Skip to main content

Local AI-powered Git auto-committer daemon

Project description

GitPilot logo

GitPilot

Every save, a meaningful commit.

AI-powered auto-committer · Zero-config · Private by default · Offline-capable

PyPI version Python versions CI Coverage License: MIT

Quick Start · Features · How It Works · CLI · Configuration · Roadmap


GitPilot watches your Git repositories and turns every save into a meaningful, Conventional Commit — automatically. It runs as a quiet background daemon, works with cloud or fully local AI, and never sends your code off-machine unless you tell it to.

- git commit -m "fix stuff"          # manual, vague, easy to forget
+ feat(login): add password reset flow   # written for you, every time
GitPilot live monitor demo

Why GitPilot?

Without GitPilot With GitPilot
You remember to commit… sometimes Every save becomes a commit
git commit -m "updates" Branch-aware, scoped messages like fix(api): handle null response
Ten trivial commits for one change Smart grouping collapses related files into one commit
Context lost between sessions A complete, searchable, meaningful history
Another tool to configure One command install, one wizard, done

Quick Start

1 — Install

pip install gitpilot

2 — Install the daemon and run the setup wizard

gitpilot install-service && gitpilot setup

3 — Done. GitPilot is now watching your chosen projects. Type gitpilot anytime to open the live dashboard.

Windows: run the same commands in PowerShell as Administrator.

Alternative: install script
curl -fsSL https://raw.githubusercontent.com/alexwakrod/gitpilot/main/install.sh | bash
Alternative: Docker
docker-compose up -d
docker exec -it gitpilot-daemon gitpilot setup

How It Works

flowchart LR
    A["File saved"] --> B["Watcher"]
    B --> C["Debounce"]
    C --> D["Smart grouping"]
    D --> E["AI provider"]
    E --> F["Conventional commit"]
    F --> G["Notification"]
  1. Watchwatchdog observes every registered project recursively, ignoring .git internals, swap files, and duplicate events.
  2. Debounce — changes settle for a configurable quiet period (default 3s).
  3. Group — related files (same directory, similar names) collapse into a single logical change.
  4. Write — the active AI provider drafts a branch-aware Conventional Commit message.
  5. Commit — git executes with automatic retry; Discord is notified if enabled.

Features

Automatic watching

Register any project with gitpilot add /path/to/project — or pick directories visually in the TUI. GitPilot watches recursively and filters out the noise so you never think about it again.

AI commit messages, your way

Every message follows the Conventional Commits spec and respects the active branch:

feat(login): add password reset flow
fix(api): handle null response from /users
chore: update dependencies
Provider Runs where API key Offline Best for
Ollama Your machine Not required Privacy-first and offline work
xAI Grok Cloud Required Default — fast and accurate
OpenAI Cloud Required GPT-class message quality
Anthropic Cloud Required Claude-class message quality

If the AI is unreachable, GitPilot falls back to sensible messages derived from file names — your history is never blocked.

Smart grouping & debounce

A burst of saves becomes one commit, not ten. Related changes are grouped automatically; the quiet period is fully configurable.

Notifications (optional)

Configure a Discord webhook per project and every commit posts a rich embed to your channel.

Interactive TUI

Launch with gitpilot:

  • Add / Select Projects — navigate your filesystem with arrow keys
  • Live Monitor — real-time commit events streamed over SSE
  • Settings — change provider, keys, and webhooks without touching a config file

Secure by Default

Measure Detail
🔒 Local-only IPC Daemon binds to 127.0.0.1 with a randomly generated bearer token
🔑 Credential storage Config written with 0600 permissions on Unix
🚫 No open ports Nothing is ever exposed to the network
🏠 Offline mode The Ollama provider runs 100% on-device — code never leaves your machine

CLI Reference

Everyday

Command Description
gitpilot Launch the interactive TUI dashboard
gitpilot add <path> Register a project for watching
gitpilot status List watched projects
gitpilot log <project_id> Show recent commit history
gitpilot remove <project_id> Stop watching a project

Configuration

Command Description
gitpilot setup Re-run the first-time wizard
gitpilot config-list Display all settings
gitpilot config-set <key> <value> Change a setting

Service

Command Description
gitpilot install-service Install the daemon as an OS service
gitpilot daemon-status Check whether the daemon is running

Planned: gitpilot force-commit <project_id> — trigger a manual commit.

The daemon (gitpilotd) starts automatically via install-service. To run it manually: gitpilotd.


Architecture

gitpilot/
├── cli/              # Click + Rich TUI
├── daemon/           # FastAPI + SSE, lifecycle
├── core/
│   ├── watcher.py        # Watchdog, debounce, smart grouping
│   ├── committer.py      # AI provider adapter (strategy pattern)
│   ├── executor.py       # Git operations + retry
│   └── notifications.py  # Discord webhook
├── domain/           # Pydantic models, policies, settings
├── infrastructure/   # DuckDB, HTTP client, repositories
└── tests/            # Unit, integration, E2E
Concern Choice Why
Database Embedded DuckDB Zero extra processes, fast queries
IPC FastAPI over loopback + bearer token Local-only, secure by default
File watching watchdog Battle-tested, cross-platform
AI integration Strategy-pattern adapter Swap providers without touching core logic

Configuration

Settings live in ~/.gitpilot/config.json and are editable via the TUI or gitpilot config-set.

Key Default Description
ai_provider grok AI backend: grok, openai, anthropic, or ollama
ai_model grok-2 Model name passed to the provider
ai_temperature 0.5 Sampling temperature for message generation
debounce_interval 3 Quiet period in seconds before changes are committed
smart_grouping true Group related file changes into one commit
branch_aware_messages true Include branch scope, e.g. feat(login): …
max_commit_retries 3 Retry attempts for failed git operations
discord_webhook_enabled false Send commit notifications to Discord
theme dark TUI color theme
Full default config.json
{
  "ai_provider": "grok",
  "ai_model": "grok-2",
  "ai_temperature": 0.5,
  "debounce_interval": 3,
  "smart_grouping": true,
  "branch_aware_messages": true,
  "max_commit_retries": 3,
  "discord_webhook_enabled": false,
  "theme": "dark"
}

Development

Setup

git clone https://github.com/alexwakrod/gitpilot
cd gitpilot
python3.11 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

Test — coverage target ≥ 80%

pytest --cov=gitpilot --cov-report=term-missing

Package

python -m build

Roadmap

  • Full-screen Textual dashboard with diff viewer
  • Plugin architecture for notifications (Slack, Telegram, email)
  • Remote daemon mode (TLS-secured, team usage)

FAQ

Does my code ever leave my machine?

Only if you choose a cloud AI provider (Grok, OpenAI, Anthropic) — and even then, only the diff metadata needed to write the message is sent. Select the Ollama provider for a fully offline, 100% on-device experience.

What happens if the AI is unreachable?

GitPilot generates a sensible fallback message from the changed file names. Your commit history is never blocked by a network issue.

Which platforms are supported?

Linux, macOS, and Windows. The daemon installs as a native user service on each platform with a single command: gitpilot install-service.

Will it create noisy, one-line commits?

No. Debouncing waits for a quiet period, and smart grouping merges related file changes into a single, well-described commit.


License

MIT © Alex Wakrod

Built with ❤️ by a developer, for developers.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

gitpilot_ai-0.1.0.tar.gz (47.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

gitpilot_ai-0.1.0-py3-none-any.whl (54.6 kB view details)

Uploaded Python 3

File details

Details for the file gitpilot_ai-0.1.0.tar.gz.

File metadata

  • Download URL: gitpilot_ai-0.1.0.tar.gz
  • Upload date:
  • Size: 47.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/7.0.0 CPython/3.12.3

File hashes

Hashes for gitpilot_ai-0.1.0.tar.gz
Algorithm Hash digest
SHA256 c94ba3939a6c2757d4580c14b40689f3f578c2a9f5bea7d07223b13d8a82f557
MD5 c0f0ca61a8f8be5888d84d3e5931f5ef
BLAKE2b-256 cb7c520de1095dd24682edbb161513807d4ef9b0f2397cf60b92622f9dcddf5d

See more details on using hashes here.

File details

Details for the file gitpilot_ai-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: gitpilot_ai-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 54.6 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/7.0.0 CPython/3.12.3

File hashes

Hashes for gitpilot_ai-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 56aeb0ba1ef0d9ae80b5a7acf60410fbcbb532bc1df05c5bb4eede86283ea4a4
MD5 264ff10e44d7aba230b6d0e0e7dfbf5a
BLAKE2b-256 643255b0515477056d9fb5bf46f8baacba38dfc88137fdd622415a49fe74c8f7

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page