A Multi Agent Memory MCP That Connect Agents Across Systems and Machines

Local-first AI memory — runs offline on any machine with 8 GB+ RAM (SBC, mini PC, laptop, workstation). Zero-loss verbatim archive, knowledge graph, hybrid retrieval. Framework-agnostic, no cloud.

Your AI forgets everything between sessions. This fixes that — 98%+ retrieval accuracy, 100% on LongMemEval, 99% token savings. 44 MCP tools. Fully local, zero cost.

Local-first Memory Framework for AI Agents · 99.2% LongMemEval-S retrieval @ k=10 · Supports Claude · Gemini · Antigravity · OpenCode · OpenClaw · Hermes · MCP-native and plugins · Hybrid search (FTS5 + vector + MMR) · GDPR · FIPS 140-3 ready · 100% local (fully offline) or cloud capable

Token-native agent memory retrieval for LLMs, without embedding APIs or vector databases.

Benchmark results, scorer, and reproducibility kit for Sibyl Memory. LongMemEval 95.6% (#2). Verify it yourself.

Reproducible benchmarks for execution-intent memory in long-horizon AI coding agents. ID-RAG cross-corpus matrix + LongMemEval-S subset; BYO API keys.

Multi-agent memory substrate for PostgreSQL — provenance-gated, vector-hybrid recall

Official Python SDK for RecallrAI – a revolutionary contextual memory system that enables AI assistants to form meaningful connections between conversations, just like human memory.

Retrain-free attention patch that makes Llama 3.3 70B ~1.3× more accurate on long-conversation memory

Public, reproducible benchmarks for Agent Brain on LongMemEval-M. 71.7% accuracy (Test 0). Companion code to https://doi.org/10.5281/zenodo.19673132 (Concept DOI → latest version, currently v3).

Smallest possible working example of CogmemAi (95.1% LongMemEval) wired into the Claude Agent SDK. Two-session demo: save in session 1, recall in session 2.

100-question 6-dimension long-conversation memory benchmark for Chinese-healthcare AI. Sivon reference: 92/100 mean (2026-05-27).

LENS - AI Memory Benchmark - Memory as Experience, Not Facts

Multi-agent strategic intelligence system with hybrid memory retrieval. Research project.

Benchmark harness for HeurChain on LongMemEval-S — reproduce the R@10, MRR, NDCG, and latency numbers from heurchain.com

LongMemEval-S benchmark results for Mnemo - 80.8% strict (404/500).

Reproducible evaluation harness for agent memory systems (LongMemEval and beyond).

Anti-RAG dual-whitebox memory for LLM agents. 2.72 MB SQLite + Markdown kernel, no vector DB, no embeddings. Lifts qwen2.5:7b from 1.79% to 60.71% on NoLiMa-32k (+58.9pp), 88.71% on LV-Eval EN 256k, 84.8% on LongMemEval-S. Restart-safe, concurrency-bullet-proof, 100% transparent.

fidelis is zero-LLM agent memory for Claude Code and AI agents: a local-first memory layer whose default retrieval path uses BM25, dense vectors, and reciprocal rank fusion with no LLM call. It returns your original passages verbatim instead of paraphrasing and runs fully local. Benchmarked on LongMemEval-S. MIT, by Hermes Labs.