https://github.com/massimilianoviola/llm-code-review
Pre-commit hook that reviews staged changes using a local LLM via Ollama
https://github.com/massimilianoviola/llm-code-review
code-review developer-tools ollama pre-commit-hook qwen
Last synced: 5 days ago
JSON representation
Pre-commit hook that reviews staged changes using a local LLM via Ollama
- Host: GitHub
- URL: https://github.com/massimilianoviola/llm-code-review
- Owner: massimilianoviola
- License: mit
- Created: 2026-03-08T15:38:46.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-03-08T22:12:11.000Z (3 months ago)
- Last Synced: 2026-03-09T02:22:16.028Z (3 months ago)
- Topics: code-review, developer-tools, ollama, pre-commit-hook, qwen
- Language: Python
- Homepage:
- Size: 55.7 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# 🔍 LLM Code Review
A git pre-commit hook that reviews your staged changes using a local LLM via [Ollama](https://ollama.com). No API keys. No cloud. Your code never leaves your machine.
### How it works
1. You run `git commit`, the pre-commit hook kicks in
2. Your staged diff is sent to a local Ollama model
3. The model reviews your code for bugs, security issues, and style
4. You get a verdict: ✅ PASS / ⚠️ WARN / ❌ FAIL
5. You choose whether to proceed or abort as normal
You can also run `llm-code-review run` manually at any time.
## 🚀 Quick start
### 1. Install Ollama
Download from [ollama.com/download](https://ollama.com/download) or `curl -fsSL https://ollama.com/install.sh | sh` (macOS, Linux).
```bash
ollama serve # start the server
ollama pull qwen3.5:4b # grab a model
```
### 2. Install llm-code-review
📌 Requires Python 3.11+
```bash
git clone https://github.com/massimilianoviola/llm-code-review.git
cd llm-code-review
pip install -e .
```
### 3. Hook it up
```bash
cd your-repo
llm-code-review install
```
### 4. Just commit
```bash
git add . && git commit -m "your message"
```
The review streams live to your terminal. You choose whether to proceed.
```
=== LLM Code Review ===
VERDICT: PASS
ISSUES: None detected
SUMMARY: README.md content updated successfully; no code or security issues found.
Verdict: PASS
Proceed with commit? [y/N]
```
### VS Code sidebar
The hook also runs (in non-interactive mode) when committing from the VS Code Source Control sidebar (takes its time). The review output is saved to `.llm-code-review.log`. On FAIL, the commit is blocked. Run `git commit` from the terminal to review interactively and accept anyway.
## 🛠️ Commands
| Command | Description |
|---|---|
| `llm-code-review run` | 📝 Review staged changes |
| `llm-code-review check` | 🏥 Verify Ollama server & model |
| `llm-code-review benchmark` | ⏱️ Compare model speeds — default: `qwen3.5:0.8b,2b,4b` |
| `llm-code-review install` | 🪝 Install the pre-commit hook |
| `llm-code-review uninstall` | 🗑️ Remove the pre-commit hook |
### Options for `run`
| Flag | Default | Description |
|---|---|---|
| `--model` | `qwen3.5:4b` | Ollama model to use |
| `--url` | `http://localhost:11434` | Ollama server URL |
| `--strict` | off | Also fail on ⚠️ WARN |
| `--no-interactive` | off | Skip prompt, auto-decide |
### Options for `benchmark`
| Flag | Default | Description |
|---|---|---|
| `--models` | `qwen3.5:0.8b,qwen3.5:2b,qwen3.5:4b` | Comma-separated list of models |
| `--url` | `http://localhost:11434` | Ollama server URL |
### Options for `check`
| Flag | Default | Description |
|---|---|---|
| `--model` | `qwen3.5:4b` | Model to check |
| `--url` | `http://localhost:11434` | Ollama server URL |
### Options for `install` / `uninstall`
| Flag | Default | Description |
|---|---|---|
| `--repo` | cwd | Path to git repository (install hook remotely without cd-ing) |
## ⚙️ Configuration
Optionally drop a `.llm-code-review.toml` in your repo root to override defaults:
```toml
model = "qwen3.5:4b"
ollama_url = "http://localhost:11434"
strict = false
timeout = 120
```
**Precedence** (each layer overrides the previous): defaults → `.llm-code-review.toml` → CLI flags
## 🧑💻 Development
```bash
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
```
```bash
pytest
ruff check --fix
ruff format
```