A pull request review queue ordered by what needs you next.

Yes. Star a pull request and it floats to the top of the queue regardless of what the ranking thinks — for the review you promised to do this morning, or the one your lead just asked about. You can also track any pull request by URL, including ones outside your normal queue.

Age is shown on the row as a 4px border: fresh, warm, or stale. Older work remains visible as newer activity arrives.

No. The default is a quiet unread count in the menu bar. Beyond that you get digest notifications at times you pick — what needs you, what has gone stale, roughly how many minutes of review are queued — rather than a ping per event. Both are configurable in Settings.

Every pull request is labelled Quick, Medium or Deep dive, worked out from the change itself rather than estimated by a model — so it is there with AI switched off. Lockfiles, generated code and vendored directories don't count toward the size wherever your provider lists the files in a change, which is why a large dependency bump reads as Quick, and long documentation counts as the reading it is rather than by the line. Code touching a migration, a schema or an authentication path is called a deep read however small it is — code specifically, so a document that merely mentions one is still Quick. Where a provider tells us nothing about the size of a change, no label is shown rather than a guess. The optional Quick wins sort order leads with the smallest changes. If you enable AI features, you also get an advisory complexity band (easy through extreme) and risk band (safe through critical), each captioned with the provider and model that produced it and marked outdated when new commits land.

Yes. Classification and ranking are built on review state, unresolved threads, age, unread activity, whether someone asked for you by name, and whether a change is conflicting — all read straight from your providers. AI adds summaries, complexity and risk estimates and inline findings on top, and if you never enable it, none of your pull request content is sent to any AI provider.