The explore agent is in beta and may change.
momentic ai explore still works
across that split, because the repo it diffs and the repo it writes to are
independent. explore diff diffs whatever repo you run it in, the current
working directory. --config picks the
momentic.config.yaml that defines the
project, and that config decides where tests are read from and written to.
This guide covers the pull-request setup: diff the app repo, write into the test
repo. If you would rather not manage two checkouts, see
Seed a separate test repo against a running app.
Run it
Check both repos out side by side, say the app at /code and the tests at
/tests, and run explore from the app checkout with --config aimed at the
test repo:
The diff is /code, so make sure that checkout has enough history to cover the
change under review (fetch-depth: 0 on GitHub Actions). In CI the range
defaults to the pull request’s window, so you rarely pass one explicitly. The
project is /tests, so the agent reads your existing tests from there and
writes new ones back into /tests.
Delivering the tests
This is the catch with the pull-request setup. The On successful explore
setting in Settings > Explore
commits and opens its PR against the repo explore ran in, which here is /code,
and it only picks up files git reports as changed there. The generated tests
landed in /tests, so that automatic PR never sees them.
Leave the setting on its default (nothing, which leaves the tests on disk) and
have your workflow commit the /tests changes and open the PR against the test
repo itself.
Notes
- Explore never touches your application code. Its analysis is read-only with
respect to the code repo.
--no-code works here too if you want the diff window without handing the
agent the code repo’s files.- A single run reads code from one repo. Pulling in source from several repos at
once is not supported yet.
