cr-reviewer

Local AI code reviewer powered by Ollama. Zero server costs. Code never leaves your machine.

Prerequisites

| Requirement | Minimum | Install | |---|---|---| | Node.js | 18.0.0 | nodejs.org | | Ollama | latest | ollama.com |

Installation

Global install (recommended)

npm install -g cr-reviewer

One-off with npx

npx cr-reviewer review my.diff

From source

git clone https://github.com/AESiR-0/webgpu-llm.git
cd webgpu-llm/packages/cli-reviewer
npm install
npm link

First-time setup

# 1. Start Ollama
ollama serve

# 2. Pull a model (one-time download ~4 GB)
ollama pull qwen2.5-coder:7b

# 3. Run your first review
git diff main...feature | cr review - --output review.md

Usage

# Review a diff file
cr review my.diff

# Read from stdin (pipe from git)
git diff --cached | cr review -

# Write to file
cr review my.diff --format md --output review.md

# Check Ollama status
cr status

# List available models
cr models

All flags

| Flag | Default | Description | |---|---|---| | --model, -m | qwen2.5-coder:7b | Ollama model name | | --host | http://localhost:11434 | Ollama base URL | | --timeout | 120000 | Per-request timeout (ms) | | --mode | fast | fast (unified) or deep (3-agent sequential) | | --agents | bug,security,performance | Comma-separated agents to enable | | --focus | "" | Context injected into every prompt | | --profile | none | security-audit, critical-only, performance, show-all | | --format, -f | md | md, json, csv, sarif, pr-desc | | --output, -o | stdout | Write to file instead of stdout | | --min-severity | info | Filter: info, warning, critical | | --stream | false | Print LLM tokens live | | --quiet, -q | false | Suppress progress (CI mode) | | --verbose, -v | false | Show per-chunk timing | | --no-tests | false | Skip test suggestion generation | | --no-commit-msg | false | Skip commit message generation |

Examples

# Security audit, JSON output
cr review feature.diff --profile security-audit --format json -o audit.json

# CI pipeline — SARIF for GitHub Code Scanning
git diff main...HEAD | cr review - --format sarif --output results.sarif --quiet

# Deep mode with focus context
cr review pr.diff --mode deep --focus "Go — check goroutine leaks"

# Stream tokens while reviewing
cr review my.diff --stream --verbose

Suggested models

| Model | Description | |---|---| | qwen2.5-coder:7b | Best for code review (default) | | llama3.2:3b | Fastest, smallest (~2 GB) | | llama3.1:8b | Best for large complex diffs | | mistral:7b | Strong general reasoning | | deepseek-coder:6.7b | Alternative code specialist |