A press-ready view of the GitHub-native review layer for teams shipping with coding agents, fast humans, and production systems that still need evidence before merge.
Critique is not a chatbot commenting on diffs. It is a review system with a visible chain from triage to specialist analysis to repair.
AI has made code cheaper to produce, but not cheaper to trust. Critique is the verification layer for teams shipping with agents.
A GitHub App that maps pull request risk, routes specialist review lanes, and returns an evidence-backed merge verdict.
Teams are adopting coding agents faster than they are upgrading review systems. That gap is becoming an engineering risk.
The first serious beta month crossed roughly a billion tokens processed across review, Chat, Builder, and Remedy surfaces.
Ray is available for short product demos, founder quotes, screenshots, and technical background.
Pick the angle that matches the outlet. The architecture stays the same; the lead changes.
The first pass decides what changed, which files matter, and which risks deserve deeper review before specialist agents spend tokens.
Security, test coverage, architecture, and performance checks are treated as different jobs instead of one generic model response.
The final response prioritizes findings with severity, evidence, affected paths, and a recommendation the maintainer can act on.
When the issue is localized and testable, Remedy can attempt a bounded patch in an isolated environment and hand back validation context.
An Ireland-built developer tool working on the verification side of the AI coding wave.
A hands-on build story with public beta volume, fast product expansion, and a clear thesis about trust.
How teams should separate coding agents, review agents, and fix agents before requiring AI checks on protected branches.
Multi-model routing, credit accounting, and evidence preservation as the operating system around generated code.
Founder interviews, technical background, screenshots, and quote checks go through Ray. Product support still goes through the support mailbox.
Search paths
These are the public pages that explain the category, compare Critique with known competitors, and give reporters or buyers the clearest product context.
Harness
Agentic coding harness, same class as Codex. TTY or critique-code web on 127.0.0.1. Binary critique-code from @critiquedotsh/harness. Trial on Verify.
Install
npm install --global @critiquedotsh/cli — local working-tree finish pass for coding agents.
Author
npm install --global @critiquedotsh/harness — author agent (critique-code). Same org as the sidecar CLI, not a rename of it.
Author
0.1.3 vs 0.1.0: login, critique/auto, voice, skills, workers, and critique-code web. Forced review did not move.
Author
critique-code web: local browser on 127.0.0.1, same forced-review author kernel.
Docs
finish.v1 contract, repair modes, integrate for Codex, Claude Code, OpenCode, and Cursor.
Alternatives
Ranked shortlist: Critique CLI finish pass vs PR comment bots and implementation agents.
Compare
Head-to-heads vs CodeRabbit, Greptile, Copilot review, Bugbot, Codex, Claude Code, Devin.
vs CodeRabbit
Pre-commit finish.v1 evidence vs seat-priced PR comments — different moment in the loop.
vs Claude Code
Claude Code writes the change; Critique checks it with structured exit codes.
vs Codex
Complementary: Codex implements, Critique finishes — not same-family self-grade.
Pricing
Shared credits (Solo $19 · Pro $49 · Team $149), student/OSS lane, BYOK harness.
Agents
Parseable plan table for AI shopping agents — no JavaScript wall.
Category
Category guide for independent finish passes vs CI and PR bots.
Essay
Why an independent finish pass differs from an agent grading its own homework.
Essay
“Looks complete” is not verified — when to call a second opinion.
llms.txt
Canonical product description, install workflow, and citation URLs for answer engines.
Essays
Product notes, buyer guides, and release updates.