A second reviewer that remembers your repo vs per-seat AI review tools
Most teams already review pull requests. The question is whether a second reviewer adds findings the first review missed. Two things separate the options: where the reviewer runs, and how it is priced.
This page compares the per-seat AI review tool category with a reviewer Teammate on a dedicated agent server. It describes categories by design, not individual products.
What a per-seat review tool does
A per-seat AI review tool runs on a hosted service. It looks at each pull request when it is opened or updated and comments on the change. It is priced per developer who uses it.
That design is simple to adopt and easy to budget for a small team. The cost grows with each developer who uses it, and how much it carries from one review to the next depends on the tool.
Why memory of the repo matters in review
Bugs repeat. A team that fixes the same class of mistake three times has a pattern, and a reviewer that does not know the pattern will meet it fresh every time. Writing the pattern down in a file in your repo means any reviewer, human or Teammate, can read it, and you can edit it.
Where a per-seat tool is the better fit
If one or two developers want comments on their own pull requests and nothing else, a per-seat tool is a direct answer. There is no server to think about and the bill is easy to predict. If you do not want a long-running server in your setup, that is a fair reason to choose it.
The two can also run together. A second reviewer is only useful if it says something the first one did not, and you can judge that by reading the findings side by side.
What a reviewer on its own server does differently
The QA Teammate reviews the pull requests your team already opens. It leaves inline findings checked against the issue and your repo's conventions, and notes whether a PR changes its own tests. It runs on an agent server: an always-on workstation one or more Teammates run on, with its own files and its own tools.
A failure-pattern file in your repo records the classes of bug your team has seen. The reviewer reads it each time it reviews, so a pattern you have written down gets caught at review next time. If you later add an engineer Teammate, it can mine past review comments into that file. Teammates also keep memory in three tiers, and private conversations are never rolled into shared summaries.
The reviewer gets a reviewer profile: it can read code and review, approve and comment, but cannot push or edit CI and workflow configuration. Once GitHub is connected, PR, review and comment events arrive without per-repo webhook setup. See Integrations. Because the server is shared by Teammates, the same server can later host an engineer too. That is optional.
How you get started once the readiness gate passes
- Connect GitHub from the reviewer's Tools page and deploy. From then on, PR, review and comment events arrive with no webhook to configure per repo.
- The reviewer comments on the pull requests your team already opens. Nothing writes code yet.
- Read the findings next to your existing review. Keep it if it earns its place.
- If you want more later, add an engineer and a product Teammate to the same server. That is the engineering lane described in the other comparison.
Side by side
| Per-seat AI review tools | Reviewer Teammate | |
|---|---|---|
| Where it runs | A hosted service, per PR | Your agent server, always on |
| What it reads | The change, with the context the service holds | The diff, the issue and the failure-pattern file |
| What carries over | Depends on the tool | Patterns in a failure-pattern file in your repo, read on every review |
| Pricing unit | Each developer who uses it | Seats for your people, plus the server, plus model usage |
| Scope | Review | Review now; the same server can host the full engineering lane |
What we have seen so far
In our own dogfood, the reviewer surfaced useful findings the existing review missed on about half the PRs it reviewed.
It is a second reviewer, not a replacement for your existing review. A measurement window is running now. We have not published a precision figure yet.
A worked monthly example
You pay for seats (the people on your team, not Teammates), plus the server, plus model usage billed at published provider list rates. Templates and setup are free. One engineering lane runs on one server, with several Teammates sharing it. A per-seat tool prices each developer who uses the reviewer.
| 2 seats (Pro Team) | $100/mo |
|---|---|
| Server: Standard Medium, 2 vCPU / 8 GB | $108/mo |
| Disk: 20 GB (default) | $2.40/mo |
| Model usage | Billed at provider list rates, depends on volume |
| Fixed monthly before model usage | $210.40/mo |
Seats, the server and disk are billed monthly. Model usage is debited from a prepaid balance; see Team and billing. Current rates are on the pricing page.
Limits
- The reviewer does not run your tests or read CI today.
- Linear is the only issue tracker supported today.
- We do not have a SOC 2 report yet.
- The reviewer and the engineering lane run on our own repositories today. Before either runs on yours, a readiness gate must pass, including an option to keep review findings private and a visible verdict on the final commit of most PRs.
Next steps
When the readiness gate passes, the entry will be a second review of the PRs you already have, before anything writes code. Until then, talk to us.