Configuration¶
Atomsh has no configuration file. Everything is a flag or an environment variable, and the defaults are meant to be right.
Environment variables¶
| Variable | Purpose |
|---|---|
ATOMSH_API_KEY |
Use this credential instead of the stored one. Useful in CI |
ATOMSH_API_BASE |
Point at a different AtomGPT deployment. Default https://atomgpt.org |
ATOMSH_API_KEY takes precedence over ~/.config/atomsh/auth.json, so you can
run a one-off against another account without logging out.
Files¶
| Path | Contents |
|---|---|
~/.config/atomsh/auth.json |
The stored credential, mode 0600 |
~/.local/share/atomsh/sessions/ |
Saved conversations, one JSON file each |
~/.local/share/atomsh/mcp-tools.json |
Cached materials tool definitions |
These follow XDG_CONFIG_HOME and XDG_DATA_HOME when those are set.
Deleting the tool cache forces a refresh on the next run. Deleting a session file removes that conversation. Neither breaks anything.
Models¶
Set a model per invocation with -m, or per session with /model. The default
is gemma-4-26b. atomsh models lists what your account can reach, marking
the one in use.
Which models exist depends on the deployment, not on Atomsh. The list is fetched at runtime.
Uninstalling¶
The second line removes your credential and saved sessions.