-
Notifications
You must be signed in to change notification settings - Fork 0
Home
Watch your AI agents work. From anywhere. They already run in tmux — Claude Code, Codex, Gemini, a script that takes two hours. Argus puts those sessions, the files they are writing and the report they just produced in one browser tab, and the same thing on your phone. Nothing to install inside the agent: if it runs in a terminal, it works.
- Getting started — install it, run it, configure it
- Reaching it from outside — Tailscale, SSH tunnels, a reverse proxy, and the case where the network only goes one way
- Starting an agent — a shell or an agent, in a folder or a fresh git worktree, with its first instruction already typed
- Sessions and the terminal — attaching, typing, copying, sizing
- Files — browsing, editing, uploading, pasting screenshots
- Documents — markdown, PDF, Word, logs
- Desks and windows — workspaces, layout, the link tray
- Two agents on one job — the two patterns, the plan file, and the review loop that closes itself
- Notifications — being told when it has finished, or when it wants you
- What each agent can do — the matrix: who rings, who can say where it is
- Keyboard shortcuts — the keys, and how to change them
- On a phone — the drawer, the window sheet, and what a thumb can reach
- The machine — health, disks, listening ports
- tmux configuration — editing it from the app, safely
- Settings and languages
- Security — what protects it, and what does not
- Troubleshooting — the things that surprise people
- Everything it does — the whole catalogue, one page, searchable
- An agent driving Argus — a key that can do five things, and the two calls that let your agents hand work to each other
- Orchestrating several agents — three worked examples: N attempts and a judge, N referees and an editor, three roles on one repository
- The API — the browser is one client of it, not a privileged one, and Swagger on your own machine
- Development — the tests, the three scripts, the vendored code
A small web app served by the machine your work runs on. It attaches to the tmux sessions you already have, the way another terminal window would, and puts a filesystem, a document viewer and the machine's own health next to them.
It is not a tmux replacement: kill Argus and every session carries on. It is not multi-user: one token, one machine, one person's work. And it is remote shell access, so read Security before putting it anywhere but a trusted network.
An editor is built for the hours you spend writing code. More of that work is now done by an agent running for two hours in a session, and what is left for you is watching it, unblocking it when it asks, and judging what came out — a job an editor is not built for, and above all a job you often want to do from somewhere else.
In a hurry, or stuck: the FAQ answers the questions people actually ask — whether it can go on the internet (no), what happens when you log out, why there is no Docker image, and what it sends anywhere (almost nothing).
Several machines? Panoptes is the board: every machine on one page, which sessions are on each, and which one is waiting for you. It holds one read-only key per machine and never gives a browser any of them. Argus does not depend on it — turn it off and every machine carries on alone.