Using Atomsh

Invocations

atomsh                             # interactive session
atomsh "fix the failing test"      # run one prompt and exit
atomsh -c                          # resume this directory's last session
git diff | atomsh "review this"    # read a prompt from stdin

With a prompt argument Atomsh runs one turn and exits, which makes it usable in scripts and pipelines. With no argument and a terminal attached it starts an interactive session.

Flags

Flag	Effect
-m, --model	Model to use. Default gemma-4-26b
-c, --continue, --resume	Resume the most recent session for this directory
--yolo	Do not ask before writing files or running commands
--readonly	Refuse all writes and shell commands
--no-materials	Leave out the AtomGPT materials tools
--no-color	Disable ANSI colour
-V, --version	Print the version

Commands

atomsh login      # connect your atomgpt.org account
atomsh logout     # forget the stored credential
atomsh whoami     # check the stored credential
atomsh models     # list available models

In a session

Input	Effect
!<command>	Run a shell command yourself, with no model round-trip and no approval prompt
/model <id>	Switch model. Without an argument, print the current one
/models	List available models, marking the one in use
/history	Replay the conversation in full
/clear	Start a fresh thread
/help	Show the command list
/exit	Quit, as does Ctrl-D
++esc++	Interrupt a response while it is streaming

Pasting multiple lines sends them as a single prompt rather than one prompt per line.

The workspace

Atomsh treats the root of your git repository as its workspace, so starting in a subdirectory still lets it reach the whole project. Outside a repository, the workspace is the current directory.

A path outside the workspace always asks for approval, even after you have allowed a tool for the session.

Sessions

Every conversation is saved under ~/.local/share/atomsh/sessions/, keyed by workspace. atomsh -c resumes the most recent one for the directory you are in and replays the last few exchanges so you can see where you left off.

Models

atomsh models lists what your account can reach. The mcp.* entries are excluded on purpose: those run the AtomGPT materials agent on the server and answer in prose rather than making tool calls, so they cannot drive a coding loop. Use them from a chat client instead.