gencommit

gencommit generates several Git commit-message candidates using your existing Codex or Claude Code subscription login, lets you choose one in the terminal, and commits the selected changes after confirmation.

Status

This is an early version. Review both the diff and generated message before approving a commit.

Requirements

• Rust 1.85 or newer
• Git
• OpenAI Codex CLI authenticated with a ChatGPT account
• or Claude Code authenticated with a Claude.ai account

ChatGPT/Claude subscriptions and provider API billing are separate. gencommit delegates generation to the selected coding CLI instead of accepting an API key, and does not read or copy authentication tokens.

Install and authenticate

Install the latest published crate:

cargo install gencommit

Alternatively, download the archive for your Linux or macOS architecture from the GitHub Releases page and place gencommit on your PATH.

For development builds:

cargo install --path .
gencommit auth login
gencommit auth status
gencommit provider
gencommit model

Usage

# Generate three variants for every changed path
gencommit

# Equivalent explicit form
gencommit --all

# Generate two variants, then stage and commit only these paths
gencommit -v 2 src/main.rs README.md

With explicit paths, unrelated staged changes remain staged and are excluded from the new commit. With no paths or --all, all worktree changes are staged and committed.

The selected patch is sent to Codex/OpenAI. Generation is read-only and ephemeral, but do not select files containing secrets you do not want to send to the model.

Configuration

Optional configuration is read from $XDG_CONFIG_HOME/gencommit/config.toml (normally ~/.config/gencommit/config.toml):

variants = 3
history_limit = 20
instructions = "Use Conventional Commits."
provider = "codex"
codex_executable = "codex"
model = "gpt-5.4-mini"
claude_executable = "claude"
claude_model = "haiku"

gencommit passes its model explicitly and ignores the global Codex model setting. This prevents an experimental model in ~/.codex/config.toml from breaking commit generation. Override model here when your ChatGPT plan and installed Codex CLI support a different model.

Run gencommit model to fetch Codex's current model catalog, choose a model with the arrow-key interface, and save it to this file.

Run gencommit provider to switch between Codex/ChatGPT and Claude Code/Claude.ai. gencommit auth and gencommit model operate on the selected provider. Claude generation uses safe mode, disables built-in tools, and does not persist sessions.

Repository configuration

Add .gencommit.toml to a repository root to share commit-message conventions with the project:

gencommit init

This creates the file from your current user-level generation settings and refuses to overwrite an existing project configuration. You can then edit and commit it. For example:

provider = "claude"
claude_model = "haiku"
variants = 3
history_limit = 20
instructions = "Use Conventional Commits with a package scope."

Repository settings override the user configuration for commit generation. The -v command-line option overrides both. For safety, repository configuration cannot change provider executable paths or authentication; gencommit provider, gencommit model, and gencommit auth continue to manage the user configuration.