Supermemory iconSupermemory

open-source Free Star27k

Memory and context engine for AI agents — #1 on LongMemEval, LoCoMo, and ConvoMem benchmarks with 99.4% context reduction

22K GitHub Stars
#1 LongMemEval
99.4% Context Reduction

Overview

Supermemory is the memory and context layer for AI, ranking #1 on LongMemEval, LoCoMo, and ConvoMem — the three major benchmarks for AI memory. Unlike RAG which retrieves static document chunks, Supermemory extracts and tracks facts about users over time, understanding that "I just moved to SF" supersedes "I live in NYC." It achieves 95% Recall@15 while adding only ~720 tokens — a 99.4% context reduction. The platform can run fully locally with one binary and zero config, supporting any model including offline operation with Ollama.

The Verdict

Who Should Use Supermemory?

Best For

  • AI agents needing persistent user memory across sessions
  • Developers wanting state-of-the-art memory benchmarks
  • Teams preferring self-hosted, open-source solutions
  • Claude Code, OpenCode, and OpenClaw users
  • Privacy-conscious deployments with local-only operation

Not Ideal For

  • Simple RAG-only use cases
  • Teams wanting fully managed cloud service
  • Non-conversational document retrieval
  • Extremely low-latency requirements

What's Great

  • #1 on LongMemEval (81.6% recall vs 71.2% next-best)
  • 99.4% context reduction with high recall
  • Fully local operation — one binary, zero config
  • Open-source with active community
  • Handles knowledge updates and contradictions automatically
  • Plugins for Claude Code, OpenCode, OpenClaw, Hermes
  • Works with any model including Ollama for offline

Watch Out For

  • Requires understanding memory vs RAG distinction
  • Self-hosted setup needed for full control
  • Newer project — ecosystem still growing
  • May require tuning for specialized domains

Pricing

View all features & details

Memory Capabilities

  • Automatic fact extraction from conversations
  • User profile building over time
  • Knowledge update and contradiction handling
  • Expired information forgetting
  • Context-aware retrieval

Benchmarks

  • LongMemEval: 81.6% (vs 71.2% Zep)
  • LoCoMo: #1 ranking
  • ConvoMem: #1 ranking
  • 95% Recall@15 with ~720 tokens

Integrations

  • Claude Code plugin
  • OpenCode plugin
  • OpenClaw plugin
  • Hermes plugin
  • Ollama support (offline)

Deployment

  • Single binary installation
  • Zero configuration required
  • Bring any model
  • Fully offline capable

How It Compares

Feature Supermemory Mem0 Zep Letta
LongMemEval 81.6% 71.2%
Open Source Yes Yes Partial Yes
Local Deploy One binary Docker Docker Docker
Memory vs RAG Both Memory Memory Memory
Context Reduction 99.4% Good Good Good
IDE Plugins CC, OC, Hermes Limited

User Reviews

Loading reviews...