Boson

You decide when a pull request gets reviewed.

You comment @useboson review on a pull request. Boson reads that diff once, traces the changed lines through their callers, and quotes exactly where it breaks. Ask again next time you want another pass.

example-org/payments-service · pull request #482illustrative example
  1. jules-lindqvist commented just now

    @useboson review

  2. boson commented moments later

    blocker

    payments/retry.py:142

    if retry_count < MAX_RETRIES:
        schedule_retry(job)

    This runs two lines before the idempotency key is set below it. A retry scheduled in that window double-charges the customer when the webhook replays.

    This comment does not approve or block the pull request.

A finding is a path, not a guess

This is the same example from the top of the page, traced end to end: the function that changed, the caller in between, and the function where it fails. Boson writes this path out in every comment it posts.

payments/retry.py:schedule_retrypayments/charges.py:charge_customer

What happens after you comment

  1. Step 1 of 4

    You ask for a review

    Comment "@useboson review" on an open pull request. That's the whole setup, no config file to add to your repo.

  2. Step 2 of 4

    Deterministic checks run first

    Boson checks the diff against the merge-base commit: is the line actually changed, does the snippet match, what does blame say, has this already been flagged. A model never sees a diff that fails these.

  3. Step 3 of 4

    What's left gets read for a reason to reject it

    Boson reads the change the way a reviewer looking to reject it would, following each line through its callers until it lands on a concrete, reproducible outcome.

  4. Step 4 of 4

    One comment, evidence attached

    Boson quotes the code and the path from that line to the failure it causes. Reading the comment is enough to check the finding yourself.

Credits, pooled at the organization

One credit pays for one review. Every review gets the same depth, there is no faster or slower tier and nothing multiplies the cost. The balance sits at the organization rather than at any one developer's seat, and it only drains when someone actually asks for a review.

Free
Try it on a handful of real pull requests.
Pro
One team running reviews most weeks.
Teams
Credits pooled across a whole organization.
Enterprise
Volume terms and procurement.

We haven't published prices yet. Running a review spends real inference and infrastructure, including whatever a follow-up thread spawns, and we want that fully measured before putting a number next to it. The free tier is open now; get started and we'll tell you the rest before anything is billed.

What you can verify yourself

We haven't run enough reviews yet to publish an accuracy number we'd stand behind, so here are none. Pull up an actual comment Boson posts and check each line below against it.

  • Every finding quotes the exact code and traces a path from that line to a concrete, user-visible failure.
  • Deterministic checks run before any model sees the diff: line in diff, snippet match, blame, duplicate detection.
  • It's a comment. Nothing else on the pull request changes because Boson ran.

Try it on your next pull request.

Install the GitHub App on a repo, then comment @useboson review the next time a pull request needs someone hunting for the way it breaks. No config file to write.