Critique Chat: Ask Your GitHub Repo Questions with Frontier Models
Connect GitHub, pick a repository, ask in plain language. Answers cite real paths and line numbers — included free for every signed-in user.
Chat with GitHub. Frontier models. No extra fee.
Pick a repo, ask in plain language, get answers backed by live GitHub search. The model menu matches in-app Critique Chat — including StepFun-3.5 Flash, Grok-4.2, GLM-5-Turbo, GPT-5.4-Mini, and more.
Why codebase chat matters
A knowledgeable person with an AI agent can outperform a strong software engineer who has zero context on a codebase. That is not a controversial claim anymore — it is the lived experience of teams onboarding into unfamiliar repos every week. The gap is not talent; it is information.
Critique Chat closes that gap. Instead of guessing where auth logic lives or grepping through a monorepo you inherited last sprint, you ask a question in plain language and get an answer grounded in actual file paths, line numbers, and recent PR history. The model is not pretending it read your repo; it is calling live GitHub search and file reads under the hood.
How it works
The lead model can call live GitHub code search, read individual files, and inspect PR metadata. Answers cite real paths so you can verify every claim. If the model searched three files to answer your question, you see exactly which three.
Free for every signed-in user
If you have a Critique account, Chat is included — no separate subscription, no credit spend for chat turns. You get access to the same selectable models we surface across the product: StepFun-3.5 Flash, GPT-5.4-Mini, GLM-5-Turbo, Grok-4.2, MiMo-V2-Pro, MiniMax M2.7, and KAT Coder Pro V2.
Why these models?
Model quality for “where does this live?” and “how does auth flow work?” is not a nice-to-have. It is the difference between a useful operator console and a generic chatbot. We want every team to feel that gap on real repos before they ever think about credits on PR review.
For context on how far open and near-open stacks have moved: we recently published a deep look at MiniMax M2.7 next to Claude Opus 4.6 — vendor benchmarks, our own head-to-head runs on bugs and security, and where each model still wins. The headline is not “M2.7 beats Opus everywhere”; it is that the distance is now close enough to force a real buying decision, especially on price-sensitive exploration workloads.
Read: Our Faith in Open Source
MiniMax M2.7 vs Claude Opus 4.6 — benchmarks, economics, and what it means for code review pricing.
Read the full breakdown →Use cases
- “Where is the auth middleware? Show me the file.”
- “Explain the data flow from API route to database.”
- “What changed in the last 5 merged PRs?”
- “Summarize the open PRs and their review status.”
- “Does this repo have CI configured? What does it run?”
- “Find all usages of the deprecated fetchUser function.”
Evidence-backed answers
Every answer that references code includes the file path and, where applicable, line anchors from live GitHub search. If the model calls three tools to answer your question — two code searches and a file read — you see all three in the activity stream. Transparency is not optional when you are making engineering decisions based on AI output.
From “understand this repo” to deeper review
Chat is the lightweight on-ramp. GitHub-grounded answers, frontier-class models while logged in, and a straight line from “understand this repo” to Critique’s deeper PR review stack when you are ready. Start with questions, graduate to automated code review, security scanning, and Remedy fix suggestions.
Try Critique Chat now.
Connect GitHub, pick a repo, and ask your first question. Frontier models included for every signed-in user.
Open Chat →