AI agents
taiku detects coding agents in your terminals, tracks their status, and tells you the moment one needs you.
When you run an AI coding agent inside a taiku terminal, the session recognizes it automatically and keeps an eye on it for everyone in the room.
An agent in a taiku terminal uses the login it already has on that machine. How to sign one in with a subscription or an API key, and what reaches guests and cloud boxes, is on its own page: Claude Code, Codex, Gemini CLI, OpenCode.
Automatic agent detection
You do not flip a switch or label anything. Start a coding agent in a terminal and taiku spots it from the agent's own output, then watches that terminal in real time. It recognizes Claude Code, Codex, Aider, GitHub Copilot, Cursor and OpenCode.
Each agent terminal shows a colored status dot in its title so you and your collaborators can read the room at a glance:
| Status | What it means |
|---|---|
| Starting | Just detected, or still coming online. |
| Running | Actively producing output. |
| Thinking | Working on a response; the terminal is quiet for now. |
| Tool use | Reading files, running commands, or making edits. |
| Waiting | Stopped and asking for your input. Needs your attention. |
| Completed | Reported its task is finished. |
| Idle | Detected earlier, but quiet for a while. |
Detection keeps up as things change. Start an agent after the session is already running and it gets picked up as soon as it speaks. Swap to a different agent in the same terminal and the status follows along.
Get pinged when an agent needs you
The most important moment is when an agent stops to ask for approval. taiku can alert you the instant a terminal switches to Waiting so you are not babysitting output. Choose how you want to be notified under Settings → Agent:
- Sound plays a short ping.
- Flash pulses the terminal's border.
- Both does each at once.
A volume slider tunes the ping, and you can turn alerts off entirely. taiku also sends a native system notification, in the browser and the desktop app alike, so you get nudged even when taiku is in the background. Alerts are spaced out per terminal, so a chatty agent will not flood you with repeats.
When every taiku tab is closed
Background notifications go one step further: turn on Notify me when taiku is closed under Settings → Agent and your browser will still surface workspace and sync updates with no taiku tab open at all. Your browser asks for permission the first time, and taiku only asks when you flip that switch, because a dismissed permission prompt cannot be re-asked from inside the app.
A few things worth knowing:
- The taiku desktop and mobile apps do not need this. They already receive the same alerts through the system, and turning it on in a browser does not duplicate them.
- On iPhone and iPad, add taiku to your Home Screen first. Safari only delivers background notifications to an installed web app. The setting tells you so until you do. On a Mac, Safari works in an ordinary tab.
- If you later block notifications in your browser's own settings, the switch turns itself back off the next time you open taiku, so it never claims to be on when it is not.
- Clicking a notification jumps to the taiku window you already have open rather than opening a second one.
The activity panel
The activity panel (the bell icon) gathers every agent and alert into one place so you can run several agents at once without losing the thread.
From the panel you can see:
- Every terminal that needs attention, with its current status.
- Per-agent stats: tokens used, cost, and how many steps it has taken.
- A running list of agent sessions, including ones that have already ended, each one click away.
- A Track diffs toggle. With it on, finished commands earn a Δ badge you can click to review exactly what that run changed. See Reviewing changes.
Jump to an unseen terminal
When you are running a handful of agents, the one that needs you is rarely the one on screen. taiku marks terminals you have not looked at yet and lets you jump straight to the next one that wants attention.
Click an entry in the activity panel to bring that terminal into focus, wherever
it lives in your layout. You can also mark a terminal as unseen yourself to come
back to it later, with Cmd+U on macOS or Ctrl+Alt+U on Linux and Windows.
Resume and relaunch an agent
Agents do not always finish in one sitting. If a terminal closes or you step away, you can pick an agent's work back up instead of starting over.
Click the agent's badge in its tile's title bar and open Sessions. Each past
session has its resume command (for Claude, claude --resume <id>) one click
away on your clipboard. Run it in any terminal and the agent carries on from
where it stopped, with its earlier work in context.
Working across many agents
taiku is built for running more than one agent at a time, even several inside a single terminal split into panes. It keeps the agents apart so each one gets its own status and stats: three agents side by side stay three separate entries, not one blurred mess. When an agent kicks off a helper agent for a parallel task, taiku links the two so you can see the parent and its helper together. An agent on your account shows its helpers too: see Subagents.
Agents that work without a terminal
An agent can also belong to your account instead of a terminal: you talk to it in a conversation, give it skills and connectors, and put it on a schedule. See Agents on your account.