account switching
Move every tab to the account with room left.
Conversation history
Resume any past Claude or Codex chat.
Panes and prefix keys
tmux-style splits and keys, scriptable from the CLI.
API tab
Agents open the endpoint they built, ready to send.
Usage and limits
Spend and plan limits beside your sessions.
Quick terminal
A drop-down terminal on a global hotkey.
iPhone companion
Answer questions from your phone. Beta.
Inline images
Kitty and iTerm2 images, locally and over SSH.
Notes and to-dos
Any to-do can start an agent.
Free and open source
AGPL-3.0, native Rust, no account.

Questions

More in the docs and guides.

Is diri free?

Yes. diri is free and open source under AGPL-3.0. You only pay for the agents you use, through your own Claude, ChatGPT or other accounts.

Which agents does it work with?

Claude Code, Codex, Cursor, Gemini, OpenCode, Pi, Copilot, Kimi and more: 22 agent definitions ship with the app, and any terminal agent can be added with a short JSON file.

What happens when I close diri?

Your agents keep working. Reopen diri and every session is where you left it. Restarting your Mac stops them, but Claude Code, Codex and most others can resume the same conversation.

Can agents work together?

Yes. Agents started in diri can start other agents, on this Mac or on any of your SSH hosts, hand them tasks, wait for them, and collect their results. Ask in plain English, starting with “Use diri to…”.

Does diri see my code or prompts?

No. Agents run on your Mac or your own servers, with no diri account and no hosted relay. diri records crash and performance diagnostics, never terminal contents or prompts, and you can turn sharing off.

Which platforms are supported?

macOS 15 or newer on Apple silicon and Intel. Linux is in beta, and a Windows version is in progress.

Start one agent. Let it run the rest.