Questions, answered.

CodeVetter is a desktop-first AI code review workbench built for agent-generated code. It runs locally as a Tauri binary with a local SQLite database, reviews diffs through pluggable LLM providers, and preserves per-finding evidence — your repo never hits a central server.

Point CodeVetter at the diff from Claude Code, Cursor, or any agent. It inspects the diff, understands repo and prior intent, exercises changed behavior, and emits findings with evidence. Reviews run in the desktop app using your own LLM API keys, fully offline from any central server.

CodeVetter wraps an LLM in a review pipeline — rubric, diff framing, finding normalization — that catches what a single "review this code" prompt misses. Its public benchmark scores 29/29 expected findings vs. 27/29 for raw Claude, isolating what the harness adds over the model alone.

Yes. CodeVetter is a Tauri desktop binary with a macOS release. The review engine runs in the app with local SQLite storage, and your repo never leaves the machine. Only the LLM provider you configure receives code, over your own API key.

CodeVetter verifies agent-written code by running reviews, emitting verification handoff proofs with per-finding evidence, and re-checking that each issue is resolved before merge. Findings are fixed one at a time so you can confirm every change rather than accepting a bulk rewrite.

CodeVetter is a desktop alternative to hosted PR review bots. It runs as a Tauri binary on your machine, reviews agent-generated diffs locally, and stores review history in SQLite. You bring your own LLM keys, so no code is uploaded to a vendor server.

CodeVetter reviews diffs entirely on your machine. The Tauri desktop app inspects the diff, frames it with repo context, and runs the review through your configured LLM provider. Aside from the provider call, nothing is networked — no central CodeVetter server exists.

CodeVetter is built around evidence-backed review. Each finding carries per-finding evidence, the review emits a verification handoff proof, and the Unpack feature produces a whole-repo evidence-backed system brief. Review history is persisted to local SQLite so prior intent is preserved across sessions.

CodeVetter's Review screen handles commit-intent reporting and verification loops that previously lived in separate IntentDebugger and QaReplay surfaces. It inspects the diff, exercises changed behavior, and fixes one finding at a time, giving you a replayable review trail stored locally.

CodeVetter inspects each diff and frames it against repo and prior intent before reviewing. The Unpack feature generates a whole-repo evidence-backed system brief, and review history persisted in SQLite lets you recover what the agent was trying to do across sessions.

CodeVetter is a macOS desktop app (Tauri binary) for verifying agent-generated code. It reviews diffs with pluggable LLM providers, emits verification handoff proofs, and tracks shipped and verification telemetry on a Roadmap dashboard — all with local SQLite and no central server.

CodeVetter preserves evidence by design: every finding carries per-finding evidence, reviews emit verification handoff proofs, and the full review history is persisted to a local SQLite database. The Unpack feature stores a whole-repo evidence-backed system brief you can revisit later.

Yes. CodeVetter supports Anthropic (Claude), OpenAI (GPT), and OpenRouter as pluggable LLM providers. Keys are stored in your OS keychain and only leave your machine in the Authorization header of the review request you initiate. There is no CodeVetter-hosted model.

CodeVetter is a Tauri 2 desktop app that functions as a code quality workbench for agent-generated code. Its Rust backend manages a local SQLite database; the React frontend runs the review engine. It ships as a macOS binary via GitHub Releases with an auto-updater.

CodeVetter is local-first: the repo, diff, notes, and review history all live in a local SQLite database. There is no CodeVetter-owned server. The only third party that sees code is the LLM provider you configure, over your own API key — and you can pick any provider.