Compare the leading AI agent memory systems: Mem0, LangMem, Memori, and agent-memory. LoCoMo benchmark, open source, encryption, and pricing.
Vectorize.io evaluated 8 frameworks on the LoCoMo and LongMemEval benchmarks — the de facto standard evaluations for AI agent memory systems. Here's what we know:
"LoCoMo and LongMemEval have become the de facto standard evaluations for AI agent memory systems." — Vectorize.io, 3 weeks ago
66.9% on LoCoMo (self-reported). Open core. Cloud-first with self-host option.
SDK for building memory into LLM applications. Open source. From the LangChain ecosystem.
Agent-native SQL-native memory infrastructure. Evaluated on LoCoMo. 3 days ago.
MCP v3.2 native. AES-256 + TTL. MIT license. Works with any MCP agent.
| Feature | Mem0 | LangMem | Memori | agent-memory |
|---|---|---|---|---|
| ★ agent-memory | — | — | — | MCP v3.2 |
| LoCoMo Benchmark | 66.9% (self) | — | Evaluated | — |
| License | Open core | Open source | — | MIT |
| Encryption | — | — | — | AES-256 |
| TTL | — | — | — | Yes |
| Cloud Required | Yes | No | No | Zero |
| Storage | Managed | Self-hosted | SQL-native | JSON/SQLite/Redis |
LangMem from the LangChain ecosystem provides an SDK for building memory into LLM applications:
"LangMem: a software development kit (SDK) from the LangChain ecosystem for building persistent memory into LLM applications." — Awesome-GraphMemory on GitHub, February 5, 2026
Memori is a new entry — agent-native memory infrastructure that turns agent execution into structured, persistent state:
"Memori is agent-native memory infrastructure. A SQL-native, LLM-agnostic layer that turns agent execution and interactions into structured, persistent state for production systems. Evaluated on the LoCoMo benchmark." — MemoriLabs/Memori on GitHub, 3 days ago
"This is a no-BS comparison of the four main options in 2026: Mem0, Zep, LangMem, and MemoClaw." — DEV Community, February 15, 2026
For developers building with MCP-compatible agents (Claude Code, Cursor, Cline, Windsurf, OpenClaw, OpenHands), agent-memory is the only option that gives you:
# Install agent-memory
pip install agent-memory
# Run the MCP server
python -m agent_memory.mcp_server --storage json --path ./memory.json
# Connect to any MCP agent (Claude Code, Cursor, Cline, Windsurf, OpenClaw...)
# Works with all MCP-compatible tools automatically