GitHub App · reviews every PR

Orvex catches bugs before production.

Orvex reviews every pull request the moment it opens. Deterministic rules run first, then a multi-model AI ensemble weighs in, and it runs your code — not just the diff. It flags the bug an instant before merge.

No passwords pasted · scoped GitHub App · priced per workspace, not per seat
131bugs caught that other bots missed
3independent models per PR
~8comments max — signal, not noise

// THE PIPELINE

Three passes. One verdict.

Every PR runs the same route — cheapest checks first, most expensive last, nothing wasted. Different layers catch different bugs.

01 · FIRST

Deterministic rules

Cheap, exact checks run before any LLM token is spent. Config-as-code — your repo sets the rules in .orvex-review.yml.

0 tokens · instant
02 · THEN

3-model ensemble

Up to three independent models review one PR. Different models catch different bugs — the merged, deduped result is the differentiator.

A · B · C → merged
03 · PROOF

Runs your code

Orvex executes the change in a sandbox instead of only reading the diff — the runtime-proof moat. Add @orvex deep for critical PRs.

sandbox execution

// HEAD-TO-HEAD · 80 REAL PRs

Flagged 131 issues the other reviewers didn't.

Run side-by-side with Codex, CodeRabbit and Qodo on real pull requests. Different models catch different bugs — the ensemble surfaces what a single pass walks right past.

131 issues only Orvex caught
— missed by every other bot tested
…and the widest coverage overall defect-cluster coverage, 80 PRs
Orvex 43%
Codex 29%
CodeRabbit 28%
Qodo 22%

No single model catches everything — that's the whole point of the ensemble. Coverage = share of all issues flagged by any reviewer across 80 real PRs; “only Orvex” = clusters no other bot flagged.

// IT REMEMBERS

Never repeats itself.

Push three times, get zero repeated nits. Orvex tracks what it already said and replies “Fixed” when you actually fix it. Roughly eight comments, max — signal, not noise.

a1f9c02 retry-queue: backoff + jitter P1 · null deref
7d2e5b8 guard null on retry path ✓ Fixed — verified
b93aa10 tidy tests + rename 0 repeated nits
  • Says it once. No re-litigating a line you already handled.
  • Confirms “Fixed” when the change actually resolves it.
  • Caps the noise at ~8 comments — a review you'll actually read.

// PRICING

Priced per workspace, not per seat.

Every account starts with 10 free reviews — no card required. Scale from a first repo to bring-your-own-LLM enterprise.

Starter
$29/mo
100 reviews · then $0.50/review
  • 2 AI models per PR
  • Deterministic rules pass
  • Config-as-code (.orvex-review.yml)
Get Starter
Pro Unlimited
$69/mo
unlimited reviews
  • 2 AI models per PR
  • No per-review overage
  • Never-repeats memory
Choose Pro
Verify Lite
$49/mo
50 reviews · then $0.75/review
  • 3 AI models per PR
  • Premium track, budget entry
  • @orvex deep on demand
Choose Lite
Verify
$99/mo
120 reviews · then $0.75/review
  • 3 AI models per PR
  • Runs your code in sandbox
  • The full runtime-proof moat
Choose Verify
Enterprise
Custom
volume + governance
  • SSO / SAML
  • Bring-your-own-LLM key
  • SLA & support
Talk to us

Every plan starts with 10 free reviews — no card required.

Ship it. Orvex already read it.

Install the GitHub App and your next pull request gets reviewed — before it merges.

Installs scoped: pull_requests read/write, contents read · no passwords pasted