La Review

Turn a diff into a review checklist.

---

LaReview turns a PR or unified diff into a task tree so you can review changes in a deliberate order instead of scrolling. It’s local-first: state in SQLite, PR fetch via gh, and plan generation via an ACP agent you run.

Demo

demo.mp4

Key Features

LaReview is designed for a local-first, secure, and focused review experience.

	Feature
	AI-Powered Plan Generation Input a GitHub PR reference (e.g., owner/repo#123) or paste a raw diff. LaReview uses your local ACP agent to analyze the changes and generate a structured review plan based on the author's intent.
	Structured Task Tree & Notes Navigate the review as a hierarchical tree. Mark tasks as To Do, In Progress, or Done. Attach contextual notes to specific tasks or lines of code to keep track of your thoughts.
	Task-Focused Diffs Stop context switching. When you select a task, the diff viewer only shows the specific hunks relevant to completing that task, isolating the noise.
	Agent-Generated Feedback Threads The AI agent doesn't just plan; it actively identifies issues. It can autonomously create feedback threads (nitpicks, blocking issues, or suggestions) anchored to specific lines of code, helping you catch bugs before they merge.
	Local Repository Context Link your local Git repositories to LaReview. This gives the AI agent full access to search your codebase and list files, providing maximum context for more accurate and insightful reviews.
	Visual Diagram View Visualize the structure and flow of changes with automatically generated diagrams, helping you understand complex refactors faster. (Requires D2).
	Export & Local-First No third-party servers. Review state is stored in a local SQLite DB. Export your summary as Markdown. PR data is fetched securely via your own GitHub CLI (gh), and planning is handled by your local ACP agent.

How it works

1. Input
   • Paste a unified diff, or
   • Paste a GitHub PR reference: owner/repo#123 or a PR URL
2. Fetch (for PRs)
   • LaReview uses the GitHub CLI (gh) locally
3. Generate a review plan
   • The review plan is generated by an AI agent you run via ACP (Agent Client Protocol)
4. Review
   • Work task-by-task, add notes, track status, and optionally clean completed tasks

Supported Agents

Claude	Codex	Gemini	Grok	Kimi	Mistral	OpenCode	Qwen

Requirements

• Rust nightly (see rust-toolchain.toml)
• GitHub CLI if you want to load PRs:
  • Install: brew install gh (or your OS package manager)
  • Auth: gh auth login
• D2 (optional): For visual diagram rendering. Install: brew install d2 (or see d2lang.com)
• ACP agent configured on your machine
  • Learn more: https://agentclientprotocol.com/overview/introduction

Note

LaReview does not require a custom LaReview server. Your PR data is fetched locally via gh, and plan generation runs via your chosen ACP agent.

Installation

Option A: Download a release

1. Go to the Releases page and download the latest binary for your operating system.
2. Follow the instructions for your OS below.

macOS

Download the lareview-macos-*.zip asset and unzip it. You should see LaReview.app.

• Drag LaReview.app into /Applications.
• If macOS blocks it on first run, open System Settings → Privacy & Security and allow it.

Linux

1. Download the lareview-linux.tar.gz file and extract it.

   tar -xzvf lareview-linux.tar.gz

2. Make the binary executable and move it to your path.

   chmod +x ./lareview
   mv ./lareview /usr/local/bin/lareview

Windows

1. Download the lareview-windows.zip file and unzip it.
2. Place the lareview.exe file in a directory that is included in your system's PATH environment variable.

Option B: Run from source

cargo run

Option C: Build a local macOS .app (dev)

scripts/build_macos_app.sh

Linux dependencies

# Debian/Ubuntu
sudo apt-get update
sudo apt-get install -y libxkbcommon-dev libxkbcommon-x11-dev

Quickstart

1. Start the app:

   cargo run

2. Open GENERATE

   • Paste a unified diff, or a GitHub PR like owner/repo#123 (or a PR URL)

3. Pick an agent and click generate

4. Switch to REVIEW

   • Work through tasks, add notes, and track status

Tip

If PR loading fails, run gh auth status and then gh auth login.

Usage examples

Review a GitHub PR

Input:

• owner/repo#123
• or https://github.com/owner/repo/pull/123

Review a unified diff

Paste a unified diff directly:

diff --git a/src/lib.rs b/src/lib.rs
index 123..456 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -1,3 +1,8 @@
+pub fn example() {
+  // ...
+}

Data, paths, and privacy

LaReview stores review state locally.

• macOS DB: ~/Library/Application Support/LaReview/db.sqlite

• Windows DB: %APPDATA%\LaReview\db.sqlite or %LOCALAPPDATA%\puemos\LaReview\db.sqlite

• Linux DB: $XDG_DATA_HOME/lareview/db.sqlite or ~/.local/share/lareview/db.sqlite

• Override DB path:

  LAREVIEW_DB_PATH=/path/to/db.sqlite <RUN_COMMAND>

Executable discovery (GUI apps can start with a minimal PATH):

• LaReview hydrates PATH from your login shell when launched outside a terminal (macOS/Linux). Restart after shell PATH changes.
• You can override individual agent executables in Settings → Agent Settings.

Wipe local state:

• Delete the DB file listed above.

UI overview

GENERATE

Paste a diff (or PR), pick an agent, and generate a plan. While the agent runs, you get a timeline of what’s happening.

REVIEW

The review plan is shown as a tree: Intent → Sub-flows → Tasks. Selecting a task shows details plus a unified diff viewer for related hunks. Add notes and mark progress.

Use Clean done to remove completed tasks (and their notes) for the current review.

REPOS

Link local Git repositories to give the AI agent deep context. Once linked, the agent can use tools to search through your codebase and list files, enabling it to understand the wider impact of changes.

EXPORT

Generate a Markdown summary of your review, including stats, metadata, task details, and diagrams. You can preview and edit the Markdown before saving it to a file.

Design goals

• DX: review flows, not file lists
• Local-first state: tasks, notes, and status stored locally in SQLite
• Security / trust model: no server in the middle; processing happens through gh and your ACP agent
• Fits your setup: use ACP instead of a new agent runtime; use gh instead of forcing new integrations

Project status

• Status: ALPHA
• Known limitations:
  • Complex nested diff structures might require manual confirmation in some agents.
• Roadmap:
  • Sync comments with GitHub
  • Support for repo search
  • Support more ACP agents

Development

• Toolchain: nightly Rust with rustfmt and clippy components (edition 2024; see rust-toolchain.toml)

• Run the app: cargo run

• Reset/seed sample data:

  • cargo run --bin reset_db
  • cargo run --bin seed_db

• Checks:

  • cargo fmt -- --check
  • cargo clippy --all-targets --all-features -- -D warnings
  • cargo test

• Tests:

  • Unit tests: Located alongside modules in src/ (usually as tests.rs)
  • Integration tests: Located in the root tests/ directory
  • Run all tests: cargo test

• Supply chain:

  • cargo deny check (used in scheduled CI)

Docs

• Docs index: docs/README.md
• Architecture overview: docs/ARCHITECTURE.md
• Development guide: docs/DEVELOPMENT.md

Contributing

Please read CONTRIBUTING.md before submitting a PR.

Security

See SECURITY.md for reporting vulnerabilities.

License

Licensed under either of:

• Apache License, Version 2.0 (LICENSE-APACHE)
• MIT license (LICENSE-MIT)

At your option.