Agents
ekatva finds AI agent sessions in tmux (Claude Code, Codex, Aider, Gemini) and reads each one’s state from its screen:
| State | What you see |
|---|---|
| working | It’s running (its status line says “esc to interrupt”), with what it’s doing. |
| waiting | It’s asking: the question and its numbered options, exactly as on screen. |
| done | It finished its turn and is waiting for you (Claude’s Stop hook marked it). |
| idle | Nothing happening. |
From the phone you can answer an option, reply with text (typed or dictated), interrupt it (Esc), read its recent output, and start a new session in any of your projects with a first prompt.
Answers land where you meant
Section titled “Answers land where you meant”Each question has a hash. An answer carries the hash of the question you saw; the daemon re-reads the screen and sends the key only if the same question is still there. A late tap can’t answer the next question by accident.
Projects
Section titled “Projects”New sessions start only in your projects: those Claude knows (~/.claude.json),
those open in tmux, and git repositories in ~/personal, ~/codes, ~/work,
~/projects, ~/src (configurable). Each runs in its own tmux session, with
a shell underneath so it stays when the agent exits.
ekatva agents