Python AI commit tools

A single self-healing script (aic.py) that turns git diff --cached into a clean, conventional commit message. One CLI, four providers — Anthropic Claude, Google Gemini, OpenAI, and Ollama (local) — selected by auto-detection (Ollama → Anthropic → Gemini → OpenAI) or forced with --provider. Flags work the same for every provider.

For a side-by-side comparison of providers, flags, costs, and setup, see the bundled comparison guide:

AI commit comparison guide (full text)

AI Commit Tools Comparison

A single script — aic.py — turns git diff --cached into a clean, conventional commit message using one of four AI providers. It auto-detects the best available provider, or you can force one with --provider. All flags work identically across providers.

Providers

Provider	Alias	API key	Default model	Notes
Ollama (local)	git aicl	none	first installed of PREFERRED_MODELS	Free, private, no network; auto-falls back to smaller models on OOM/timeout
Anthropic Claude	git aica	ANTHROPIC_API_KEY	claude-haiku-4-5	Fast, high quality
Google Gemini	git aicg	GOOGLE_API_KEY / GEMINI_API_KEY	gemini-2.0-flash	Generous free tier
OpenAI	git aico	OPENAI_API_KEY	low-cost auto-pick (-m to choose)	State-of-the-art quality

git aic (no provider) auto-detects in priority order: Ollama → Anthropic → Gemini → OpenAI (Ollama first because it's free and local). Use git aic --cloud (or -c) to skip Ollama and use the best available cloud provider. If Ollama runs but every local model is exhausted, aic offers to switch to a cloud provider.

Quick start

```powershell

Set whichever cloud key(s) you want (PowerShell; use setx to persist)

$env:ANTHROPIC_API_KEY = "..." $env:GOOGLE_API_KEY = "..." $env:OPENAI_API_KEY = "..."

Ollama needs no key — just run ollama serve and ollama pull <model>

```

bash git aic # auto-detect provider, commit staged changes git aica # force Anthropic git aicg # force Gemini git aico # force OpenAI git aicl # force Ollama (local) git aic --cloud # auto-detect, but skip Ollama git aic --provider openai -m # equivalent to: force OpenAI, pick model interactively

Command flags (identical for every provider)

Flag	Description
FILES...	Stage the named files first
-a, --add-all	Stage all changes before committing
-c, --cloud	Skip Ollama; use the best available cloud provider
-d, --dry-run	Show generated message without committing
-f, --fill-placeholders	Fill the cinfo placeholders for extra context
-i, --interactive-add	Interactive file staging with fzf (git adi)
-m, --model	Select a model interactively
-p, --push	Push after a successful commit
--provider	Force a provider: ollama / anthropic / gemini / openai
-r, --review	Review and approve message before commit [y/n/e/r]
-v, --debug	Verbose logging
-w, --watermark	Add a "Generated by ..." signature

Flags combine, e.g. git aic -arp = stage all → review → commit → push.

Which provider?

• Ollama — free, private, offline; quality depends on your installed models and RAM.
• Anthropic / OpenAI — best quality when you have credits.
• Gemini — generous free tier, fast, good for high-volume use.

Installation

1. Ensure Python is installed.
2. Run gitinit.cmd to set up the aic / aica / aicg / aico / aicl aliases.
3. Dependencies auto-install on first run via dependency_manager.py — and only for the provider you actually use (an OpenAI-only user never installs the Anthropic SDK).

Dependencies (installed lazily, per provider)

• Ollama → requests
• Anthropic → anthropic, httpx
• Gemini → requests
• OpenAI → openai, httpx

Troubleshooting

• Rate limit (429): wait, check the provider's usage dashboard, or add credits / upgrade.
• API key not set: aic prints the exact env var to set and where to get the key.
• Ollama not running / no models: start it with ollama serve and ollama pull llama3.2:3b.
• Out of RAM with Ollama: lower OLLAMA_NUM_CTX, or let the automatic small-model fallback handle it; raise OLLAMA_TIMEOUT for slow cold starts.
• List models: git aic --provider <name> -m.

License

Part of the power-user-scripts repository.

---

The tool

aic.py

git aic auto-detects; git aica / git aicg / git aico / git aicl are thin aliases that pin Anthropic / Gemini / OpenAI / Ollama via --provider. The provider classes (AnthropicProvider, GeminiProvider, OpenAIProvider, OllamaProvider) all live inside this one module.

aic

AI-powered Git commit message generator — one tool, many providers.

This single module replaces the old family of per-provider scripts (aica.py / aicg.py / aico.py / aicl.py). It generates a conventional-commit message from your staged changes using whichever AI provider you pick.

Providers (the "implementations" behind one shared framework): - Ollama (local) — no API key, runs at localhost:11434 (--provider ollama) - Anthropic Claude — ANTHROPIC_API_KEY (--provider anthropic) - Google Gemini — GOOGLE_API_KEY / GEMINI_API_KEY (--provider gemini) - OpenAI — OPENAI_API_KEY (--provider openai)

Usage

git aic # auto-detect: Ollama (local) > Anthropic > Gemini > OpenAI git aic --cloud # skip Ollama, use the best available cloud provider git aic --provider X # force a specific provider (X = ollama|anthropic|gemini|openai)

All commit flags work the same across every provider

FILES... stage these files first -a / --add-all stage all changes (git add .) -i / --interactive-add stage interactively via 'git adi' (fzf) -d / --dry-run print the message, do not commit -r / --review review before committing [y/n/e/r] -p / --push push after a successful commit -m / --model pick the model interactively -f / --fill-placeholders fill the 'cinfo' placeholders for extra context -w / --watermark append a "Generated by ..." signature -v / --debug verbose logging

Design: a Provider abstract base class declares the contract; each concrete provider implements only its own API calls. The shared git framework (staging, prompt building, the review/commit loop) lives here exactly once. Third-party SDKs are imported lazily — only the chosen provider's packages are installed/loaded, so an OpenAI-only user never pulls in the Anthropic SDK.

---

Companion utilities

chat.py

chat

aipr.py

aipr

dependency_manager.py

dependency_manager

🛡️ Dependency Manager (Self-Healing Environment Engine)

A drop-in module to ensure Python scripts automatically install their own dependencies before execution. It uses a "Hierarchy of Trust" security model to prevent dependency confusion attacks.

Usage

import dependency_manager

Method 1: The Smart List (Recommended)

Automatically resolves pip names using the Internal Golden Map.

dependency_manager.require(["requests", "cv2", "yaml"])

Method 2: The Explicit Override (For private packages or version pinning)

dependency_manager.require({ "my_module": "my-private-package>=1.0.0", "legacy_lib": "legacy-pkg==2.4.1" })

Debug Mode

Run your main script with '--debug' to see verbose installation logs. e.g., python my_script.py --debug