Runs today · For engineering teams
AI code review on every pull request
A reviewer agent that reads the diff in a clean context, cites file and line, runs your gate, and reports only what it can prove. Bring your own model key.
The job to be done
Teams shipping to GitHub who want a first pass on every PR without paying per developer for it.
Review queue grows faster than review time. The first pass — style, obvious bugs, missing tests, a migration without a rollback — is mechanical, and it is the pass that slips when the sprint gets loud. Hiring another reviewer does not scale; neither does asking the author to review themselves.
How it works
- 1
The PR opens
A webhook hits an automation trigger. The run carries the repo, the PR number and the head SHA — nothing else is trusted from the payload.
- 2
The diff is read in isolation
The reviewer phase runs with `isolatedContext`, so it reviews the code and not the author’s narration of the code. Only the changed files and their immediate neighbours enter the context.
- 3
The gate decides
Lint and tests run as literal commands. A run that ends with a red gate is a failed run, whatever the model said on the way out.
- 4
Findings land on the PR
Every finding cites a file and a line, or it is dropped. Comments post through the GitHub connector; a human still merges.
Works with
Each entry is a connector or channel available on the platform today.
What it costs
What the incumbent charges
CodeRabbit Essentials: $24 per developer/month billed annually; Team $48, Advanced $72. Cloud agent minutes are $0.40 each on top.
Where they are the better choice
One-click fixes applied to your branch, learnings that persist across reviews, and Jira/Linear integration ship today. We do not offer any of those yet.
Source, checked 2026-10-07 — CodeRabbit pricing (Essentials/Team/Advanced, agent minutes)
On SynthHires
SynthHires
Free tier, then $9/month. You pay your model provider directly — we take no markup on tokens.
- No per-developer seat: the platform is free to start and $9/month after, while the model cost lands on your own key.
- Same repository, same pull request, no vendor-hosted copy of your code beyond the diff you send.
- The reviewer is a configurable agent: change the prompt, the gate, or the model without waiting for a vendor roadmap.
Your API keys are encrypted locally with AES-GCM and never travel to our servers.
See pricingWhere this is not the right fit
- CodeRabbit ships one-click fixes, cross-repo context and issue-tracker integration today. This workflow comments and gates; it does not patch your branch.
- You assemble and maintain the workflow yourself. That is freedom and it is also your maintenance.
What it handles
Proof, not opinions
A review comment that cannot point at a line is discarded. The gate’s exit code — not the model’s self-report — decides whether the run passed.
Cost on your key
The review runs against the provider key you already configure. We never see it, and there is no per-developer seat to buy.
Repository-scoped access
The agent gets the diff and the files it needs. Capability policy and allow-listed hosts keep it from wandering off the repo.
Audit trail per run
Each run stores its phases, its gate output and its cost, so a review can be argued with months later.
Availability: this workflow is assembled from primitives that ship today. You configure the trigger, the agents and the gate; nothing is hidden behind a feature flag.
Frequently asked questions
Does this replace my human code review?
No, and the design assumes it should not. The agent takes the mechanical first pass — missing tests, obvious defects, style — and every merge still goes through a person. The gate decides whether a run passed; a human decides whether the change ships.
Where do my provider keys live during a review?
Tus API keys se cifran localmente con AES-GCM y jamás viajan a nuestros servidores.
What does it cost to review 200 pull requests a month?
The platform is free to start and $9/month after that. The tokens are billed by your provider at your provider’s rates — we add no markup, so the only variable is the model you choose and how large the diffs are.
Which languages and frameworks does it understand?
It is model-dependent rather than language-dependent: whatever your key can run, the reviewer can read. Lint and test commands are yours to configure per repository, which is also what makes the gate trustworthy.
Claims on this page last checked against the source on 2026-10-07.
- Greptile pricing (Starter free, Pro $30/seat/month) (2026-10-07)
- SynthHires pricing (free tier, $9/month) (2026-10-07)
Deploy this workflow with your own keys
Free tier, no card. Bring your provider key — it is encrypted in your browser and never reaches our servers.