Competitive landscape

How CodeVetter is positioned against hosted PR-review bots, agent orchestrators, and Claude Code /review.

Competitive landscape

CodeVetter is a local-first desktop workbench for verifying agent-generated code (no server, no auth, local SQLite). This page positions that against the market. The exhaustive March 2026 competitor survey and the (now-obsolete) April 2026 hosted-GitHub-App go-to-market plan are archived in stale-competitive-landscape-2026-03.md — kept for history, not current.

Two market categories

A. Hosted PR-review bots — Greptile, CodeRabbit, Ellipsis, Qodo, Sourcery, Bito. Install as a GitHub/GitLab App, review every PR automatically, compete on accuracy / false-positive rate / learning-from-feedback / custom rules / platform breadth. Pricing roughly $12–40/user/month. Greptile leads on catch accuracy (~82% claimed, codebase graph); CodeRabbit leads on adoption (2M+ repos).

B. Agent orchestrators — commercial, open-source, and multi-agent desktop apps that run multiple AI agents in parallel with basic built-in diff viewers to skim agent output. Free to ~$20/month. No deep AI review.

The built-in baseline — Anthropic Claude Code /review: multi-agent parallel review with a verification step, deduped and posted as inline GitHub comments. Cloud-run, ~20 min/review, ~$15–25 per review on Claude Teams/Enterprise quota. This is the tool every other reviewer justifies itself against, and it is CodeVetter’s most direct comparison because CodeVetter also runs multi-pass specialist review with a verification/proof step.

Where CodeVetter sits

CodeVetter straddles both categories: a desktop app (like B) that does deep code review (like A), specialized for agent-generated code.

• Opportunity: no single tool does both well. Orchestrators have shallow review; PR bots have no agent orchestration and are cloud-hosted.
• Local-first is the wedge, not a limitation. The archived plan proposed a hosted GitHub App to compete on distribution; the product deliberately went the other way — everything runs on the user’s machine, offline, against local SQLite. That is the differentiator vs. every Category A tool, which requires granting a cloud service repo access.
• Risk to manage: being seen as “worse than an orchestrator at orchestration and worse than Greptile at review.” The answer is the verification loop below, not feature parity.

What CodeVetter does that the bots don’t

CodeVetter’s on-strategy angle is trustworthy agent output, not generic PR review (see product/overview.md):

• Risk-tiered multi-pass review with security/product/agent specialists and coordinator dedup — see architecture/review-pipeline.md.
• Verification proof: review-proof + agent-fix-packet with per-finding fixed/reproduced/unchecked tallies, not just static diff judgment.
• Repo understanding + history: evidence-backed repo briefs (repo-unpacked.md) and a release-history workbench (graph-and-history.md).
• Synthetic user QA to exercise the changed behavior (product/synthetic-user-qa.md).
• Benchmarked catch-rate evidence rather than marketing accuracy claims — see development/benchmark.md.

Table stakes (present in most Category A tools)

GitHub/GitLab bot integration, custom review rules (English or YAML), learning from feedback, PR summaries, and one-click inline suggestions. CodeVetter’s local-first posture means it does not compete on cloud-bot distribution; it competes on depth of verification and on keeping code on the user’s machine.

Open competitive questions

• What does CodeVetter’s review catch that Claude Code /review does not, at a lower incremental cost? (This is the head-to-head the benchmark exists to answer; real agent-PR case curation is still pending — see ../../STATUS.md.)
• Is the agent-generated-code specialization a durable moat, given that codebase-aware bots apply their graph to agent code too?

For the full per-competitor breakdown (pricing, features, unique angles as of March 2026), see the archived survey.