turboCommit

A CLI tool that uses the GPT-6 family to write Conventional Commit messages from Git and Jujutsu (JJ) changes. GPT-6 Luna is the default for its speed and low cost; Sol and Astra are available for harder changes.

Features

• GPT-6 Luna by default, with Sol and Astra available via --model
• One suggestion by default; request more with -n
• Prompt tuned for commit intent, JJ descriptions, and concise output
• Conventional Commit suggestions from staged Git changes or a JJ revision
• Interactive review, editing, and revision when --auto-commit is not set
• Reasoning effort controls: none, low, medium, high, xhigh, max (Astra starts at low)
• Verbosity controls: low, medium, high
• Structured JSON outputs for stable multi-suggestion parsing
• Debug logging for requests, responses, token usage, and timing
• YAML configuration via ~/.turbocommit.yaml

Installation

cargo install --git https://github.com/dikkadev/turboCommit

For the latest version published to crates.io, use cargo install turbocommit. The published release may lag behind the repository.

Optional shell alias:

alias tc='turbocommit'

Usage

1. For Git, stage your changes. JJ uses the selected revision's changes directly.

git add .

2. Generate one suggestion and review it.

turbocommit

The interactive flow lets you accept, edit, revise, or abort. For the common automatic flow, use turbocommit --auto-commit. With Git, it creates a commit; with JJ, it sets the selected revision's description. It always requests one suggestion, regardless of -n order.

# Describe the current JJ change without prompts
turbocommit --auto-commit

# Describe a specific JJ revision, using its current description as a hint
turbocommit --auto-commit --revision <rev> --rw

Options

• -n <number>: number of suggestions, default 1 in interactive mode
• -m, --model <model>: gpt-6-luna (default), gpt-6-sol, or gpt-6-astra
• -e, --reasoning-effort <level>: none, low (default), medium, high, xhigh, max; Astra does not support none
• -v, --verbosity <level>: low, medium, high
• -d, --debug: print request and usage details
• --debug-file <path>: write detailed debug logs to a file, or - for stdout
• -a, --auto-commit: write the generated Git commit or JJ description without interactive review
• --amend: regenerate the last Git commit message from its diff
• --api-key <key>: provide API key directly
• --api-endpoint <url>: override the API endpoint
• -c, --config <path>: load a non-default config file
• -r, --revision <rev>: select a JJ revision to describe
• --rw: toggle JJ rewrite mode, which passes the current description to the model as context

Reasoning

Reasoning effort defaults to low for fast commit generation. Use a higher level for ambiguous or broad diffs.

turbocommit --auto-commit
turbocommit --reasoning-effort high -m gpt-6-sol
turbocommit --reasoning-effort none -m gpt-6-luna

Verbosity

turbocommit --verbosity low
turbocommit --verbosity medium
turbocommit --verbosity high

Debugging

turbocommit -d
turbocommit --debug-file debug.log
turbocommit --debug-file -

Debug logs include request parameters, API responses or errors, token counts, and elapsed time.

Pricing

Standard GPT-6 Luna API pricing per 1M text tokens:

• Input: $0.10
• Cached input: $0.01
• Output: $0.50

All three supported models have a 1.05M token context window. OpenAI charges higher rates for prompts above 272K input tokens; see the model pages. The CLI uses v1/chat/completions, which supports these models and the current structured JSON response format. OpenAI recommends Responses for new text generation applications; this CLI keeps its working Chat Completions integration and configurable compatible endpoint.

Configuration

turboCommit creates ~/.turbocommit.yaml on first run.

Example:

model: "gpt-6-luna"
default_number_of_choices: 1
reasoning_effort: "low"
verbosity: "medium"
disable_auto_update_check: false
api_endpoint: "https://api.openai.com/v1/chat/completions"
api_key_env_var: "OPENAI_API_KEY"

Omit system_msg to use the current built-in prompt. A custom nonempty system_msg overrides it. New generated config files omit the default prompt, so future prompt improvements apply automatically. Existing config files created by version 3.x are upgraded in memory from the GPT-5.4 defaults to Luna, one suggestion, and the new prompt. Custom prompts and nondefault suggestion counts are preserved. This migration does not rewrite your config file; edit the stored model or remove the old default prompt when you want the file itself to reflect the new settings.

Multiple Config Files

turbocommit -c ./local-config.yaml
turbocommit -c ~/.turbocommit-azure.yaml
turbocommit

Amend Flow

Use --amend when you want to improve the last commit message without staged changes.

git status
turbocommit --amend
turbocommit --amend --auto-commit

Constraints:

• no staged changes when using --amend
• the tool analyzes the previous commit diff only

Git Hooks

Recommended workflow:

1. Stage and commit normally.
2. Fix any hook failures.
3. Re-stage fixes if needed.
4. Use turbocommit --amend after checks pass if you want a better message.

JJ does not use Git staging. Run turbocommit --auto-commit in a JJ workspace to describe @, or pass --revision to describe another change. The command updates the description; it does not create a new JJ change.

Dev Container Test Environment

A disposable Dev Container is included for validating Git and JJ integration without touching real repositories.

devcontainer up --workspace-folder .
devcontainer exec --workspace-folder . bash

Contributing

Issues and pull requests are welcome.

License

Licensed under MIT. See LICENSE.