Git-AI

Git-AI is a professional, provider-agnostic CLI that turns your staged Git changes into high-quality commit messages and AI-powered code reviews. This Rust rewrite delivers faster startup, safer error handling, and multi-platform release builds.

Features

• AI-generated commit messages with ticket-prefix detection
• AI reviews with target-branch readiness checks, fast local mode, task/comment verification, and large-diff exclusions
• Provider-agnostic HTTP client (no vendor SDKs)
• Environment-based configuration for secrets
• Arrow-key interactive configuration for provider, model, commit format, and themes
• Saved terminal color themes with custom RGB palettes and automatic dark/light detection
• Cross-platform release builds for Linux, Windows, and macOS

Installation

From Source

git clone https://github.com/samrand96/git-ai.git
cd git-ai
cargo build --release

The binary will be available at:

target/release/gai

Cargo Install (local)

cargo install --path .

Usage

Generate a Commit Message

git add .
gai commit

Optional flags:

gai commit --format one-line
gai commit --provider openai --model gpt-4o-mini
gai commit --push
gai commit --no-verify

AI Review

gai review
gai review --target develop
gai review --target origin/staging --task --comments
gai review --fast
gai review --fast --task --comments
gai review --large-threshold 800 --max-changed-files 50 --max-added-lines 800 --max-deleted-lines 500
gai review --no-large-check
gai review --markdown --output review.md

By default, gai review fetches remotes, lets you navigate origin branches with arrow-key folder navigation, compares the selected target branch against HEAD, detects large or binary file diffs, and lets you exclude those files before the AI review payload is sent.

Use --fast to skip remote fetching and target-branch comparison and review only current staged and unstaged changes. Use --task to paste the original ticket or acceptance criteria and have the report verify whether the branch appears to complete it. Use --comments to paste previous PR comments or reviewer feedback and check whether they were addressed; --coments is accepted as an alias. Large-diff checks are enabled by default with a 5000 changed-line threshold; use --large-threshold or the other max options to tune the policy, or --no-large-check to disable it explicitly.

List Models

gai list-models

Configure Settings

gai config --interactive
gai config --theme
gai config --set PROVIDER openai
gai config --set THEME matrix
gai config --set-provider openai MODEL gpt-4o-mini

The interactive config flow uses arrow-key menus for provider, commit format, theme selection, and model selection. It checks the configured endpoint for available models after the API key and base URL are set; if discovery fails or returns no models, it falls back to manual model entry. Custom themes can be created from RGB triples such as 80,255,145 or hex values such as #50ff91.

Environment Variables

Secrets must be stored in environment variables. Git-AI loads them from your shell environment or from ~/.config/gai/.env. Provider-specific keys take priority over the generic AI_API_KEY, so switching providers cannot silently reuse the wrong key.

Preferred variables:

• AI_API_KEY – API key for the active provider
• AI_BASE_URL – Override base URL (optional)
• AI_MODEL – Override model (optional)
• AI_PROVIDER – Override provider (optional)
• AI_COMMIT_FORMAT – detailed or one-line
• AI_THEME – Override saved theme (optional, defaults to auto)

Provider-specific variables:

• OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, GROQ_API_KEY, DEEPSEEK_API_KEY, PERPLEXITY_API_KEY
• OPENAI_BASE_URL, OLLAMA_BASE_URL, LMSTUDIO_BASE_URL, GROQ_BASE_URL, DEEPSEEK_BASE_URL, PERPLEXITY_BASE_URL

Configuration File

Git-AI stores non-secret defaults in:

~/.config/gai/config.toml

Example:

provider = "openai"
commit_format = "detailed"
theme = "auto"

[providers.openai]
model = "gpt-4o-mini"
base_url = "https://api.openai.com/v1"

[themes.lab-green]
error = [255, 92, 92]
success = [60, 220, 110]
warning = [190, 210, 95]
info = [96, 190, 135]
header = [80, 255, 145]
highlight = [226, 255, 230]
dim = [86, 122, 96]

Development Setup

cargo fmt
cargo clippy -- -D warnings
cargo test

Project Structure

src/
  main.rs
  lib.rs
  cli/
    config/
    review/
  config/
  services/
    ai/
  theme/
  utils/

For module ownership and extension points, see ARCHITECTURE.md.

GitHub Actions Builds

• .github/workflows/build.yml builds, formats, lints, tests, and uploads artifacts for Linux, Windows, and macOS.
• .github/workflows/release.yml builds tagged releases and uploads binaries to GitHub Releases.

Release Builds

Tag a release to trigger automated binaries:

git tag v1.0.0
git push origin v1.0.0

Artifacts are packaged per target and attached to the GitHub release.

Troubleshooting

• Missing API key: set the provider-specific key such as OPENAI_API_KEY or DEEPSEEK_API_KEY in your environment or ~/.config/gai/.env; AI_API_KEY is only a fallback.
• No changes to review/commit: stage or modify files and retry.
• Provider errors: verify AI_BASE_URL and model name for your provider.

To-Do / Future Plans

• AI Review: Professional code analysis with CLI colors, Markdown reports, target-branch readiness checks, and fast local review mode.
• Support for Conventional Commits: automatically enforce and generate messages in the type(scope): description format (feat, fix, docs, etc.).
• Pre-commit and commit-msg hooks: integrate Git hooks to validate or reformat the AI-generated commit before it’s saved.
• Issue tracker integration: detect and link JIRA, GitHub, or GitLab issue IDs to commits, and fetch issue titles for context.
• Automated semantic version tagging: analyze commit types and suggest or apply version bumps (major, minor, patch) and tag accordingly.
• Changelog automation: aggregate and format commit messages into a CHANGELOG.md, grouped by version and type.
• Custom branch naming rules: allow teams to define and validate branch patterns for consistent ticket prefixes.
• Offline/fallback mode: cache recent commit templates or model outputs so basic messages can be generated without internet.
• Performance profiling and optimization: track and reduce latency for commit generation in large repositories.
• AI security check-up: scan generated commit text for potential secrets, PII, or insecure patterns before committing.
• Template configuration: allow users to define custom commit templates with optional footers or co-author tags.
• Multi-language support: enable commit messages in different languages based on project locale or user preference.

License

MIT. See LICENSE.