A self-hostable 'memory operating system' that packages long-term memory into MemCube units and manages their lifecycle (store / retrieve / update / schedule) outside the model. v2.0 adds multi-cube knowledge base management, multi-modal memory, tool-use history, async ingestion via MemScheduler, and natural-language feedback and correction.
- Storage
- Pluggable backends; the local plugin uses persistent SQLite. Supports text, images, tool traces, and personas as memory. Graph memory via Neo4j and vector storage via Qdrant for the full self-hosted deployment.
- Retrieval
- Hybrid retrieval combining full-text search (FTS5) and vector similarity, with task summarization and cross-task skill reuse. Search pipeline hooks added in v2.0.20 for context rendering.
- Self-host
- Self-host: moderate
- License
- Apache-2.0
- Pricing
- Open source (Apache-2.0), free to self-host. Hosted cloud tier at memos.openmem.net/pricing: Free $0/mo (50K memory adds, 20K searches/mo, 10 knowledge bases); Starter $0/mo promo (list price $19; 600K adds, 200K searches/mo, 30 KBs); Pro $0/mo promo (list price $286; 80M adds, 30M searches/mo, 100 KBs); Enterprise custom/unlimited with private deployment. · Freemium
- GitHub stars
- 10,664
- Last release
- 2026-07-31
- Last commit
- 2026-08-07
- First catalogued
- 2026-06-28
Strengths
- Active, high-traction project (10k+ stars)
- Hybrid full-text + vector retrieval
- Multi-modal memory (text, images, tool traces, personas)
- Vendor reports ~35% token savings and +43.7% accuracy vs OpenAI Memory
- Both self-hosted (Apache-2.0) and managed cloud (OpenMem) paths available
- MCP, REST API, and Python SDK all ship in the same repo
Watch out
- 'Memory OS' framing overlaps confusingly with the separate BAI-LAB MemoryOS project — they are different systems
- Hosted cloud pricing page now lists Starter/Pro at $0/mo as a promotional discount off list prices ($19/$286) — confirm current promo status before quoting a rate to readers
- Full self-hosted stack requires Neo4j + Qdrant + Redis; lightweight mode uses SQLite only but has reduced features
Best for
- Teams wanting a self-hosted memory layer with hybrid retrieval and skill reuse
- Projects needing an MCP-accessible memory server that can run locally or in the cloud
How it integrates
Benchmark results
| Benchmark | Value | Backbone | Trust | Source |
|---|---|---|---|---|
| locomo | 75.8 accuracy | GPT-4o-mini | Self-reported | MemOS (MemTensor et al.) ↗ |
| longmemeval | 77.8 accuracy | GPT-4o-mini | Self-reported | MemOS (MemTensor et al.) ↗ |
Sources
- Pricing page re-checked — unchanged: Free $0/mo (50K adds, 20K searches, 3M input / 1M output Chat API tokens); Starter $0/mo promo against a $19 list price (600K adds, 200K searches); Pro $0/mo promo against a $286 list price (80M adds, 30M searches); Enterprise custom with private deployment (vendor)
- Monthly card re-verification — GitHub API repo metadata: 10,534 stars, license Apache-2.0 confirmed, archived=false, last commit on the default branch 2026-07-31 (third-party)
- https://github.com/MemTensor/MemOS (vendor)
- MemOS: A Memory OS for AI System (paper)
- GitHub API — stars, pushed_at, license, archived (vendor)
- MemoryOS on PyPI — version 2.0.20 (vendor)
- mcp_serve.py — FastMCP stdio + HTTP transport (vendor)
- memos-local-plugin for OpenClaw and Hermes (vendor)
- MemOS REST API server setup docs (vendor)
- MemOS Cloud pricing page — Free/Starter/Pro/Enterprise tiers; Starter and Pro shown at $0/mo promo (list price $19/$286) (vendor)
- GitHub API monthly re-check — license Apache-2.0 confirmed, not archived (third-party)
Last verified 2026-08-01 · updated by refresh-framework-cards