RHOBEARReviews Install on GitHub
Reviewing live · 78% SWE-bench

AI code review
that doesn't sleep.

One free review per day. Every PR. Every repo. No per-seat tax, no minimums, no sign-up beyond installing the GitHub App. RHOBEAR Reviews lands a verdict in under 2 minutes — half the time of CodeRabbit, a third of CodeAnt — so you fix, push, and merge while you're still in the headspace.

No credit card · Install in 20 seconds · Free forever
rhobear · review.log · live
gh pr view 142 → @rhobear review
→ fetching diff · 87 files · +1,243 / -318
→ scoring 9 categories of risk
 
[HIGH]src/auth.py:142 — token compared with ==
Risk: timing attack on credential check
Fix: hmac.compare_digest()
 
[HIGH]src/api/upload.py:88 — path traversal
[MED]tests/__init__.py:1 — missing coverage
 
VERDICT: REQUEST_CHANGES · 2 blocking
posted to PR #142 · 89s total
78%
SWE-bench score
Real GitHub issues solved
204K
Context window
Full PR + repo context
<2 min
Median review time
Mention to verdict
$0
First daily review
No card, no signup
How it works

Install. Open a PR. Tag @rhobear. Done.

No new dashboard to learn. No webhook to configure. RHOBEAR Reviews lives where your code already does — in the PR thread, posting structured findings with file:line evidence and a binary verdict your team can act on.

01

Install the GitHub App

Pick the repos you want reviewed. Free covers public; Pro covers private. Takes about 20 seconds.

02

Open a PR — or comment

New PRs trigger us automatically. Existing PRs: drop @rhobear review in any comment.

03

Reviewed with scrutiny

Security, correctness, scope creep, missing tests. Every finding gets file:line evidence and a fix.

04

Binary verdict on the PR

ACCEPT, REQUEST_CHANGES, or COMMENT. Posted as a PR review your team can resolve.

Full codebase context · NEW

Reviews that know your whole codebase, not just your diff.

Most AI reviewers can only see what's in the pull request. They miss the function callers your change just broke. They miss the invariant this method was designed to protect. They miss the test pattern your codebase has used for three years. We don't. On install, we read your entire repo — every file, every function, every cross-reference — and we keep it up to date as you ship.

1 · INSTALL

We read every file in your repo

Deep ingestion on day one. We map the structure — imports, exports, call sites, type relationships. Happens once.

2 · DAILY

Incremental updates as you ship

When you push, we re-read only what changed. The map self-updates without re-ingesting the whole codebase. No daily cost spikes. No latency tax.

3 · REVIEW

Every review pulls from full context

Reviews query the map for the callers, the invariants, the patterns. The verdict reflects your whole codebase, not 87 lines of diff.

The result: a 12-line PR can surface a bug in the caller it just broke.
your_repo · live map live
Files mapped
341
Functions tracked
12,407
Last full read
47 days ago
Updates today
23 changed
Pricing

Built for the team that ships, not the team that signs seats.

Per-review, not per-developer. Your 12-person OSS project pays the same as your solo side project.

Free
$0 / forever
No card. No trial. Just review.
  • 1 review per day, any repo
  • Full reviewer — same engine as Pro
  • Public repos · GitHub App install
  • Posted as a PR review with verdict
  • Private repos require Pro
Install free →
Pro
$19 / month
100 reviews · all repos · no per-seat fee
  • 100 reviews per month, public or private
  • Unlimited team members — no per-dev billing
  • Auto-review on every new PR (configurable)
  • @rhobear escalate for higher-budget scrutiny
  • Priority queue when traffic spikes
  • Email + Slack digest of weekly findings
Start Pro · $19/mo →

CodeRabbit Pro = $24 / dev / mo  ·  CodeAnt = $24 / user / mo  ·  They charge per seat. We charge per review. Ship faster, pay for what you use.

Benchmarks · receipts

The exact scores. No hand-waving.

We don't hide our receipts. RHOBEAR Reviews is built on a top-tier coding model (MiniMax-M2.7). Here's where it sits on the benchmarks that matter for code review — solving real software engineering tasks.

Model
SWE-bench
SWE-Pro
Context
MiniMax-M2.7 RHOBEAR
78.0%
56.22%
204K
Claude Opus 4.6
~73%
~57%
200K
Claude Sonnet 4.5
~69%
~48%
200K
GPT-5 (gpt-5-turbo)
~71%
~52%
256K
DeepSeek V4
~68%
~46%
128K

SWE-bench / SWE-Pro: % of real GitHub issues from open-source repos that the model can solve from a PR diff. Higher is better. Sources at the foot of this page.

Receipts, not marketing

The same brain reviews our code. We just sell the seat next to ours.

RHOBEAR Reviews isn't a side hustle. The exact reviewer you install is the one we trust on our own ship-blocking PRs, every day. You're not getting a stripped-down trial — you're getting the production engine.

Battle-tested on our own code

Every line we ship runs through this reviewer first. If it's not good enough for our merges, it's not good enough for yours.

Your code never trains anyone

Reviews run inference-only. No diff retained past the review. No data sold. No fine-tuning on your repo. Ever.

Hosted on our always-on infra

Cloud-resident, owner-operated. No SaaS middleman. No Zapier in the loop. If we're down, we already know — it's our pager too.

receipt · single review v1.0
ReviewerRHOBEAR Reviews
PR scanned87 files · ~25K LOC
Categories checked9
Latency p50< 2 min
Findings emitted3–12 per PR
SWE-bench score78.0%
Your cost$0.19
At Pro · $19/mo ÷ 100 reviews. Free tier: $0.
FAQ

Honest answers to the obvious questions.

We're a small shop. We answer real emails. Anything not here, ping us.

What's a "review"?

One pass over one pull request. Mention @rhobear ten times on the same PR — still one review. Open a new PR — that's review two.

Do I need a credit card for the free tier?

No. Install the GitHub App, get one review per day, period. We'll only ask for payment if you want Pro.

How is this different from CodeRabbit or CodeAnt?

Three things. Pricing — we charge per review, not per developer. A 10-person team pays the same as a solo dev. Speed — verdicts land in under 2 minutes, where competitors take 2-5 minutes of thinking before they post anything. Context — our Codebase tier reads your whole repo, so reviews catch the bugs in callers most tools never see.

What does it actually look for?

Correctness bugs, security flaws (injection, auth bypass, secret leaks), scope creep, missing tests, broken contracts, performance regressions, error-handling gaps. Each finding gets a severity, evidence, and a one-line fix recommendation.

How does full codebase reading work?

On install, we read every file in your repo and build a queryable map — imports, exports, call sites, type relationships. After that, pushes only update what changed. Every PR review reflects your whole codebase, not just the diff.

What happens to my code?

Pulled via GitHub App permission, sent inference-only for the review, dropped from memory the moment the review posts. We don't store diffs. We don't train on your code. We don't sell anything to anyone.

Does it work on monorepos / large PRs?

A 204K-token context covers the vast majority of real PRs. For monorepo behemoths, the Codebase tier's repo map lets us review one subsystem at a time with full cross-reference awareness.

What if I burn 100 reviews mid-month?

You can buy a top-up pack (50 reviews for $9) or wait for the next billing cycle. No surprise charges, no auto-overage.

Stop paying per developer.
Start paying per review.

Install the GitHub App. First review is on us — every day, forever.