Skip to content

regit: safely rewrite the commit history of your current git branch — TUI + headless CLI

Bursts of work — often AI-assisted, or history that was squashed and recreated — leave every commit stamped on the same day, with messy messages, oversized commits, and sometimes secrets buried in past revisions. Cleaning that up with raw git means hand-editing rebase todo lists with no preview, no date strategies, and no guarantee you didn't corrupt anything.

regit rewrites the current branch's history under a shared safety core: strategy-driven date spreading, LLM-assisted commit splitting, rewording, and squashing, plus history hygiene (purge files, redact secrets). File contents never change — every rewrite reuses the original tree objects and is verified byte-for-byte before any ref moves — and every apply is reversible via a backup ref and undo. Everything works both as an interactive terminal UI and as a scriptable headless CLI.


Get started View on GitHub

Features

  • Date spreading (redate) — 10 day-assignment strategies × 7 time-of-day models to spread same-day commit bursts across a realistic calendar.
  • LLM commit splitting (split) — split one commit into 2–4 semantic commits via LLM hunk grouping, with optional --test-cmd validation and repair loop.
  • LLM rewording and squashing (reword, squash) — regenerated messages with ranked alternatives; consecutive commit runs collapsed into one per chain.
  • History hygiene (purge, redact) — remove a file or directory as if never added, or remove/replace a literal secret across the branch's history.
  • Safety core — byte-for-byte tree preservation, per-commit change-set verification, monotonic date ordering, backup refs, and in-app undo.
  • Preview by default — every mutating command shows its plan first; nothing is applied without --execute (and --force for already-pushed commits).
  • TUI + headless parity — every TUI action is projected from one capability registry into a non-interactive command, with a command-log footer showing the exact equivalent.
  • Flexible LLM backends — any OpenAI-compatible HTTP endpoint (Ollama, LM Studio, OpenAI, OpenRouter) or a local agent CLI, with env-only API keys.

Quick install

go install github.com/wlame/regit/cmd/regit@latest

What's next?