Architecture

Architecture

MemoV is a Git-powered version control layer designed specifically for AI-assisted development. It captures three bound elements of each interaction: user prompts, agent plans, and code changes, storing them together in a shadow .mem/memov.git repository separate from standard Git.

Three-Layer Design

  • Interface Layer - CLI, Web UI, MCP clients
  • Core Management - MemovManager (central orchestrator), GitManager (subprocess-based Git operations)
  • Storage - Bare Git repository with JSON metadata

Six Core Features

FeatureDescription
Context-Bound MemoryAutomatically captures prompts, plans, and code changes as atomic units via the snap tool
Context-Aware DebuggingIsolates faulty interactions and enables testing with different LLM providers
Team Context SharingThe .mem/memov.git bare repository can be pushed/pulled like standard Git
Change ReuseQuery past commits by description to reapply proven solutions without regeneration
Loop GuardDetects infinite agent loops through pattern analysis of recent commits
History-Driven OptimizationUses past interactions as training corpus to improve future generation quality

MCP Integration

The MemMCPTools class translates Model Context Protocol invocations into MemovManager method calls, enabling transparent integration with Claude, VS Code, and Cursor.

File Classification

MemoV distinguishes between three types of file changes:

TypeDescriptionAttribution
Untracked filesNew files not yet trackedRequire track()
Manual editsModified but not in declared files_changedby_user=True
AI modificationsModified and in files_changedby_user=False

This enables accurate attribution and prevents misclassification of code provenance.

Separation from Git

MemoV maintains complete independence from standard Git:

  • No .git/ modification or pollution
  • Manual commits control production history
  • MemoV tracks conversational context at per-interaction granularity
  • Both timelines coexist without interference

Technology Stack

Core Dependencies

PackagePurpose
FastMCPModel Context Protocol
FastAPIWeb framework
TyperCLI framework
Git CLISubprocess-based operations

Optional RAG Features

PackagePurpose
ChromaDBVector storage
sentence-transformersEmbeddings
LiteLLMMulti-model LLM support

Supported Python Versions: 3.11 - 3.13