Agent Monitor

An AI agent rewriting your codebase looks like scrolling text in a terminal.
Agent Monitor shows you what it is actually doing — every file, class, and
function it touches, lighting up as it happens.

Three panes, updating while the agent works:

• Narrative / Review notes — a streamed running commentary on what the agent

appears to be doing, and a file-by-file compilation of everything the AI has
said about the change, with risk levels.

• Structure — repository to directories to files to classes to functions, as

a force-directed graph. Changed symbols light up and pulse.

• Flow — the static call graph, entry points fanning right through your

functions into external packages.

You can also point it at uncommitted work, a specific commit, or this branch
against another, to review the shape of a change before opening a PR.

Symbol identity is a hash of the AST rather than the source text, so
reformatting produces no changes at all — a formatter run will not light up
the whole graph, while a docstring edit still counts. Python is fully
supported; JavaScript and TypeScript via tree-sitter; every other language
still appears at file level, so nothing is invisible.

AI insights are optional and inert until you add an Anthropic API key in
Settings. The key never reaches the frontend and nothing is sent without it.

Note: as a strictly-confined snap, Agent Monitor can read repositories under
your home directory and on removable media. Repositories elsewhere on the
filesystem are not visible to it.