MemOS vs OpenViking
MemTensor's memory OS for LLM agents: one API over graph-structured, multi-modal memory with hybrid retrieval and skill evolution — hosted, self-hosted (Neo4j + Qdrant) or local plugins. — versus — Volcengine's context database: memories, resources and skills as one `viking://` filesystem agents ls, tree and grep — L0/L1/L2 tiers, traceable retrieval, sessions distilled into memory.
Both unify memories, resources and skills for agents and distill sessions into long-term memory. OpenViking exposes it as a viking:// filesystem; MemOS as a graph-structured memory API.
| MemOS | OpenViking | |
|---|---|---|
| Stars | 12k | 39k |
| Forks | 1.1k | 3.0k |
| Language | TypeScript | Python |
| License | Apache-2.0 | AGPL-3.0 |
| Last activity | 5 days ago | 4 days ago |
| Topics | memory, skills | memory, rag, skills |
| Curated connections | 5 | 8 |
MemOS — the curator's take
Pick MemOS when you want a full memory stack rather than a vector wrapper: graph-structured memories you can inspect and correct in natural language, isolated or shared 'memory cubes' across users and agents, async ingestion, and traces that crystallize into reusable skills. The fastest path is a plugin — local SQLite for Hermes, OpenClaw or DeepSeek Harness, zero infra. Self-hosting the service means running Neo4j and Qdrant plus LLM and embedder keys, which is heavy for a single-user bot. Read the headline LoCoMo/LongMemEval numbers with care: the comparison runs on OmniMemEval, MemTensor's own harness. Skip it if you only need session recall for one coding agent — claude-mem or engrim is far less machinery.
OpenViking — the curator's take
Use it when opaque vector recall has burned you: every query keeps the directory-browsing trajectory that produced it, so a wrong answer is debuggable, and the L0/L1/L2 tiers let an agent judge relevance before paying for full content. It pays off most for long-lived agents with heavy reference material — repos, docs, per-user preferences — and there are documented hooks for Claude Code, Codex, Cursor, OpenCode, LangChain/LangGraph and plain MCP clients. Costs: it's a server plus semantic pre-processing on every write, so ingest is slow and not free; AGPLv3 rules it out of many closed products; and there's an obvious managed-SaaS path on Volcano Engine behind it. Wrong tool if you just want an in-process vector index.