Local AI · No Cloud · No Subscriptions

Git Operations.Intelligently Managed.

GitPilot keeps your work, personal, and open-source Git identities perfectly organized — and writes your commit messages using on-device AI.

0.5B–7B
Model Sizes
100%
Offline AI
Git Identities
macOS + Win
Platforms
GitPilot v0.1.0 — Dashboard
GitPilot main dashboard showing repos grouped by Git identity

Everything your Git workflow needs.

One desktop app to manage identities, generate commits, resolve conflicts, and stay in sync — without ever leaving your flow.

👤

Smart Identity Management

Define work, personal, and OSS identities. GitPilot auto-applies the right name and email per repo based on path rules — no more wrong commits, no more embarrassing email leaks.

🤖

On-Device AI Commits

Powered by Qwen 2.5 Coder running locally via llama.cpp. Analyzes your staged diff and generates meaningful, conventional commit messages — zero cloud, zero API keys, zero latency.

🌿

Visual Git Operations

Stage and unstage files, commit, push, pull, manage branches and stashes — all with animated progress modals and a clean dark UI that respects your eyes at 2am.

🔀

Merge & Rebase Wizards

Step-by-step guided flows for merges and rebases. Preview conflicts before they explode, and get AI-powered resolution suggestions when they do.

🔍

Repo Scanner

Recursively scans your filesystem for Git repos and groups them by active identity. Identity mismatches are flagged immediately — one click to fix.

👁

Live File Watching

Filesystem watcher with debounced events auto-detects new repos and refreshes status in real time. Your dashboard is always current, always accurate.

Up and running in minutes.

01

Scan Your Machine

Point GitPilot at a root directory. It recursively discovers every Git repo and surfaces them in a unified dashboard, grouped by detected identity.

02

Configure Identities

Define path-based rules per identity — work, personal, OSS. GitPilot silently applies the correct user.name and email for each repo, automatically.

03

Commit with AI

Stage your changes, click Generate. The on-device model reads your diff and produces a meaningful commit message. Review, edit if needed, ship.

Your diffs stay on
your machine. Always.

GitPilot bundles llama-server (llama.cpp) as a Tauri sidecar. No OpenAI. No Anthropic. No subscriptions. Nothing leaves your device.

🔒

Zero Cloud Exposure

Your code, diffs, and commit history never leave your machine. Private repos stay private, always.

Streaming Token Output

SSE-based streaming delivers commit message tokens in real-time as the model generates them.

🎛

Choose Your Model

From 0.5B (fast, low RAM) to 7B (richer output). Download once from the AI panel, switch anytime.

📦

No Ollama Required

llama-server ships bundled inside the app. No Homebrew, no Python, no external runtimes.

// Staged diff → AI commit message

model: "qwen2.5-coder-1.5b"
transport: "llama-server (bundled)"
streaming: true
api_key: null
cloud: false
// Generated output

feat(auth): add token refresh logic

Implements silent refresh using stored
refresh token. Handles 401 responses
with automatic retry queue.
// Identity auto-applied

path: "~/work/adcb/**"
user.name: "Krishnaraj Nair"
user.email: "k@adcb.work"
GitPilot AI model selector panel showing Qwen 2.5 Coder models
AI Model Panel — select from 0.5B to 7B Qwen 2.5 Coder models, downloaded on demand

The better way to manage Git.

FeatureManual .gitconfigOther GUI ToolsGitPilot ✦
Multiple Git identitiesManual includes
Auto identity per repo
AI commit messagesCloud only✓ On-device
No API key required
Merge & rebase wizardsBasic only✓ AI-assisted
Identity mismatch warnings
Works fully offlinePartial

Get GitPilot for free.

One-time download. No subscription. No telemetry. No account required.

🍎

macOS

Apple Silicon · arm64 · ~18MB
Download .dmg
🪟

Windows

x64 · Windows 10/11 · ~16MB
Download .exe

AI models downloaded separately on first run (0.5B–7B GGUF) · Storage: 400MB–4.4GB

Common questions.

Never. The AI runs entirely on-device via a bundled llama-server binary. Your diffs, code, and commit messages are processed locally and never transmitted to any server. GitPilot has no analytics, no telemetry, and no account system.
No. GitPilot bundles llama-server directly as a Tauri sidecar process. No Homebrew, no Ollama, no Python — just the app. You pick and download a GGUF model on the first-run wizard and you're done.
GitPilot uses path-based rules you configure. For example, all repos under ~/work/ can be auto-assigned your work identity. If no rule matches, the repo is flagged with a mismatch warning and you can assign one with a single click.
The 1.5B Qwen 2.5 Coder hits the sweet spot for most machines — fast generation and solid commit quality. The 3B model is recommended for the best balance of quality and speed. The 7B produces the richest messages but needs 16GB+ RAM.
Yes. GitPilot uses libgit2 under the hood, which is remote-agnostic. Push and pull work with any Git remote — GitHub, GitLab, Bitbucket, self-hosted, or plain SSH. Authentication uses your existing SSH keys or credential helpers.
GitPilot is a portfolio/showcase project built with Tauri 2, Rust, and React. Source code is available on GitHub. Feel free to explore the architecture or reach out if you want to discuss the implementation.