Skip to main content

Changelog

v1.3.0

Init Command Improvements

  • agit init --update now shows "Already up to date." when protocol version matches
  • Added restart warning after init/update changes: "Restart your AI assistant to apply the updated AGIT memory protocol."
  • Improved version detection for protocol updates

MCP Tool Enhancements

agit_log_step improvements:

  • New repo_path parameter for cross-repo awareness - validates target repo matches current repo
  • Entries with all-invalid locations are now rejected entirely (not just locations stripped)
  • Improved error messages distinguishing rejected entries vs skipped locations

Example with repo_path:

{
"name": "agit_log_step",
"arguments": {
"repo_path": "/path/to/repo",
"batch": [...]
}
}

v1.1.1

File Location Tracking

Neural context entries now include file path and line number where thoughts originated:

{"role":"ai","category":"reasoning","content":"...","file":"/src/auth.rs","line":42}

This enables better traceability between reasoning and specific code locations.


v1.1.0

Git Workflow Resilience

Major improvements to handle advanced Git workflows:

  • Merge/Rebase Guard - Blocks mutating commands during merge or rebase to prevent graph corruption
  • Amend Detection - Automatically detects git commit --amend and migrates neural memory to the new hash
  • Per-Branch Index Stashing - Preserves pending thoughts when switching branches
  • Rewind Detection - Handles git reset --hard by snapping to valid ancestor

See Git Resilience for details.

Semantic Conflict Detection

Protects against "hallucinated" commits by detecting when external Git commits have modified files mentioned in pending thoughts:

  • Detects "ghost commits" (external changes since last neural commit)
  • Extracts file references from pending thoughts
  • Blocks commit if files overlap
  • Use --force to override

Search Functionality

Full-text search of reasoning history:

agit search query "authentication"
agit search rebuild

Features:

  • Tantivy-based full-text indexing
  • Auto-indexing on commit
  • Incremental updates after pull
  • Manual rebuild support

See Search Command for details.

File History MCP Tool

New agit_get_file_history tool for retrieving neural commit history for specific files:

{
"name": "agit_get_file_history",
"arguments": {
"filepath": "src/auth.rs",
"limit": 5
}
}

Enables AI auto-context injection before modifying files.


v1.0.0

Initial Release

Core neural commit system:

  • Neural Graph - Parallel commit graph linking reasoning to code
  • Content-Addressable Storage - SHA-256 object store
  • MCP Integration - Model Context Protocol server for AI editors
  • Staging Area - JSONL-based index for pending thoughts
  • Synthesizer - Deterministic summary generation

CLI Commands

  • agit init - Initialize AGIT in a repository
  • agit record - Manually record thoughts
  • agit add - Stage files and freeze context
  • agit status - View current status
  • agit commit - Create neural + git commit
  • agit log - View commit history
  • agit show - Show full commit details

Editor Integrations

  • Cursor (via .cursorrules and .cursor/mcp.json)
  • Claude Code (via CLAUDE.md and .mcp.json)

Installation Methods

  • Homebrew (macOS)
  • Scoop (Windows)
  • Shell script (Linux/CI)
  • Cargo (from source)