Below is the full README as normal formatted text (no outer source/code fence), so you can copy it directly as markdown.

---

LlamaDiff

Private AI code review for local Git repositories. Analyze diffs between branches using local or cloud LLMs — without requiring GitHub, PRs, or code uploads.

LlamaDiff runs directly on your machine, works with local repos, and supports offline-first workflows using Ollama.

---

Why LlamaDiff?

Most AI code review tools require:

• pushing code to GitHub/GitLab
• sending diffs to cloud APIs
• exposing proprietary code

LlamaDiff is built for:

• local Git workflows
• privacy-sensitive projects
• offline environments
• self-hosted development setups
• developers working with code under NDA

Run reviews on:

• local repositories
• uncommitted changes
• internal projects
• air-gapped systems (with local models)

---

Features

• Compare any two git branches and get AI-powered code review

• Works directly with local Git repositories (no PR required)

• Review uncommitted changes

• Supports multiple LLM providers via LiteLLM:

  • Ollama (local, default)
  • OpenAI
  • Anthropic
  • OpenRouter

• Filter review to specific files or patterns

• Export reviews as Markdown

• Interactive HTML review output

• CLI-first workflow

---

Installation

Via pipx (recommended)

# Run directly without installing
pipx run llama-diff -s feature -t main

# Or install globally
pipx install llama-diff

Via pip

pip install llama-diff

From source

git clone https://github.com/ernestp/LlamaDiff.git
cd LlamaDiff
pip install -e .

---

Configuration

Copy .env.example to .env and configure your preferred provider:

cp .env.example .env

By default, LlamaDiff uses Ollama for local model inference.

---

Usage

Basic usage

# Compare current branch to main
llama-diff --source feature-branch --target main

Use a specific model

llama-diff -s feature -t main --model ollama/codellama

Use OpenAI

llama-diff -s feature -t main --model gpt-4o

Review specific files only

llama-diff -s feature -t main --files "*.py"

Output review to file

llama-diff -s feature -t main --output review.md

Interactive HTML review

llama-diff -s feature -t main --html

Review uncommitted changes

llama-diff --uncommitted

---

Model Examples

Provider	Model Example
Ollama	ollama/llama3.2, ollama/codellama
OpenAI	gpt-5.2, gpt-5.1, gpt-4o, gpt-4o-mini
Anthropic	claude-sonnet-4.5, claude-opus-4.5
OpenRouter	openrouter/anthropic/claude-sonnet-4.5, openrouter/anthropic/claude-opus-4.5

---

Example Workflow

1. Make changes locally
2. Run LlamaDiff
3. Get structured review before pushing

git diff → llama-diff → AI review → fix → commit/push

No PR required. No cloud dependency (when using local models).

---

Who is this for?

• Developers working with code under NDA
• Teams with proprietary or confidential repositories
• Self-hosted / privacy-first environments
• Offline or restricted networks
• OSS maintainers reviewing patches locally

---

Roadmap

• Git hooks (pre-commit / pre-push)
• CI integrations
• Better diff summarization
• Model-specific review tuning
• IDE integrations

---

License

Licensed under the Apache License, Version 2.0.