Automate

Agents

Claude Code and other agents in bateri: what each one is doing at a glance, and the panes they open for you with the bateri command.

An agent such as Claude Code runs in a bateri terminal like any other program. bateri shows you what it is doing, and the bateri command lets it work beside you: open a pane for a dev server, run the tests next to you and read the result, and tell you when something needs you.

What an agent can do

You ask The agent runs
“Start the dev server below and tell me when it’s up.” bateri split --below -- npm run dev, then bateri capture p9 until the server says it is listening.
“Start the dev server.” With no place named, bateri split --float --minimized --label 'dev server · app' -- npm run dev: a capsule you open with a click, read with bateri capture p9.
“Show me the whole diff.” bateri split --float --keep-open -- git diff, a card over the tab.
“Log in to GitHub.” It tells you that you have to type, then bateri split --float --focus --wait -- gh auth login.
“Run the tests beside you and fix what fails.” bateri split --right --keep-open -- make test, then bateri capture p9 to read the failures.
“Quit the pager in the left pane.” bateri send p3 'q'
“Run the tests on prod too.” bateri split -S prod -- make test, if this Mac’s level is full.
“Let me know when it’s done.” bateri notify "tests failed"
“Close what you opened.” bateri close pane p9

A pane an agent opens doesn’t take your keyboard. A command that would move the keyboard, such as focus, move or zoom, runs only from the pane the cursor is in, while bateri is behind, or in a window other than the one you are typing in: see the keyboard rule.

Telling it about bateri

Turn on Settings ▸ Agents ▸ bateri command and each Claude Code you start in bateri learns the command, for that session alone: nothing is written to Claude Code’s own files. It then runs a dev server or a watcher in a capsule on its own, and opens a login it cannot answer in a floating pane for you to type in.

Any other agent learns the command the way you would, from its documentation. Every page here has a Markdown copy for language models, such as /docs/cli.md, and /llms.txt lists them. Add two lines to its instructions, such as AGENTS.md:

In bateri, open panes beside you with the `bateri` command.
How it works: https://bateri.dev/docs/cli.md

What bateri shows you

  • What each one is doing. The tab of a Claude Code session shows what the agent said last, right under its title. The reply comes from a hook that a Claude Code started in a bateri terminal gets, whether you type claude or a program opens it with bateri split -- claude, for that session alone, so nothing is written to Claude Code’s own files; on a server over ssh too. Settings ▸ Agents ▸ Claude Code turns it off.
  • Activity. Show Activity (⇧⌘A, or the eye at the head of the side column) lists every program that reports its state, as Claude Code does: its work, what it is doing and for how long, and what it asks while it waits for you.
  • What waits for you. The side column counts the tabs that ask, failed or finished while you were elsewhere. Click a count to go to the one that waited longest.
  • The ring on its tab. A program that reports its status, as Claude Code does, turns the ring only while it works, and shows a ? when it’s waiting on you.
  • Screenshots. ⌘V pastes a screenshot into Claude Code, Codex, Gemini CLI and OpenCode.