Skip to content

FAQ

Does MergeMate replace human reviewers?

No — it works alongside them.

Code review is essential but exhausting. Long PRs get short feedback. MergeMate fills the gaps: it catches oversights, suggests improvements, and gives reviewers a running start. But the final call always belongs to a person.

A few safeguards built into the design:

  • Your original PR description always stays on top
  • MergeMate never approves a PR
  • Suggestions are optional and structured so you can scan them fast
  • The goal is to encourage self-review, not to automate it away
I got a suggestion that doesn't make sense. What gives?

AI models are powerful but not perfect. Even the best ones occasionally misfire.

The real value isn't in blindly accepting every suggestion — it's in the moments where the model spots something you missed. Spending 30 seconds skimming suggestions is worth it when it catches a bug before production.

Quick filter technique: 1. Glance at the category header. If it's irrelevant, skip. 2. Read the one-line summary. Still irrelevant? Skip. 3. Expand only the suggestions that matter.

Want better results? Use extra_instructions to steer the model toward what your project cares about. Learn how →

Can I customize the suggestions I get?

Absolutely. The extra_instructions and best_practices knobs let you tune the output. Details here →

Do you keep my code?

No storage. No training. No exceptions. See the data privacy page for the full breakdown.

Can MergeMate review draft PRs?

Yes — just trigger it manually with a comment (/review, /describe, etc.). Draft PRs won't trigger automatic runs unless you opt in. More on automations →

Can I calibrate the review effort estimates?

Yes. Use extra_instructions to map effort levels to your team's expectations. Example:

  • Effort 1 → under 30 minutes
  • Effort 2 → 30–60 minutes
  • Effort 3 → 60–90 minutes

The scale (1–5) is meant for quick comparison between PRs, not exact timekeeping. Configuration reference →

MergeMate feels noisy. How do I dial it down?

The defaults are tuned for signal over noise, but every team has different tolerances. Here's what's already in place:

  • Structured, scannable output (not wall-of-text comments)
  • Suggestions grouped in tables, not inline commits
  • Verbose sections folded by default
  • No "I'm working on it…" placeholder messages

If you still want less:

Prefer more output? Flip the knobs the other way — dual-publishing mode and interactive usage are good starting points.