La Review

The Code Review Workbench.

demo.mp4

---

LaReview is a dev-first code review workbench for complex changes. It turns a PR or diff into a structured review plan, grouped by flows and ordered by risk, so you can review with depth and understanding.

Unlike auto-review bots that post comment spam, LaReview is a local-first tool that works with your existing AI coding agent to help you ship high-signal feedback.

Example review

The feature screenshots follow an illustrative payment-webhook review: six changed files, four tasks, and two blockers. See the example scenario.

Key Features

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

Feature	Workflow
AI-Powered Planning	Load a GitHub or GitLab pull request, compare local Git refs, or paste a diff. LaReview acts as a staff engineer, looking for flows (e.g. auth, api) and risks to build a structured review plan.
Review Tasks	Follow the review plan by flow, track progress, and focus on the highest-risk concerns first.
Task Analysis	Work through a focused verification plan for each logical concern. 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	Review one logical concern at a time. The workbench isolates relevant code hunks for each task, eliminating noise.
High-Signal Framework	The AI proactively identifies bugs and authenticates them against your rules. No comment spam—just focused feedback threads anchored to specific lines.
Local Context	Link local Git repos to give the agent full access to search your codebase. Works with your existing AI coding agent using context from your local checkout.
Visual Diagrams	Visualize the flow. Automatically generate diagrams to see architectural changes before you read a single line of code. Rendered with Mermaid.
Team Rules	Define rules like "DB queries must have timeouts" or "API changes need a migration note" to enforce standards automatically.
Learning Patterns	The AI learns from rejected feedback. Mark suggestions as "ignored" during reviews, then analyze rejections to discover patterns that calibrate future reviews—fewer nitpicks, more signal.
Git Host Sync	Submit your review feedback directly to GitHub or GitLab PRs with automatic summary generation.
Export & Share	Export your review summary to Markdown with diagrams and code insights. Copy the selected tasks and feedback to your clipboard.
CLI Support	Launch reviews from the terminal: lareview, lareview pr owner/repo#123, git diff | lareview, or lareview --agent claude.

Inside the workbench

A focused review plan

A finding with its code and evidence

The task’s relevant code changes

How it works

1. Input
   • Run lareview from terminal (opens GUI with current repo linked), or
   • Paste a unified diff, or
   • Paste a code review link from GitHub or GitLab, or a PR URL
2. Fetch (for PRs)
   • LaReview fetches the change locally and prepares the review workspace. No intermediate servers.
3. Generate a review plan
   • Starting creates the review immediately and opens its live Activity page while the selected AI agent works in the background.
   • You can return to Generate and start another review; two generations can run concurrently and additional reviews wait in a visible queue.
   • When generation finishes, the same review page automatically becomes the full review workbench.
4. Review
   • Review task by task with your AI coding agent.
   • Ship focused feedback that authors can act on, without comment spam.

Terminal Workflow

LaReview includes a CLI that integrates naturally into your developer workflow:

# Open GUI with current repository linked
lareview

# Review changes between branches
lareview main feature

# Pipe a diff directly to the GUI
git diff HEAD | lareview

# Review a GitHub/GitLab PR
lareview pr owner/repo#123

# Specify an agent
lareview --agent claude

The CLI launches the GUI in the background and passes your diff/repo info seamlessly.

On-prem and other Git hosts

Clone your repository using your usual Git credentials, then link the local folder in Repositories. In Generate Review, select that repository, choose Local Git, enter a base and head branch, tag, or commit, and click Compare changes. Inspect the diff, then click Start Review with your chosen agent.

Comparison uses the two commit snapshots (git diff BASE HEAD) already present locally. Fetch remote branches with Git first. LaReview does not change your checkout or staged changes. The review records both resolved commit IDs and can use the head commit through the existing snapshot permission flow. This works with on-prem Git, Bitbucket, Gitea, Azure DevOps, GitHub Enterprise, and repositories without a remote. Host-specific PR metadata, inbox, and posting comments remain available through the GitHub/GitLab integrations.

The terminal workflow also works from any local clone: lareview main feature or git diff main feature | lareview.

Supported Agents

LaReview launches compatible harnesses through commands already available on your PATH, or an executable override configured in Settings. It does not provide an in-app harness installer.

Agent	Launch command
Claude	npx -y @agentclientprotocol/[email protected]
Codex	npx -y @agentclientprotocol/[email protected]
GitHub Copilot	copilot --acp --stdio
Gemini	gemini --acp
Grok	grok --no-auto-update agent stdio
Kimi	kimi acp
Mistral	vibe-acp
OpenCode	opencode acp
Qwen	qwen --acp --experimental-skills

General ACP support is not sufficient: the harness must forward LaReview's client-provided MCP server and call its tools. pi-acp is intentionally unsupported because it does not provide that required MCP behavior.

Requirements

• Rust 1.99.1 (installed automatically by rustup from rust-toolchain.toml; MSRV 1.91)
• Node.js 24.22.2 and pnpm 11.28.9 for frontend development
• GitHub CLI (optional):
  • Install: brew install gh (or your OS package manager)
  • Auth: gh auth login
• GitLab CLI (optional):
  • Install: brew install glab (or your OS package manager)
  • Auth: glab auth login
• AI 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 or glab, and plan generation runs via your chosen AI agent.

GitHub Copilot setup

Install the GitHub Copilot CLI and sign in by running copilot and using /login. Restart LaReview, then select GitHub Copilot in the Agent picker. If the executable is outside your PATH, set its path in Settings. LaReview uses Copilot's native ACP stdio server; no additional adapter or gh copilot extension is needed. Models and session options are discovered from the installed CLI, and require an authenticated Copilot account or a configured bring-your-own-key provider.

For each review, LaReview supplies its local task server through Copilot's --additional-mcp-config launch option, because Copilot's ACP session API accepts only HTTP/SSE MCP entries. This uses the same local stdio task server as the other agents.

Installation

Option A: Homebrew (Recommended)

Install LaReview via Homebrew with a single command:

macOS

# First, tap the repository
brew tap puemos/tap
# Then install
brew install --cask lareview

Linux

brew install puemos/tap/lareview

Option B: 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

1. Download LaReview_<version>_aarch64.dmg for Apple silicon or LaReview_<version>_x64.dmg for Intel.
2. Open the disk image and drag LaReview.app into /Applications.
3. If macOS blocks it on first run, open System Settings → Privacy & Security and allow it.
4. To use from terminal, add to PATH (or use the CLI Installation button in Settings):

   echo 'export PATH="$PATH:/Applications/LaReview.app/Contents/MacOS"' >> ~/.zshrc

Linux

1. Install required dependencies:

   # Debian/Ubuntu
   sudo apt-get update
   sudo apt-get install -y \
     libwebkit2gtk-4.1-0 \
     libappindicator3-1 \
     librsvg2-2 \
     libxdo3
   
   # For other distros, install equivalent WebKit2GTK 4.1 packages

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

   tar -xzvf lareview-linux.tar.gz

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

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

WSL (Windows Subsystem for Linux)

LaReview runs natively inside WSL2 as a Linux app. Follow the Linux installation steps above (including dependency installation).

Optional: For better Windows integration (opening URLs in your Windows browser), install the wslu package:

# Ubuntu/Debian
sudo apt install wslu

# Other distros: see https://wslutiliti.es/wslu/install.html

If wslu is not installed, LaReview will fall back to using cmd.exe for opening URLs.

Option C: One-click CLI installation

Launch LaReview and go to Settings → CLI Installation for a one-click setup to add lareview to your PATH.

Option D: Run from source

cargo run

Option E: 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:

   lareview  # From terminal with repo linked
   # OR
   cargo run  # From source

2. Open GENERATE

   • Paste a unified diff, or a GitHub/GitLab PR like owner/repo#123 (or a PR URL)
   • Or use the terminal: lareview pr owner/repo#123

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 or glab auth status and then login.

Usage examples

Review a GitHub/GitLab PR

Input:

• owner/repo#123
• https://github.com/owner/repo/pull/123
• https://gitlab.com/owner/repo/-/merge_requests/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

• 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.

Development

• Toolchain: Rust 1.99.1 with rustfmt and clippy components (edition 2024; MSRV 1.91)

• Run the app: cargo run

• Reset/seed sample data:

  If you need more advanced control, use the CLI.

  CLI Reference

• Checks:

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

• Logging:

  • LaReview uses the log crate with env_logger.
  • Set RUST_LOG to control log levels:
    • RUST_LOG=debug cargo run - Full debug output
    • RUST_LOG=acp=debug cargo run - ACP agent debug output only
    • RUST_LOG=error cargo run - Errors only
  • Logs appear in the terminal when running from source.

• 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.