DiffMem addresses the challenge of managing evolving memory for conversational AI agents by leveraging Git as a version control system for knowledge storage. It offers a human-readable, durable, and portable memory backend that prioritizes current-state efficiency while enabling deep historical analysis through Git's differential capabilities. This project is ideal for developers building long-horizon AI systems that require auditable and scalable memory management.

How It Works

DiffMem treats AI memory as a versioned repository, storing the current state of knowledge in Markdown files and utilizing Git for tracking temporal changes. This design separates the "surface" (current information) for fast, token-efficient queries from the "depth" (historical data) accessible on demand via Git's commit graph. An in-memory BM25 index facilitates quick, explainable retrieval, while Git diffs allow agents to efficiently query how specific facts have evolved over time.

Quick Start & Requirements

• Install: pip install -r requirements.txt (after cloning the repository).
• Prerequisites: Git, Python, alexmrval/DiffMem repository, OPENROUTER_API_KEY environment variable.
• Setup: Clone the repository and install dependencies. Running examples is straightforward.
• Links: Examples directory within the repository.

Highlighted Details

• Leverages Git for efficient, scalable tracking of memory evolution.
• Reduces query surface by indexing only current-state files for faster, more token-efficient LLM interactions.
• Enables "differential intelligence" by allowing agents to query specific changes over time using Git diffs.
• Uses Markdown for human-readable and tool-agnostic memory storage.

Maintenance & Community

This is an R&D project from Growth Kinetics, a boutique data solutions agency. They are open to collaborations, pull requests, and feedback. Contributions are welcomed for Git sync optimizations, advanced search plugins, and real-world integrations.

Licensing & Compatibility

• License: MIT
• Compatibility: Permissive for commercial use and integration with closed-source applications.

Limitations & Caveats

This is an early Proof-of-Concept (PoC) and is not production-hardened. It lacks automatic Git synchronization, has basic error handling, rebuilds the index on every initialization (caching is recommended for production), and does not implement multi-user concurrency locks.