git-ac AI Commit Message (Bash + Ollama or tgpt)

A tiny Bash helper that generates ultra-short Git commit messages from your staged changes, asks you to confirm/edit, and commits.

• Default engine: Ollama with the gpt-oss:20b-cloud model (ollama cloud).
• Alternative engine: tgpt (CLI chatbot without API keys).
• If generation fails, it falls back to manual input.

---

How it works

1. git add .
2. Read the staged diff (git diff --cached). If empty → exit.
3. Ask an LLM for a very short commit message.
4. Show it, then:
   • Y: commit,
   • n: cancel,
   • e: open in $EDITOR, then commit.

---

Requirements

• Git, Bash (Linux/macOS; on Windows use Git Bash or WSL).
• For Ollama mode:
  • Install Ollama and have the mistral model available. :contentReference[oaicite:0]{index=0}
• For tgpt mode (optional):
  • Install tgpt (cross-platform CLI; no API keys required). :contentReference[oaicite:1]{index=1}

---

Install

1. Save the script as git-ac in your repo (or somewhere in PATH).

2. Make it executable:

   chmod +x git-ac
   

3. (Optional) choose the editor used by e:

   export EDITOR=micro   # or nano, geany, vim, ...
   

---

Using Ollama (default)

1. Install Ollama

   • Linux/macOS:

     curl -fsSL https://ollama.com/install.sh | sh
     

     ([Ollama][1])

   • Windows: download the installer from the official site. ([Ollama][2])

2. Get the model (pulled automatically on first run):

   ollama run mistral "hello"
   

   ([Ollama][3])

3. Run the script inside your Git repo:

   ./git-ac
   

---

Switching to tgpt (optional)

If you prefer tgpt instead of Ollama, install it and swap two lines in the script.

Install tgpt

Pick one method (see repo for others):

• Linux/macOS (install script to /usr/local/bin):

  curl -sSL https://raw.githubusercontent.com/aandrew-me/tgpt/main/install | bash -s /usr/local/bin
  

• Homebrew:

  brew install tgpt
  

• Go:

  go install github.com/aandrew-me/tgpt/v2@latest
  

• Windows (PowerShell):

  irm https://raw.githubusercontent.com/aandrew-me/tgpt/refs/heads/main/install-win.ps1 | iex
  

  ([GitHub][4])

Minimal code change

Replace the Ollama line:

msg=$(ollama run mistral "Write a very short git commit message from:\n$diff")

with the tgpt block that also cleans spinners/fences:

raw=$(tgpt "Write ONLY a very short git commit message from:\n$diff")
msg=$(echo "$raw" | sed 's/.*Loading[[:space:]]*//' | sed 's/^[[:space:]]*//' | sed -n '/^```/{n;:a;/^```/q;p;n;ba}')

Notes

• tgpt sometimes prints a “Loading …” spinner or wraps output in ````` code fences; that one-liner strips both.
• Keep the prompt strict (“ONLY a very short…”) to avoid verbose outputs.

---

Tips

• Want even shorter messages? Soften the prompt to “ONE line, max 8 words”.
• Prefer imperative style? Add “Use imperative tense” to the prompt.
• Different Ollama model? Replace mistral with any installed model, e.g. gemma:2b, llama3.2:1b. ([GitHub][5])

---

Troubleshooting

• No changes to commit. You forgot to stage files. Run git add -p or git add ..
• ollama: command not found Install Ollama and/or restart your shell so PATH is updated. ([Ollama][1])
• model not found or long first run Ollama downloads the model on first use; wait for mistral to pull. ([Ollama][3])
• tgpt: command not found Install tgpt via script/Homebrew/Go/Windows installer. ([GitHub][4])

---

License

MIT (for this script). Ollama, models, and tgpt are licensed separately—see their repos/websites.

[1]: https://ollama.com/download?utm_source=chatgpt.com "Download Ollama on Linux"
[2]: https://ollama.com/download/windows?utm_source=chatgpt.com "Download Ollama on Windows"
[3]: https://ollama.com/library/mistral?utm_source=chatgpt.com "mistral"
[4]: https://github.com/aandrew-me/tgpt "GitHub - aandrew-me/tgpt: AI Chatbots in terminal without needing API keys"
[5]: https://github.com/ollama/ollama?utm_source=chatgpt.com "ollama/ollama: Get up and running with OpenAI gpt-oss, ..."