Question 1

How does the git commit message generator work?

Accepted Answer

Paste your staged git diff (the output of `git diff --cached`) and the tool parses it into file-level changes — for each file it captures the path, change kind (added / modified / deleted / renamed), language, and the line counts of additions and deletions. It then classifies the overall change type (feat, fix, docs, refactor, test, chore, perf, style, ci, or build) using filename and content heuristics, detects breaking changes (BREAKING CHANGE: markers, !-suffix, or deleted public APIs), extracts issue refs (#123, JIRA-456), and composes a Conventional Commits message with subject, body, and footer in the style you choose.

Question 2

What styles does it support?

Accepted Answer

Four: (1) Conventional Commits subject-only — `feat(parser): add support for nested arrays`. (2) Conventional Commits with body — adds a wrapped paragraph describing what and why. (3) Gitmoji — `✨ feat(parser): add support for nested arrays` with an emoji picked by change type. (4) Plain — a normal imperative sentence without the type/scope prefix. All four also include the optional body, footer (issue refs + breaking change note), and a copy-paste-ready message.

Question 3

How does it handle large diffs?

Accepted Answer

Large diffs (over ~5,000 lines or 50 files) are chunked: each file is summarized to its kind, language, additions/deletions, and a top-line content hint (function names touched, primary keywords). The classifier then operates on the chunk summaries rather than the raw diff — so a 50,000-line refactor still produces a single readable message. The tool flags when it has summarized and reminds you to review the body.

Question 4

What extra features does this tool have compared to other commit generators?

Accepted Answer

(1) Diff parser for unified-diff format (added / modified / deleted / renamed, with hunk counts). (2) Per-file language detection from extension. (3) Change-type classifier (feat / fix / docs / refactor / test / chore / perf / style / ci / build) with filename + content heuristics. (4) Breaking-change detection (BREAKING CHANGE:, ! suffix, deleted public exports). (5) Issue-ref extraction (#123, ABC-123). (6) Four commit styles: conventional, conventional+body, gitmoji, plain. (7) Conventional Commits validator (subject length, type, scope, tense). (8) Large-diff chunked summarization (map-reduce). (9) prepare-commit-msg hook snippet generator. (10) Gitmoji map (12 types). (11) Team convention preset persistence (localStorage). (12) History (localStorage, last 20). (13) Shareable URL with diff encoded (size permitting). (14) Copy / download as .txt and .md. (15) Optional BYO-key LLM polish (OpenAI / Anthropic). (16) Honesty disclaimer (AI infers intent from the diff — review before committing).

Question 5

Is my diff sent anywhere?

Accepted Answer

No. All parsing, classification, and message generation run locally in your browser. Your diff (which often contains proprietary code) never leaves this device — no account, no upload. The only network path is if you explicitly paste your own LLM API key and click 'Polish with LLM' — that request goes directly from your browser to your chosen LLM provider (OpenAI or Anthropic) and never touches UnQTools servers. Skip the LLM step for 100% offline use.

AI Git Commit Message Generator

Style + overrides

How to use

Style + overrides