🤖 aicommit

Generate Conventional Commit messages from your staged diff, powered by a local LLM. No API keys. No cloud.

---

Contents

• Features
• Requirements
• Quick start
• Configuration
• Alternatives
• License

---

Features

• Generates Conventional Commit messages from staged changes
• Powered by a local LLM via Ollama (or llama-cpp-python)
• Opens the proposal in $EDITOR for a quick approval / edit / regenerate loop
• changelog subcommand: summarizes a tag range into a CHANGELOG.md entry
• Installs as a git subcommand — call it as git aicommit

---

Requirements

• Python 3.11+
• An LLM backend, either:
  • Ollama running locally or over the network (default), or
  • llama-cpp-python with a local GGUF model — pip install 'aicommit[llama-cpp]'

---

Quick start

# 1. install a model in Ollama (any chat-instruct model works)
ollama pull qwen2.5-coder:7b

# 2. install aicommit straight from GitHub
pip install git+https://github.com/yumiaura/AICommit

# 3. stage and let it write the message
git add -A
git aicommit

Sample interaction:

Text fallback (if the image above doesn't render)

────────────────────────────────────────────────────────
staged changes:
────────────────────────────────────────────────────────
 parser.py | 2 ++
 1 file changed, 2 insertions(+)
────────────────────────────────────────────────────────
proposed commit message:
────────────────────────────────────────────────────────
fix(parser): handle empty input gracefully

when input text is None, empty, or contains only
whitespace, return empty list instead of processing
invalid input
────────────────────────────────────────────────────────

[ Enter = commit · e = edit · r = regenerate · q = quit ]

Changelog mode:

git aicommit changelog v0.4.0..HEAD --out CHANGELOG.md

Copy the generated message to your clipboard (useful when you commit from another tool — IDE, GUI git client, GitHub web UI):

aicommit --print | pbcopy              # macOS
aicommit --print | wl-copy             # Wayland
aicommit --print | xclip -sel clip     # X11
aicommit --print | clip                # Windows (Git Bash / WSL)

---

Configuration

Quickest way to get a config file:

aicommit config

Creates ~/.config/aicommit/config.toml from the default template (if missing) and opens it in $EDITOR (fallback $VISUAL → nano). Existing files are never clobbered.

Config keys — all optional, defaults shown:

Section	Key	Env var	Default	Description
llm	backend	AICOMMIT_BACKEND	ollama	ollama or llama-cpp
llm	model	AICOMMIT_MODEL	qwen2.5-coder:7b	Ollama tag or GGUF file path
llm	url	AICOMMIT_OLLAMA_URL	http://localhost:11434	Ollama base URL
llm	temperature	AICOMMIT_TEMPERATURE	0.1	Sampling temperature
llm	max_tokens	AICOMMIT_MAX_TOKENS	512	Max tokens in response
commit	style	AICOMMIT_STYLE	conventional	conventional or plain
commit	include_body	AICOMMIT_INCLUDE_BODY	true	Emit a body under the subject
review	enabled	—	false	Run --review on every commit
changelog	skip_conventional	—	true	Use the deterministic fast path when possible

Precedence: defaults → ~/.config/aicommit/config.toml → <repo>/.aicommit.toml → env vars → CLI flags.

CLI flags — all override config:

aicommit [--backend {ollama,llama-cpp}] [--model M] [--url URL]
         [--temperature T] [--max-tokens N]
         [--style {conventional,plain}] [--no-body]
         [--review] [--review-only]
         [--print] [--no-stream] [-y/--yes] [--debug] [--version]
aicommit changelog <rev-range> [--out CHANGELOG.md]
aicommit config

---

Alternatives

aicommit lives in a crowded space, and that's fine — these are all good:

Tool	Stack	Notes
aicommits	Node	most popular; many providers, Ollama too
opencommit	Node	feature-rich; Claude/GPT/Ollama, GitHub Actions
gptcommit	Rust	prepare-commit-msg hook
CodeGPT	Go	commits + short code review

Why aicommit, then? A few deliberate choices:

• Local-first, offline by default — Ollama or in-process llama-cpp; no cloud provider to configure, no API keys, nothing leaves your machine.
• Zero runtime dependencies for the Ollama backend — just the Python standard library (urllib for HTTP, subprocess for git, argparse for the CLI, tomllib for config), so it's easy to audit and quick to install.
• Conventional Commits with an approve / edit / regenerate loop in $EDITOR.
• Changelog mode — summarize a tag range straight into CHANGELOG.md.
• Installs as a native git aicommit subcommand.

Want the biggest ecosystem? Use aicommits or opencommit. Want a tiny, auditable, Python-native, fully-offline tool? That's aicommit.

---

License

MIT · @yumiaura