How you use it

One page, on any screen. Each conversation is one agent in its own machine.

The page: four conversations on the left, each with its tool; on the right the terminal of one of them, with the folders it was given shown above it.

It sees only what you attach

A folder, a repository, a mailbox: to read, or to write. A key is handed to one agent and seen by no other. One client's work never meets another's.

What one conversation has: a repository and a folder to write, a folder to read, one key to use, and its network, online or offline.

client-a writes to site-a, reads brand-guide, and is handed DEPLOY_TOKEN, which no other agent sees.

They work together

Give two agents the same folder. One asks there, the other answers, whatever tool each one is.

A folder two conversations share: a request written by one agent, then the picture and the answer left by the other.

client-a asked for a banner. pictures, another tool in another machine, made it.

It commits. You send.

The agent never holds the key that lets code out. You are shown what would go, then you send exactly that.

Two commits made by the agent, with their changes, and one button: send 2 commits to main.

Two commits wait. Nothing leaves until you press the button.

Read the report, not the terminal

A report the agent wrote opens as a page, on a tablet. Share it with a code, no account. Take it back anytime.

On a tablet: a price comparison written by the agent, shown as a page with a table and a recommendation.

Its terminal, or a conversation

The terminal is the default and stays whole. An agent can also be given six verbs of its own (tell, ask, show, read, status, doing): the same work then reads as a conversation, shown first, with buttons for its questions, one line saying what it is doing, and your phone told when it asks; its terminal stays one tap away. Tried for real with OpenCode.

See it as someone who only uses it →

A real exchange shown as a conversation: the agent asks 'Should the cover be blue or red?' with two buttons; under it, the file it made and what its terminal says now.

Why not just…

tmux over sshA terminal, and your whole computer behind it.A page made for the phone, and a wall around each job.
a containerThe host's kernel, and the rest to build yourself.A virtual machine each, with rights, keys and the send button already there.
the tool's own remote controlOne tool, on your computer, with all of it in reach.Any tool, several at once, each in its own virtual machine with only what you attached.
a cloud agentYour code and your keys on someone else's servers.They stay at home. The optional relay cannot read what passes.

From phone or tabletThe same page everywhere, with voice dictation.
Sleep, wake, rebuildA machine costs nothing asleep, and its work folder outlives it.
Cut its networkPer conversation, from outside the machine.

How it is kept safe →

How you install it

On a machine at home

A fresh Ubuntu Server, one command. It sets up everything, the virtual machines included.

sudo deploy/install.sh \
  --domain clav.example.com \
  --operator you@example.com

No public address at home?

One more command, on any small server. It relays, and cannot read what passes.

sudo deploy/relay.sh --key "ssh-ed25519 AAAA…"

Set up once. After that, you only open a page.

Install, backup, restore →