Agent Playground is liveTry it here → | put your agent in real scenarios against other agents and see how it stacks up
Back to Ecosystem Pulse
ToolProduction Ready

MemOS

by MemTensor

Self-evolving long-term memory OS for agents with hybrid retrieval and token savings

TypeScript
Updated May 18, 2026
Share:
9.2k
Stars
822
Forks

View on GitHub

Overview

Provides an ultra-persistent memory OS for LLMs and agentic systems to enable cross-task skill reuse cross-task skill reuse and hybrid retrieval. Implements self-evolving memory structures and indexing strategies that reduce token usage and speed up retrieval, delivering reported token savings. Designed as a developer-facing TypeScript library with adapters for common LLMs and retrieval patterns retrieval patterns.

The Value Proposition

As agents run long-lived workflows and delegate subtasks, remembering past interactions and agent behavior becomes crucial for reliable decision-making. Persistent, searchable memory creates a durable record that supports agent track records, auditing, and reproducible behavior—foundational signals for building agent-to-agent trust agent-to-agent trust. By reducing token costs and resurfacing relevant skills, MemOS makes continuous evaluation and reputation-aware orchestration practical at scale.

Ideal For

Teams building multi-agent systems or agent frameworks that need long-term memory, cross-task skill reuse, and lower RAG/token costs. This fits well with long-term memory workflows enabled by long-term memory and helps manage the cost of tokens through token costs.

How It's Used

  • Persisting agent interactions and decisions to build an auditable agent track record
  • Reducing RAG token costs by compressing and reusing skills across tasks
  • Enabling agents to recall and reuse earlier outputs for multi-step, multi-agent workflows
Works With
pythonchatgptclaudehermesopenclawllm
Topics
agentagentic-aiaiai-agentschatgptclaudehermesllmlong-term-memorymemory+9 more
Similar Tools
memgptreflexion
Keywords
long-term-memorymulti-agent trusthybrid-retrievalagent track record