GitSage AI

AI-powered Git commit assistant — one command, production-quality commits.

GitSage analyses your staged changes and returns a perfectly structured commit message plus a full intelligence report, powered by a hosted backend that makes a single AI round-trip so you never wait for a local model to spin up.

Features

Quick start

1 — Install

pip install gitsage

2 — Get an API key

Visit https://gitsage-ai.vercel.app/docs to generate a free key.

3 — Authenticate

gitsage auth --token gs_YOUR_KEY_HERE

Your key is saved to ~/.gitsage_auth with restricted file permissions (600 on Unix).

4 — Commit

git add .
gitsage commit
# or the shorthand
gitsage -c

Commands

gitsage commit

Analyses staged changes and opens an interactive prompt:

y     — accept and commit
edit  — revise the message before committing
n     — abort

gitsage auth

gitsage auth --token <KEY>   # save / replace API key
gitsage auth                 # show current key status (masked)
gitsage auth --status        # same as above, explicit
gitsage auth --logout        # remove stored key

gitsage config

gitsage config                        # show current preferences
gitsage config --style emoji          # change commit style
gitsage config --no-telemetry         # disable anonymous telemetry
gitsage config --reset                # restore all defaults

gitsage --version

Prints the installed version and exits.

Configuration

Preferences are stored in ~/.git-sage.json. The API key is kept separately in ~/.gitsage_auth and is never written to the preferences file.

Architecture

cli/main.py          — Typer app, auth command, commit workflow, rich UI
│
├── config/loader.py — Preferences (~/.git-sage.json) + secure key (~/.gitsage_auth)
├── git/diff.py      — Staged diff retrieval, sensitive-file filtering, async wrappers
│
├── engine/
│   ├── core.py      — Orchestrator: cache → fast-path (API) or legacy (local)
│   ├── cache.py     — SHA-256 keyed result cache in ~/.gitsage_cache
│   ├── analyzer.py  — Diff parser → DiffSummary (files, intent, cleaned content)
│   ├── models.py    — CommitResult, DiffSummary dataclasses
│   ├── orchestrator.py — Single-prompt generator for local providers
│   ├── explainer.py — Explanation generator + confidence heuristic (local path)
│   └── formatter.py — Output formatting utilities
│
└── providers/
    ├── base.py      — AIProvider ABC (generate / generate_async)
    ├── gitsage.py   — GitSageAPIProvider — async httpx, AnalysisResult, _clean_commit_message
    ├── gemini.py    — Google Gemini (optional, pip install gitsage[gemini])
    └── local.py     — Ollama local provider (optional, pip install gitsage[local])

Request flow

gitsage commit
    │
    ├─ git diff --cached          (async subprocess)
    ├─ sensitive-file filter
    ├─ truncate to 3 000 tokens
    ├─ SHA-256 cache lookup       → hit: return immediately
    │
    └─ POST /v1/intelligence/analyze
           X-API-Key: gs_...
           { diff, context, style }
           ↓
       { commit_message, explanation, confidence, provider, model }
           ↓
       _clean_commit_message()    (strip markdown from message)
       cache.save()
       display_result()           (rich panels + confidence bar)
       prompt: y / edit / n

Development

git clone https://github.com/iamAgbaCoder/gitsage
cd gitsage
python -m venv .dev && source .dev/bin/activate   # or .dev\Scripts\activate on Windows
pip install -e ".[dev]"

Run tests

pytest tests/ -v

Lint & format

ruff check .
black .

Build

python -m build

Publishing

Trusted Publishing (OIDC) is used — no PYPI_TOKEN secret needed. Set up the PyPI and TestPyPI environments in your GitHub repo settings.

License

Apache 2.0 — see LICENSE.