Queue the next prompt while it works
Write the follow-up while the agent is mid-turn. It sends the moment the turn ends, and you can pull it back to edit until then.
Ten agents.
One window.
Haven runs your Claude Code sessions across your Mac and your servers, and marks the ones waiting on you.
Search every terminal at once
⌘⇧F runs one query across all open scrollback and the
files in your working directory. Terminals with a match light up in
the tab strip as you type.
Claude renders inline
A built-in MCP (haven-display) lets Claude drop images,
video, audio, diffs, and live HTML into the session, instead of
writing paths you have to go open yourself.
⌘K writes the command
Describe what you want in plain English. The command lands in the prompt for you to read, edit, and run.
Sessions survive the disconnect
Remote sessions run under Haven's open-source daemon, not your SSH connection. Close the lid, change networks, come back to full scrollback.
The rest of it
- SFTP file browser and editor, opened in the session's cwd.
- Per-host CPU, memory, GPU, VRAM, and temperature, with the session that owns each GPU process.
- Notifications when an agent finishes or needs an answer.
Haven for iPhone
Answer from your phone
The iPhone app shows the same live stream as the Mac, for local and remote sessions both. Read what happened, send the next prompt, or start Claude on a terminal that isn't running one.
- Images, video, audio, and HTML render inline Whatever Claude shows you on the desktop with
haven-displayarrives on the phone, fetched when you open it. - Works when your Mac is offline Remote machines stream straight to your phone, end to end encrypted. We can't read it.
FAQ
Which agents does it work with?
Claude Code. Haven reads its hooks and transcript, which is where the status, recaps, and diffs come from. Any other CLI runs fine in the terminal, it just doesn't get the agent view.
Do I have to give up my terminal?
No. Haven is a real terminal: your shell, your dotfiles, your iTerm color themes.
What does a remote host need?
Linux on x86_64 or ARM64, and SSH access. Haven detects the architecture and uploads the daemon on first connect.
How does session persistence work?
Haven runs a lightweight open-source daemon on each remote host that manages PTY sessions independently of your SSH connection. Disconnect and the sessions keep running. Reconnect and Haven reattaches and restores scrollback. No tmux, no shell wrapping.
Do local sessions persist?
The shell process dies when Haven quits, but the session is still in the sidebar when you come back, with its working directory and recent scrollback. Remote sessions are the ones whose processes actually keep running.
Is my data private?
Haven runs on your machines. There's no Haven-operated backend your sessions pass through, so your output, commands, and credentials never reach us. The exceptions are opt-in AI features like host nicknames and agent-response summaries, which send a small amount of data through a Haven backend to a third-party model; we don't store those requests and the provider doesn't train on them. Past that we collect anonymous product analytics: app starts, feature usage, crash reports. Details on the Privacy page.
Your workspace should be beautiful.