Commits and suspects
What Plavn records from your GitHub pushes, how it finds the commits behind a problem reported by email, and what it sends to the AI.
Once a repository is linked to a project, Plavn keeps track of what is pushed to it, so that a bug report can be matched to the change that probably caused it.
What is recorded
Section titled “What is recorded”Every time you push to a linked repository, GitHub tells Plavn, and Plavn keeps for each commit:
- the commit message, its author’s name and GitHub username, and the date,
- the branch it was pushed to and a link to it on GitHub,
- the names of the files it added, changed or removed.
Plavn does not keep the code or the diffs. It keeps pushes to branches, not tags. A commit pushed again (to another branch, say) keeps its first record.
For every connected repository, linked or not, Plavn updates the repository’s name, whether it is private, its default branch and the time of its last push, which is what Settings → GitHub shows. An unlinked repository’s commits are not kept.
When Plavn looks for suspects
Section titled “When Plavn looks for suspects”Only when an email needs a reply and Plavn judges it to be work mail, and Prepare replies and tasks is on. Personal mail never loads or sends any work data, code included. See Tasks and prepared replies.
Plavn then:
- looks at the recent commits of your linked repositories: up to the latest 40 from the last 45 days, with no more than 15 from any one repository, so a busy repository cannot push the others out;
- decides which of them, if any, could be behind the problem the email describes;
- for those suspect commits only, fetches their diffs, and may read up to three files as they were at that commit, for code the diff cuts off;
- writes the reply and the task with that in mind.
If it finds no likely commit, the task simply has none. If GitHub cannot be reached for a diff, Plavn goes on without it.
What is sent to the AI
Section titled “What is sent to the AI”For step 1 and 2, the AI sees for each recent commit: its message (the first 300 characters), the first 12 file names, the author and the date, along with the email, your project names and the repositories linked to them. For steps 3 and 4 it also sees the diffs of the suspect commits (up to 12,000 characters each) and short excerpts of the files it chose.
All of it goes to Plavn’s AI provider, or to your own provider if you added your own key, only to answer that request. If your repository is private, this includes pieces of its code. To stop it:
- turn off Prepare replies and tasks in Settings → AI & Automation,
- turn AI off for the email account, or
- unlink the repository (or disconnect GitHub), so its commits are no longer used.
See AI and privacy for everything else Plavn sends.
Where you see the result
Section titled “Where you see the result”Open the task. In From email, Suspect commits lists each suspect with:
- the short commit ID (a link to it on GitHub) and the repository,
- the author and date,
- the commit message’s first line,
- the reason Plavn suspects it,
- the first few files it changed.
A badge says how sure Plavn is: High confidence, Medium confidence or Low confidence. Treat it as a lead to check, not a verdict: Plavn compares what the customer describes with what changed recently, and it can be wrong.
The same suspects go into a GitHub issue if you create one from the task.
GitHub issues: turn the task into an issue and keep its state in step.