API Reference / agent / tools /

system/

The tools for the installation itself, and for serving models.

Housekeeping rather than modelling: what version is running, how to update it, and how to put a trained model behind an HTTP endpoint.

Tools

  • tuiml_system_info: Installed version, install method, Python and
platform, optionally checking PyPI for a newer release.
  • tuiml_self_update: Reinstall TuiML at the latest or a pinned version.
Supports dry_run to report what it would do without doing it.
  • tuiml_restart: Kill running tuiml-mcp processes so their parent
clients respawn them on the freshly installed code.
  • tuiml_serve_model / tuiml_stop_server / tuiml_server_status: Run a
trained model as a REST API, and manage those servers.

Notes

tuiml_self_update refuses to upgrade an editable checkout, since the fix there is git pull. tuiml_restart is genuinely destructive: it terminates processes, and it excludes the calling process so an agent does not kill the server handling its own request. The serving tools are thin wrappers over serving, sharing one registry with it, so a server started from Python can be stopped by an agent and vice versa.

Modules