Your reviewers approved these PRs. They broke anyway.

Precall learns which changes kept needing a fix in your repo and comments on the risky ones on the next PR.

github-actionsbotExample

Precall: this PR breaches a rule from your repo

Changes src/middleware/etag/index.ts

6 of 21 earlier PRs that changed this needed a fix

#5111 feat(etag): support conditional requests for the QUERY method fixed by #5196, #5199, #5234
#4198 fix(etag): fallback if `res.clone()` is not supported fixed by #4200

How it works

1. Connect GitHub
Pick the repos you want Precall to watch.
2. Precall reads your repo's history
It finds the files and habits that kept needing a fix after merge.
3. It comments on risky PRs
When a new PR touches one of those spots, it says so on the PR, with the earlier PRs that broke there.

What you get

Two examples we picked from public repos: approved PRs where the repo's own history pointed at the risk, and what happened after merge.

supabase/supabase-js
1. 1. ApprovedApproved by 1 reviewer, merged May 18, 2026
2. 2. Precall would have flagged itChanges packages/core/auth-js/src/GoTrueClient.ts. 15 of 33 earlier PRs that changed it needed a fix.
3. 3. Users hit itsignUp() returned no user, even though the user was created (#2390), 3 days later
4. 4. HotfixFixed in #2391, shipped in v2.106.2 7 days later
sveltejs/kit
1. 1. ApprovedApproved by 1 reviewer, merged Jun 2, 2026
2. 2. Precall would have flagged itChanges packages/kit/src/runtime/server/page/render.js. 8 of 29 earlier PRs that changed it needed a fix.
3. 3. Users hit itThe generated env.d.ts had a broken import path on Windows (#15959), 3 days later
4. 4. HotfixFixed in #15953, #15971 and #15977, shipped in 2.63.1 5 days later

Across 3,009 new PRs in 28 repos, PRs with any rule hit needed a fix 1.7x as often. How we tested

What Precall learned from Hono

983 merged PRs read · 87 needed a fix (9%)

Files that kept needing a fix

Changes to src/middleware/etag/index.ts: 6 of 21 needed a fix29%
Changes to src/types.ts: 15 of 47 needed a fix32%
Changes to src/helper/adapter/index.ts: 5 of 9 needed a fix56%

Full Hono report →

More sample reports, 15 public repos

Pricing

Priced per active developer.

Free

For open source maintainers and small teams.

$0no card

Runs as a GitHub App. One read-only clone of each repo to build rules, deleted when the run ends.

Unlimited public repos
Private repos free up to 3 active devs. Over that, private-repo comments pause the next month unless you upgrade
Comments on PRs that match your repo's rules
Rules refresh weekly, plus up to 2 manual reruns a day per repo

Team

For teams that want every PR checked against their own rules.

$20per active dev
a month

Only devs who opened a PR that month count. 14-day free trial, no card, once per account. You start it in the app after you sign up.

Runs as a GitHub App. One read-only clone of each repo to build rules, deleted when the run ends.

Unlimited active devs
Rules updated on every merge
Slack digest
Admin and audit log
Priority support

Enterprise: runs fully on your own machines, SSO, invoice billing. Talk to us.

Compare Precall with Qodo, CodeRabbit, Copilot code review and CODEOWNERS

Questions

Anything else? Talk to us.

Is this another AI review bot?

No. Precall does not write review text. It comments on risky PRs, using rules learned from your repo's own history, with the counts behind each rule.

How much will it comment?

One summary comment, edited on each push, and at most 5 inline notes. Nothing when no rule is breached or the PR only changes docs.

How do you know the rules work?

We tested the rules on 3,009 recent PRs in 28 public repos that the rules had never seen. PRs with any rule hit needed a fix 1.7 times as often. The result held in 10 of the 28 repos. See the proof page.

Does our code leave our machines?

The GitHub App reads your pull requests through the GitHub API. To build a repo's rules we make one read-only clone of it, then delete the clone and its token when the run ends. We keep the rules and reports, not a copy of your code. If your code must stay on your own machines, that is the Enterprise setup.

What if a rule is wrong for our repo?

Reply /precall ignore to mute it on that PR, or add it to muted in .precall/config.json to turn it off. Only people with write access can mute.

What if our repo is small or new?

A repo needs at least 30 merged PRs for a report; with fewer, the run stops and says so. If it has 30 but the history is thin, Precall says so and starts with clearly marked starter rules for your stack. Repos up to 1 GB.