
vibemit
AI-generated Git commit messages using a local LLM via Ollama.
How It Works
Reads your staged diff via
git diff --stagedSends the diff to a local Ollama model with a carefully tuned prompt
Parses the response into 3 distinct commit message candidates
Presents an interactive selection menu
Commits with the selected message (or copies/prints based on flags)
The model runs locally — no data leaves your machine.
Prerequisites
Node.js >= 20
Ollama installed and running
A pulled model (default:
qwen3:8b)Install
Or run directly:
Usage
Stage your changes, then run:
vibemit reads your staged diff, generates 3 commit message candidates using a local LLM, and lets you pick one interactively.Flags
--model <name> Ollama model to use (default: qwen3:8b)
--intent <text> High-priority intent guidance for commit wording
--conventional Conventional Commit format (type(scope): subject)
--body Include subject + body (1-3 bullet points)
--dry-run | -d Print selected message, do not commit
--clipboard | -c Copy selected message to clipboard, do not commit
--yes | -y Auto-select the first option (skip prompt)
--add-rule <text> | -r Add a persistent rule
--rules Print saved rules
--clear-rules Delete all saved rulesExamples
Flag Combinations
Flags Behavior (none) Select message, then
git commit --dry-run Print only, no commit --clipboard Copy only, no commit --dry-run --clipboard Print and copy, no commit --yes Auto-select first, then commit --yes --dry-run Auto-select first, print onlyRules
vibemit supports persistent rules that are included in every prompt. Rules are stored per-repo when inside a Git repository, with a global fallback.
Storage locations:
Per-repo:
.git/vibemit.jsonGlobal fallback:
~/.config/vibemit/config.jsonDevelopment
Watch mode:
License
MIT
0 likes · 11 views · Feb 17, 2026
Created a CLI tool that generates AI commit messages using a local LLM via Ollama. No data leaves your machine.