LGTM

Automated code review from your terminal. Point it at a git diff, pick your models, get comments with severity ratings.

██       ██████  ████████ ███    ███
██      ██          ██    ████  ████
██      ██   ███    ██    ██ ████ ██
██      ██    ██    ██    ██  ██  ██
███████  ██████     ██    ██      ██

Install

npm

npm install -g lgtm

From source

git clone <repo-url>
cd lgtm
npm install
npm run build
npm link

Linux dependencies

LGTM stores your API key using your system's secret service. On Linux, you need secret-tool (from libsecret):

# Arch/CachyOS
sudo pacman -S libsecret

# Debian/Ubuntu
sudo apt install libsecret-tools

# Fedora
sudo dnf install libsecret

On macOS, Keychain is used automatically. On Windows, a config file fallback is used.

Usage

lgtm [options]

Run lgtm inside any git repository. On first run, you'll be prompted for an OpenRouter API key.

Flags

Examples

# Interactive — pick branches in the TUI
lgtm

# Diff two branches directly
lgtm --base main --head feature/auth

# Check system setup
lgtm doctor

How it works

1. You select a base and head ref (branches, tags, commits, or staged changes)
2. LGTM generates the diff and splits it into chunks
3. Each configured agent sends the diff to its assigned model via OpenRouter
4. Models stream back comments with file, line number, severity (1-5), and explanation
5. A dedup agent consolidates overlapping comments across models
6. Results are grouped by file with severity ratings: Info, Suggestion, Issue, Warning, Critical

Smart filtering

Binary files, lockfiles (package-lock.json, yarn.lock, etc.), and .env files are automatically excluded from the diff before it reaches the models. This reduces token usage and keeps reviews focused on meaningful changes.

Focus modes

Choose a review focus: Normal (default), Security, Performance, Best Practices, or Accessibility. Each mode tailors the AI's system prompt with domain-specific checklists. Set with r on the launch screen.

Framework detection

LGTM auto-detects your project's language and frameworks from manifest files (package.json, requirements.txt, Cargo.toml, go.mod). This context is injected into the review prompt for more relevant feedback.

Confidence scores

Each comment includes an optional confidence score (0-100%) indicating how certain the model is about the issue.

Time tracking

Review duration is tracked per agent and total. Displayed on the results screen and saved to history.

Multi-model review

Configure multiple agents, each with a different model. Run openrouter/auto alongside specific models to get varied perspectives on the same diff. The dedup pass removes duplicates so you don't read the same comment three times.

Cost tracking

Token usage and cost are tracked per session. Free models show $0.00. Cost is displayed on the results screen after review completes.

Keyboard shortcuts

Welcome screen

Launch screen

Results screen

History screen

Configuration

API key

Stored securely — macOS Keychain, Linux secret-tool, or an encrypted config file fallback. Never stored in .env or plaintext config.

Models

Any model available on OpenRouter works. The model picker shows all available models with pricing info. Default for new agents: openrouter/auto.

Requirements

• Node.js 18+
• Git
• An OpenRouter API key
• A terminal with mouse support (most modern terminals)

Development

# Run in dev mode
bun run src/index.tsx

# Build
npm run build

# Type check
npm run typecheck

# Lint
npm run lint

# Format
npm run format

# Run tests
npm run test

# Full validation (typecheck + lint + test + format)
npm run check

Project structure

src/
  components/    UI screens and widgets (OpenTUI/React)
  services/      Business logic (git, review, storage, cost)
  config/        Model configuration
  state/         React context state management
  theme/         Colors and styling (Dracula palette)
  types/         TypeScript type definitions

License

ISC