Providers compete for every request
The buyer scores providers on price, latency and reputation and sends each request to the best one. When a provider gets cheaper, your reviews do too.
AI code review for GitHub, on Antseed
The cheapest AI code reviewer.
reviewant buys its models on Antseed, an open peer-to-peer AI network where independent providers compete on price. You get a thorough review of every pull request and pay what the models cost, with no markup.
Install the GitHub App, pick the repositories. Every new PR head is reviewed; tag
@reviewant to ask for more.
reviewantcommented · 2 minutes after the push
Review summary
Moves session refresh into the token middleware and drops the per-route check.
- P1Refresh races a concurrent sign-out: a revoked session is re-issued for up to 60 s.
- P2
/admin/exportno longer checks the role it used to.
3/5
Fix the refresh race and the export check before merging.
- if (session.expiresSoon()) return refresh(session)+ if (session.expiresSoon() && !session.revoked)+ return refresh(session, { ifVersion: session.version })
Tests: 412 passed · Files read in full: 6 · Model cost: $0.02
Reviewing every pull request of the Antseed monorepo, and its own.
Why it’s cheap
Models from an open market, not a reseller.
Antseed is a peer-to-peer network for AI services. Independent providers offer models on it, and requests go straight to them over an encrypted connection, with no company in the middle. reviewant is a buyer on that network.
You pay what the models cost
Model usage is passed through at the network’s price, with no markup and no per-seat fee on top.
Paid per request, by card or USDC
Every request is paid from a balance funded by card or with USDC on Base. No commitment, no unused capacity.
How it works
From install to the first review in minutes.
-
01
Install the GitHub App
Choose the repositories. They are watched from that moment on, and repositories you add later follow.
-
02
Open a pull request
Each new head commit gets a review once pushes settle. A burst of pushes is reviewed once, not ten times.
-
03
Talk to it in the PR
Tag
@reviewantwith a focus (“check the retry path”) or “re-review” after a fix. The new review replaces the old one.
What you get
A reviewer that does the reading.
Not a diff summariser. reviewant works in a real checkout of your branch, with the tools a person would use.
Context beyond the diff
Changed core files are read in full, callers and tests are looked up, and regressions come first.
Your tests, actually run
Dependencies are installed and the suite runs in a sandbox, so a finding can say what failed.
A confidence score
Every review ends with N/5 and one sentence on what stands between this PR and a merge.
Findings with fixes
P0–P3 severity, the exact lines, why it breaks and a suggested change. Nothing vague.
A code quality pass
A second agent looks only for reuse, simplification and consistency, kept apart from the score.
Your rules
Guidelines per repository or organization, plus your CLAUDE.md or AGENTS.md.
Large PRs, split
Big changes are reviewed in parallel passes by area and merged into one review.
Diagrams when it matters
When a PR changes how components talk, the review draws the new flow.
It never merges
No approvals, no pushes, no requested changes. It comments; your people decide.
Security
Your code runs in its own sandbox.
Each review of a hosted organization runs as its own job in a separate cluster, and is deleted when it ends.
- gVisor sandbox per review. PR code, installs and tests run in a user-space kernel, never on a shared machine.
- Internet only. No route to private networks, cloud metadata or the cluster’s API.
- Read-only, one repository. The review gets a GitHub token that can read that repository and nothing else. Posting stays outside the sandbox.
- No shared secrets. The service’s own credentials never enter a sandbox.
- Redacted output. Emails and token-shaped strings are removed before anything is posted.
- Your organization only. Admins see their own repositories, settings and history in the dashboard, and sign in with GitHub or a passkey.
Pricing
Hosted, or on your own infrastructure.
Either way, models are bought on Antseed at the network’s price: you pay for what your reviews use, not a markup.
Hosted
Subscription
per organization, plus model usage at cost
- Install the GitHub App and go
- Isolated sandbox for every review
- Tests run in the sandbox
- Dashboard for your organization
- Guidelines per repository
- Updates and support
Deploy your own
Open source
GPL-3.0, on your own Kubernetes
- The same reviewer, on your machines
- Your own GitHub App or bot account
- Python standard library, no build step
- Models bought on Antseed
- Community support
FAQ
Questions.
Which models does it use?
A chain of models bought on Antseed, with fallbacks when one is slow or down. The chain is set per repository; the hosted service picks a tested one for you.
Why is it cheaper than other reviewers?
There is no per-seat price with model costs built in. reviewant buys its models on Antseed, where independent providers compete for each request, and passes the cost through.
Does it store our code?
No. Hosted reviews check the branch out inside a sandbox that is deleted when the review ends. What is kept is the review itself and its history in your dashboard.
Who can trigger a review?
Every new PR head is reviewed automatically. Tags are accepted from people with write access, or from an allowlist you set.
Can it run our tests?
Yes, for pull requests from branches of the same repository, inside the review’s sandbox. Results go into the review.
Does it work with GitLab or Bitbucket?
Not yet. reviewant is GitHub only today.
Can we try it before paying?
Yes. Ask for access and we will set up your organization for a trial.