Critique PR Review v4: The Beta Is Getting Real

Founder note

v4 is the release where the beta stops feeling conceptual

There is a moment in an infrastructure product when the language around it has to change. In the earliest phase, you are mostly explaining the idea. Then you are explaining the architecture. Then, if things go well, you reach the point where the product starts generating enough real-world movement that the story is no longer theoretical. Critique PR Review v4 feels like that moment for us. We are still in beta, yes, but we are no longer talking about a speculative direction. We are talking about a live system that has now reviewed nearly 4,000 pull requests and is getting stronger in a way that is visible both inside the product and outside it.

That is the frame I want for this post. Not hype for hype’s sake. Not pretend-final language. Just a clean statement of where the product is. v4 is the release where the operating model gets tighter, the public proof gets louder, and the beta starts feeling like a product with momentum rather than a product with a pitch deck.

What v4 means

The review engine is getting closer to its own execution plane

The most important architectural direction in Critique PR Review has been consistent for a while now: the app should be the control plane, and the sandbox should be the execution plane. Identity, policy, routing, persistence, GitHub publication, review-run pages, and downstream remedy orchestration belong in the app. Repository inspection, deterministic collection, targeted commands, and model execution belong as close to the checked-out code as possible. v4 keeps pushing in that direction.

In practical terms, this release tightens the runtime around official OpenCode server semantics inside E2B-backed sandboxes rather than relying only on one-shot CLI behavior. That sounds like a narrow implementation detail, but it matters because it makes the review path more native to the execution environment that actually inspected the code. Cleaner session semantics. Cleaner runtime truth. Less translation between the place that saw the repository and the place that speaks for the review.

The usage signal

Nearly 4,000 PRs reviewed is the real headline

Products like this are easy to over-explain in technical language and under-explain in operational language. So I want to say the operational part clearly: Critique has now reviewed nearly 4,000 pull requests during beta. For me, that number matters more than any single internal diagram. It means the system is being exercised enough that edge cases, architectural pressure, product sharpness, and trust surfaces are no longer abstract concerns. They are product realities. They show up in the work every day.

That scale is still tiny relative to where we want to go, but it is already large enough to force honesty. You cannot hide behind a nice landing page when real repositories, real diffs, real engineering teams, and real expectations keep pushing against the system. v4 is stronger because the beta has now had enough mileage to demand it.

The public proof

OpenRouter now shows Critique at #2 for apps using Embed V1 4B this month

There is another reason I wanted this note to exist now. On OpenRouter’s activity view around `perplexity/pplx-embed-v1-4b`, Critique is currently showing up as the number two public app this month in the “Apps using Embed V1 4B” list, sitting just behind AnythingLLM. I love that signal because it is specific, external, and hard to fake. It is not us declaring ourselves important. It is a public ecosystem surface reflecting that Critique is actually being used in the wild on top of that embedding lane.

That does not mean “we won.” It does not mean the product is done. It does not mean every part of the review stack is perfectly proven. But it does mean we are past the stage where the company has to invent all of its own validation. When a public model ecosystem starts surfacing your product in usage-ranked views, the market is quietly telling you something. In our case, the message is simple: Critique is becoming real enough to show up on other people’s infrastructure.

Why the beta language stays

Because trust is earned by precision, not by pretending

I do not want to remove the word beta until it stops being true. There is still hardening work ahead. More runtime coverage. More proof across broader repository shapes. More product smoothing around the review-to-remedy boundary. More repetition. More edge-case mileage. The thing I want Critique to become is not “the loudest AI code review product.” I want it to become one of the most trustworthy ones. That requires resisting the temptation to use overconfident language one release too early.

So yes: PR Review v4 is live. Yes: we have reviewed nearly 4,000 pull requests. Yes: public usage surfaces are starting to notice. And yes: we are still in beta. All four of those statements belong together. That combination is exactly what makes this stage interesting.

What I think v4 really is