Two outcomes. One rule.
You set the confidence threshold once per project. Every AI fix is scored on the way out. Above the line — merged. Below the line — goes to your review queue.
Pin/comment mode feedback always goes to review — auto-merge only applies to text highlight and element click feedback.
How auto-merge works
From the moment your client clicks Submit to the moment the fix is live — no one on your team has to do anything.
Client submits feedback using text or element mode
The widget captures structured data — exact before/after text, or the element HTML and a description. Pin/comment mode feedback doesn't trigger auto-merge because it's too open-ended for high-confidence AI fixes.
AI reads the feedback and writes a code fix
Pensl's AI reads the structured payload, searches your GitHub repository for the matching file, and writes the minimal change — a text replacement or a component update. The change is as small as possible.
The fix gets a confidence score
The AI scores its own fix from 0 to 100%. The score reflects how certain it is that it found the right file and made the right change — based on how specific the selector was, how exact the text match was, and how cleanly the file was found.
Above threshold: pull request is created and merged
If the score is above your threshold and auto-merge is on, Pensl creates a branch, commits the change, opens a pull request, and merges it — all automatically. The feedback card moves to "Done" in your Pensl dashboard.
Below threshold: goes to your review queue
If the score is below threshold, the feedback moves to the "Review" column in your Pensl Kanban. You see the suggested diff, the confidence score, and the original feedback. One click to approve, one click to reject.
One toggle per project
Auto-merge is a setting in your Pensl project dashboard. You turn it on or off per project — so you can run it on low-risk client sites while keeping manual review on anything more sensitive.
- Per-project setting — mix and match across clients
- Threshold adjustable from 0% to 100%
- Start at 80% and lower it as you build trust
- Every merged PR shows in the dashboard with a diff link
- Turn it off instantly — no code change needed
Is it safe to let code merge automatically?
Every auto-merged change is a Git commit. Nothing is irreversible. Here is why it works without breaking things.
Everything lives in Git
Each auto-merge creates a regular Git commit with a clear message. If you ever want to undo it, one git revert brings the site back to where it was before the change landed.
You set the bar for what gets through
The confidence threshold is your safety valve. Start at 85% and lower it over time as you see that the fixes are accurate. Anything below the threshold always goes to you for review first.
Only applies to specific feedback types
Auto-merge only triggers on text highlight and element click feedback — the two modes where the AI has exact before/after data. Pin and comment feedback always goes to manual review.
You see every change in the dashboard
Every auto-merged fix appears in your Pensl Kanban as Resolved. The PR link, confidence score, diff, and original feedback are all there. Nothing happens silently.
Your client is none the wiser
The widget gives your client no indication of what happens after they submit. They just see that their feedback was received. You decide what gets merged and when.
Turn it off at any time
Auto-merge is a toggle in your project settings. Flip it off at any time — no code change, no redeployment needed. Future fixes will go to your review queue instead.
Which feedback can auto-merge?
Only feedback that gives the AI precise, structured data can safely auto-merge. Vague or open-ended feedback always goes to you.
Text highlight
Client selects text and types the replacement. Exact before/after strings captured. AI finds the string in your codebase with very high accuracy. Highest confidence scores of any mode.
Element click
Client clicks a button, card, or section and describes the change. Full element HTML captured. AI reads the structure and applies the described change. Good confidence scores on most codebases.
Drop a pin
Client clicks anywhere on the page and leaves a free-form comment. No structured code data. The AI has no selector, no before/after text. Feedback goes to your dashboard for manual interpretation.