// developer tool · early preview

Your PR queue,
always in view.

Perch Review sits above your other windows and shows every pull request waiting for your attention — with local AI analysis that never leaves your machine.

⬇ Download for Windows See features

// Early preview  ·  v0.2.0 · Windows x64 · macOS coming soon

Perch Review icon
Perch Review
// v0.2.0  ·  Windows x64
Windows 10 / 11  ·  MSIX installer
Download .msix
Note: Signed with a developer certificate — Windows will show a SmartScreen prompt on first install. See step 2 below. A commercially trusted certificate is planned for a future release.
1
Download & open the installer — double-click PerchReview-0.2.0.msix to launch the installer.
2
Trust the certificate (first install only) — right-click the .msix → Properties → Digital Signatures → Details → View Certificate → Install Certificate → Local Machine → Trusted Root Certification Authorities. Then run the installer.
3
Launch & connect — open Perch Review from the Start Menu. On first launch, connect a GitHub or GitLab Personal Access Token in Settings. Your PRs appear within seconds. The 14-day free trial starts automatically — no key needed.
🔒

AI that runs on your machine. Not ours.

Perch Review uses Ollama — a local model runner — for code analysis. Your source code, pull request content, and review history never leave your computer. No cloud AI, no telemetry, no data retention. Built for teams where privacy isn't optional.

// what it does

Everything you need to
stay on top of reviews.

A focused tool that does one thing well: shows you what needs your attention, right now.

⚡
Always-on-top HUD
Frameless overlay that sits above your editor and browser. See your PR queue without switching context.
🤖
Local AI Analysis
Powered by Ollama. Review summaries and code quality signals generated on your machine — offline capable.
🔗
GitHub & GitLab
Connect multiple accounts. Track PRs, assigned issues, and project board items across organizations.
📊
Review Age Tracking
Color-coded urgency signals. Instantly see which PRs have been waiting days without a review.
🎓
Train Your Own Model
Feed Perch examples from your codebase. The AI learns your team's conventions and review style.
💻
Windows & macOS
Native desktop app. Starts with your OS, stays out of your way, and uses minimal system resources.

// faq

Frequently asked questions.

Everything you need to get Perch Review running with local AI — and keep your team in sync.

Do I need Ollama to use the AI features?

Yes — but only for the AI analysis. Ollama is a free, open-source local model runner. Install it from ollama.com, pull any compatible model, and Perch connects to it automatically.

The PR monitor, GitHub & GitLab integration, review age tracking, and all other features work completely without Ollama. The AI panel is fully optional and can be toggled off at any time.

Which AI model gives the best results for code reviews?

We recommend qwen2.5-coder by Alibaba — currently the strongest open-source model for understanding code structure, following team conventions, and producing actionable review feedback across Go, TypeScript, Python and more.

GPU VRAM Model Pull command
No dedicated GPU qwen2.5-coder:7b ollama pull qwen2.5-coder:7b
6–8 GB VRAM qwen2.5-coder:14b ollama pull qwen2.5-coder:14b
16+ GB VRAM qwen2.5-coder:32b ollama pull qwen2.5-coder:32b

The 14b model offers the best balance of review quality and inference speed for most developer machines. If your Ollama instance runs on a server shared across the team, the 32b model delivers noticeably sharper results.

How do I switch or upgrade the base model?

Switching to a better model takes about two minutes and your entire team profile — naming conventions, architecture rules, code commitments — is automatically carried over.

  1. 1Pull the new model in your terminal:
    ollama pull qwen2.5-coder:14b
  2. 2Open Perch Review and navigate to KI-Training.
  3. 3Select the new model under Base Model and keep your existing custom model name.
  4. 4Click „Modell aktualisieren" — all rules are re-embedded into the upgraded model. The version counter increments automatically.
Can my whole team share one trained AI model?

Yes — this is one of Perch's core collaboration features. When you build a team model, your complete profile (naming conventions, architecture decisions, code commitments, review examples) is embedded directly into the model file.

Any team member who points their Perch instance at the same Ollama server — on the local network or via VPN — can click the model name in KI-Training and have all fields restored automatically. A version badge shows everyone which iteration of the model they're running, making it easy to see who's up to date.

Does my source code ever leave my machine?

Never. Ollama runs entirely on your hardware. PR diffs are sent to the local Ollama process via localhost — they never touch an external server, API, or cloud service.

Perch does not collect telemetry, usage statistics, or code snippets. Your GitHub and GitLab tokens are stored locally in your system's config directory and are never transmitted anywhere except directly to GitHub's and GitLab's own APIs.

When is the macOS version available?

macOS support is in active development. The core app is built with Flutter and already compiles for macOS — we're finalising the always-on-top window behaviour and the macOS release pipeline.

Send us an email and we'll notify you as soon as it's ready.