Git tells you what.
Anubis tells you why.

Automatic reasoning traces for AI-assisted coding. Every change documented with context. Code review that finally makes sense.

$ pip install anubis-vcs

Watch Claude Code auto-checkpoint with reasoning

Terminal — anubis
Anubis auto-checkpointing with Claude Code

The Problem

× Code review shows what changed, never why
× AI makes 50 changes—no explanation of the approach
× Context lost when sessions end or conversation resets
× Commit messages like "fix bug" or "update code"

The Solution

Every checkpoint captures the reasoning behind changes
Native Claude Code integration via MCP—no copy/paste
Resume any session with full context restored
Semantic analysis: "added function", "renamed class"

How it works

1

Connect to Claude Code

One command adds Anubis as an MCP server. Zero configuration.

2

Code with AI

Work normally. Anubis captures checkpoints with reasoning as you go.

3

Resume anytime

Claude Code reads context directly. Pick up exactly where you left off.

Quick start

# Initialize in any git repo
anubis init

# Connect to Claude Code via MCP
claude mcp add anubis -- anubis-mcp

# Create checkpoints with reasoning
anubis checkpoint "Refactoring auth" -r "JWT approach - sessions don't scale"

# View checkpoint history
anubis log
→ 3f8a2b1 (10 min ago) Refactoring auth
  Reasoning: JWT approach - sessions don't scale

# Claude Code can now access checkpoints directly!