NORTHTEKDevs/genome

Overview

Open source Python memory layer for AI agents. It ingests messages locally without LLM calls, roughly 10ms per message, and can run air-gapped. It provides a bi-temporal belief-state and an MCP server, with benchmark results against Mem0 and LoCoMo/LongMemEval.

Best for

Best for
Developers building cost-sensitive, privacy-focused AI agents

Use cases

• Add persistent memory to an AI agent without per-message LLM costs
• Run agent memory in an air-gapped environment
• Expose memory to MCP-compatible clients via the built-in server

How to use

Install

pip install -e . && python -m genome.verify

Tools exposed

• gpt-4o-mini

Tested with

Claude Desktop, Claude Code, Cursor, ChatGPT

Notes

Open source Python memory layer for AI agents. It ingests messages locally without LLM calls, roughly 10ms per message, and can run air-gapped. It provides a bi-temporal belief-state and an MCP server, with benchmark results against Mem0 and LoCoMo/LongMemEval.

6 stars on GitHub. Last updated 2026-09-04. Licensed Apache-2.0.

Use cases

• Add persistent memory to an AI agent without per-message LLM costs
• Run agent memory in an air-gapped environment
• Expose memory to MCP-compatible clients via the built-in server

Pros

• Zero LLM calls during ingest, so low cost
• Air-gapped capable
• Apache-2.0 open source

Cons

• Low GitHub stars (6) suggest early-stage project
• Bi-temporal model may have a learning curve
• Benchmarks are self-reported

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.