Back to Gap Map Product / UX / Agent memory & context stores

MemOS

MemTensor
open source / Overall score: 3.2

MemOS is a memory operating system for LLM agents that extracts structured memories from conversations, documents and tool traces, organizes them as graph-based textual, preference and activation memory in composable memory cubes, and schedules updates, merging and forgetting in the background. It runs as a self-hosted REST service or as a local memory plugin for OpenClaw, Hermes Agent and DeepSeek Harness. MemTensor builds it.

Openness

5 high confidence
5.0
license
Apache-2.0 (the MemOS service, scheduler and memory modules)+MIT (the memos-local-plugin agent memory core)
source
public
core features withheld
no — the 2,061-path tree at 3a16e5f holds the Python service, scheduler, memory modules, Docker and Helm deployment and the local plugin with no ee/, enterprise/, commercial/ or proprietary/ path; the one cloud plugin in it is a client of MemOS Cloud
commercial
MemOS Cloud and MemMate Enterprise

MemOS's memory service, scheduler and local agent plugin are published in full, and the service self-hosts with Docker beside its graph and vector stores. MemTensor sells MemOS Cloud and an enterprise agent platform as separate hosted products rather than features withheld from the code.

Adoption

2 low confidence
2.0

Adoption is read from monthly downloads of MemoryOS on PyPI and @memtensor/memos-local-plugin on npm, the two packages that install MemOS itself, most of them for the local plugin. Docker deployments, MemOS Cloud and the cloud plugin pass through neither package and are not counted.

Capability

5 medium confidence
5.0

MemOS performs every part of the cycle itself: it extracts memories from raw input, merges and corrects them in the background, forgets and decays them over time, and keeps several typed memories in a graph.

Verified 2026-11-20, except some scores, still awaiting re-confirmation