gac - Git Auto Commit

Smart, succinct Git commit messages - free & local-first

Quick Start

# Install globally
npm install -g @thehelpfultipper/gac

# Or try without installing
npx @thehelpfultipper/gac

# Initialize configuration (optional)
gac init

[!IMPORTANT] > Update Required for Notifications

Version 1.0.6 introduces automatic update checks. If you installed gac before this version, you must update to at least this version if not newer to receive future notifications:

npm install -g @thehelpfultipper/[email protected]

Usage

# Stage your changes
git add .

# Run gac
gac

# OR stage all tracked changes automatically
gac -a

Features

• 🆓 Free by default - Uses local Ollama (no API costs)
• 🔒 Privacy-first - Code never leaves your machine
• 🎯 Smart fallback - Works without any LLM installed
• ⚡ Fast - Generates 3 options in seconds
• 📏 Convention-aware - Supports Conventional Commits, Gitmoji, plain style
• 🎨 Interactive - Choose, edit, regenerate, or customize on the fly

Command Options

Basic Usage

gac                          # Uses Ollama (default), auto-falls back to heuristic
gac init                     # Interactively create .gacrc configuration
gac -a                       # Stage all tracked files and generate message
gac --engine none            # Force heuristic mode (no LLM)
gac --engine ollama          # Explicitly use Ollama
gac --dry-run                # Preview without committing
gac --help or -h                  # Show help and all available options

Styling Options

The --style flag controls what format the commit messages use:

gac --style mix       # Default: 3 diverse formats (recommended)
gac --style conv      # Only Conventional Commits format
gac --style plain     # Only plain imperative format
gac --style gitmoji   # Only gitmoji format

Example outputs for each style:

--style mix (default - 3 different formats):

1. feat(auth): add session refresh on 401 response
2. Add session refresh on 401 response
3. ✨ auth: add session refresh on 401 response

--style conv (Conventional Commits only):

1. feat(auth): add session refresh on 401 response
2. fix(auth): handle expired tokens correctly
3. refactor(auth): simplify token validation logic

--style plain (plain imperative only):

1. Add session refresh on 401 response
2. Handle expired tokens correctly
3. Simplify token validation logic

--style gitmoji (emoji-prefixed only):

1. ✨ add session refresh on 401 response
2. 🐛 handle expired tokens correctly
3. ♻️ simplify token validation logic

Prefix Options

Add a prefix to all generated messages (useful for ticket tracking):

gac --prefix "JIRA-123: "    # Manual prefix
export GAC_PREFIX="JIRA-123: " && gac  # Via environment variable

Auto-detection from branch names: If your branch is named like feature/ABC-123-add-auth, gac automatically detects and suggests ABC-123: as the prefix.

Model Selection (Ollama)

Important: If using a model other than the default (mistral:7b), you must specify it:

# Via command flag
gac --model llama3.2:3b

# Via environment variable
export GAC_MODEL="llama3.2:3b" && gac

# Via config file (see Configuration section)

Length Control

gac --max-len 50     # Shorter messages (default: 72)
gac --max-len 100    # Longer messages (not recommended)

Complete Flag Reference

| Flag | Description | Default | | ------------------------- | ---------------------------------------------------------- | ------------------------------- | | -a, --all | Automatically stage all tracked modified/deleted files | false | | --style <type> | Message style: plain, conv, gitmoji, or mix | mix | | --engine <name> | Engine: ollama, openai, gemini, or none | ollama | | --model <name> | Model name (for Ollama, OpenAI, or Gemini) | mistral:7b | | --max-len <num> | Max subject line length | 72 | | --dry-run | Preview without committing or releasing | false | | --prefix <text> | Prefix for commit message | "" (auto-detects from branch) | | --changelog [version] | Generate/update CHANGELOG.md (optional version label) | — | | --changelog-path <path> | Override changelog file path | auto-detect | | --since <ref> | Generate changelog since a specific Git ref (tag/commit) | latest tag/heading | | --release | Auto-bump version, update changelog, and create Git tag | — | | --bump <level> | With --release, force bump: patch, minor, or major | auto-detect | | --release-as <version> | With --release, set exact version (e.g., v1.2.3) | auto-detect | | --update-pkg | With --release, also update package.json version | false | | -h, --help | Show help and all available options | — | | -V, --version | Show version number | — |

Engine Behavior

Default Behavior (Ollama with Automatic Fallback)

When you run gac without specifying an engine:

1. Tries Ollama first - Attempts to connect to http://127.0.0.1:11434
2. Auto-falls back to heuristic - If Ollama is unavailable/not running
3. No error shown - Seamless fallback with a notice: "Ollama unavailable, using heuristic fallback"

# These all behave the same way:
gac
gac --engine ollama

Force Heuristic Mode

To skip Ollama entirely and use the rule-based heuristic:

gac --engine none

The heuristic mode is smart enough that many users prefer it over LLM generation - it's instant and produces high-quality conventional commits.

OpenAI Engine

Security warning:

• Do not commit API keys to version control.
• Prefer environment variables; if you use a local .gacrc, keep it out of git (this repo's .gitignore includes .gacrc).

# Requires OPENAI_API_KEY environment variable
export OPENAI_API_KEY=sk-... && gac --engine openai

# Specify a model (examples):
gac --engine openai --model gpt-4o-mini
gac --engine openai --model gpt-4o

Notes:

• If --model looks like an Ollama model (contains :), gac defaults to gpt-4o-mini for OpenAI.
• The same prompt/formatting as Ollama is used; output is exactly 3 one-line subjects.
• You can also store the key in config (see Configuration) as openaiApiKey.

Gemini Engine

Security warning:

• Do not commit API keys to version control.
• Prefer environment variables; if you use a local .gacrc, keep it out of git.

# Requires GEMINI_API_KEY or GOOGLE_API_KEY environment variable
export GEMINI_API_KEY=ya29... && gac --engine gemini

# Specify a model (examples):
gac --engine gemini --model gemini-1.5-flash
gac --engine gemini --model gemini-1.5-pro

Notes:

• If --model looks like an Ollama model (contains :), gac defaults to gemini-1.5-flash for Gemini.
• Same prompt/formatting as other engines; output is exactly 3 one-line subjects.
• You can also store the key in config (see Configuration) as geminiApiKey.

Anthropic Engine (Placeholder)

The Anthropic engine is not implemented yet in this version. Use --engine ollama or --engine none instead.

Configuration

Automatic Configuration

Run the init command to create a .gacrc file interactively:

gac init

Option 1: .gacrc file

Create .gacrc in your project root:

{
  "engine": "ollama",
  "model": "llama3.2:3b",
  "style": "conv",
  "maxLen": 72,
  "prefix": ""
}

For OpenAI, you may include your API key (avoid committing this file):

{
  "engine": "openai",
  "model": "gpt-4o-mini",
  "openaiApiKey": "sk-..."
}

For Gemini, you may include your API key (avoid committing this file):

{
  "engine": "gemini",
  "model": "gemini-1.5-flash",
  "geminiApiKey": "ya29-..."
}

Option 2: package.json

Add a "gac" field to your package.json:

{
  "name": "my-project",
  "gac": {
    "style": "conv",
    "model": "phi3:3.8b",
    "engine": "ollama"
  }
}

Configuration Priority

Settings are merged in this order (highest priority last):

1. Default values
2. .gacrc file
3. package.json "gac" field
4. Environment variables (GAC_PREFIX, GAC_MODEL, OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY/GOOGLE_API_KEY)
5. Command-line flags (highest priority)

Example:

# .gacrc says model: "mistral:7b"
# But CLI flag overrides it:
gac --model llama3.2:3b  # Uses llama3.2:3b

Ignoring Files

Create a .gacignore file in your project root to prevent specific files from being analyzed by the AI.

Note on .gitignore vs .gacignore: gac only analyzes staged files. If a file is already in your .gitignore, it usually cannot be staged, so gac won't see it anyway. You only need .gacignore for files that are tracked by Git (and currently staged) but contain "noise" you don't want the AI to process (e.g., massive lockfiles, auto-generated code, or minified assets).

Example .gacignore:

# Lock files (tracked, but huge and noisy for AI)
package-lock.json
yarn.lock
pnpm-lock.yaml

# Auto-generated source files
src/gql/types.ts
src/aws-exports.js

# Minified assets (if you commit them)
public/js/*.min.js

.gacignore supports standard glob patterns (like .gitignore).

Installing Ollama

To use the default Ollama engine (these are bound to change, check website for updated installation instructions):

# macOS/Linux
curl -fsSL https://ollama.ai/install.sh | sh

# Windows
# Download from https://ollama.ai

# Start Ollama service
ollama serve

# Pull a model (in another terminal)
ollama pull mistral:7b
# or for faster generation:
ollama pull llama3.2:3b

Keep Model Loaded for Speed

By default, Ollama unloads models after 5 minutes of inactivity. To keep them loaded:

export OLLAMA_KEEP_ALIVE=1h
ollama serve

Interactive Mode

When you run gac, you'll see a two-step interactive menu.

1. Choose a Message

First, select a starting point from the generated options.

◇ Choose a commit message:

❯ ✓ 1. feat(auth): add session refresh on 401 response (50/72 chars)
  ✓ 2. Add session refresh on 401 response (38/72 chars)
  ⚠ 3. ✨ auth: implement automatic session refresh on token expiry (73/72 chars)
  ↻ Regenerate new options (r)
  ✎ Set/change prefix (p)
  ✕ Quit without committing (q)

  Use ↑/↓, Enter, or shortcuts (1-3, r, p, q)

Keys:

• ↑ / ↓ - Navigate options
• 1, 2, 3 / Enter - Select an option to proceed to the edit/confirm step
• r - Regenerate 3 new options
• p - Add/change prefix (e.g., add ticket number)
• q / Ctrl+C - Quit without committing

Indicators:

• ✓ Green checkmark - Under 72 characters (good)
• ⚠ Yellow warning - Over 72 characters (consider regenerating)

2. Edit and Confirm

After selecting a message, you get a final chance to edit it before committing. This allows for quick tweaks or additions.

? Edit commit message: (Enter to commit, Ctrl+C to cancel)
❯ feat(auth): add session refresh and handle 401 response

• Enter - Confirms and commits the message.
• Ctrl+C - Cancels the edit and returns to the selection menu.

Changelog & Releases

When to use which flag:

• gac --changelog [version]

  • Use to generate or update CHANGELOG.md without tagging.
  • Great for previewing (--dry-run) or preparing notes before a release.
  • If version is provided (e.g., v1.2.3), inserts/replaces that section.
  • If omitted, uses an “Unreleased - YYYY-MM-DD” heading and the latest tag as the “since” point.
  • Optional: --since <ref> to choose a custom starting point, and --changelog-path to write somewhere else (e.g., docs/CHANGELOG.md).

• gac --release

  • One step release: determines the next semver version from commits since the last tag, updates CHANGELOG.md, and creates an annotated Git tag (vX.Y.Z). If an ## Unreleased section exists, it's automatically replaced by the new versioned section.
  • Prefer explicit control:
    • --bump patch|minor|major to increment from the last version
    • --release-as vX.Y.Z to set the exact target version
  • Use --dry-run to preview without writing anything.
  • Add --update-pkg to also bump package.json’s version to X.Y.Z.

Are these redundant?

• No. --release includes changelog generation but also handles bumping and tagging. --changelog is for manual or ad‑hoc changelog updates (preview, custom label/path/range) without touching tags or versions.

How versioning is determined (--release)

• Base version source: the most recent semver Git tag matching vX.Y.Z. If none, falls back to package.json’s version or 0.0.0.
• You can control the target in three ways:
  • Explicit target: --release-as vX.Y.Z
  • Guided increment: --bump patch|minor|major
  • Auto (optional): If neither is provided, a simple Conventional-Commits-aware detection runs; if your commits aren’t conventional, it defaults to patch.
• Next version: tags as vX.Y.Z and optionally updates package.json with --update-pkg.

Changelog formatting basics

• Groups entries into sections (Breaking Changes, Added, Changed, Fixed, etc.).
• Adds date to headings; includes a repository compare link when possible.
• Detects breaking changes via ! after type or BREAKING CHANGE(S): in body.

Examples

# Preview changelog entries since last tag without writing
gac --changelog --dry-run

# Generate/replace a specific section label (no tag)
gac --changelog v1.2.3

# Create a release: compute next version, update changelog, create tag (auto bump)
gac --release

# Also update package.json version
gac --release --update-pkg

# Explicit minor bump from last version
gac --release --bump minor

# Set exact version
gac --release --release-as v2.0.0

# Use a custom changelog path
gac --changelog --changelog-path docs/CHANGELOG.md

# Generate changes since a specific tag/commit
gac --changelog --since v1.0.0

Example 1: Quick commit with auto-prefix

# On branch: feature/PROJ-456-add-login
git add src/auth/login.ts
gac
# Auto-detects "PROJ-456: " prefix
# Select an option, then confirm or edit the final message:
# ? Edit commit message: (Enter to commit)
# ❯ PROJ-456: feat(auth): add login form validation

Example 2: Force specific style

git add .
gac --style conv --model llama3.2:3b
# All 3 options will be Conventional Commits format
# Using faster 3B model

Example 3: Preview without committing

git add README.md
gac --dry-run --prefix "docs: "
# Shows what would be committed, but doesn't actually commit

Example 4: Heuristic mode (no AI)

git add src/components/Button.tsx
gac --engine none
# Instant results using smart rule-based generation

Why gac?

• Most commit message generators require paid API keys
• Many produce verbose, multi-line messages
• gac keeps it short (≤72 chars), structured, and scannable
• Perfect for teams using Conventional Commits or changelog automation
• Works offline with zero cost

Troubleshooting

"Ollama unavailable, using heuristic fallback"

• Ollama isn't running - start with ollama serve
• Wrong model name - check available models with ollama list
• Port conflict - ensure 11434 is free

"Nothing staged"

• Run git add <files> before using gac

Messages are too long

• Use --max-len 50 for shorter messages
• Regenerate until you get a shorter option

Slow generation

• Switch to smaller model: --model llama3.2:3b
• Keep model loaded: export OLLAMA_KEEP_ALIVE=1h
• Use --engine none for instant heuristic mode

Secrets in config

• Prefer environment variables for API keys in shared repos.
• If you store keys in .gacrc, add it to .gitignore so it isn’t committed.

License

MIT