EverOS vs OpenViking
One portable memory layer for every agent: conversations, files and trajectories kept as canonical Markdown, indexed locally by SQLite and LanceDB, with offline reflection that refines them. — 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 memory, files and skills into one human-readable store rather than an opaque index. OpenViking runs a `viking://` filesystem with L0/L1/L2 tiers and traceable retrieval; EverOS keeps plain Markdown as truth with local indexes and offline reflection. Filesystem metaphor versus editable files.
| EverOS | OpenViking | |
|---|---|---|
| Stars | 12k | 32k |
| Forks | 901 | 2.4k |
| Language | Python | Python |
| License | Apache-2.0 | AGPL-3.0 |
| Last activity | 6 days ago | yesterday |
| Topics | memory, skills | memory, rag, skills |
| Curated connections | 4 | 5 |
EverOS — the curator's take
Pick it if you want memory you can open in an editor: `.md` files are the source of truth, edited directly and picked up by a cascade watcher, with SQLite plus LanceDB as disposable indexes and no MongoDB, Elasticsearch or Redis in the stack. User episodes and profile sit apart from agent cases and skills as first-class tracks, retrieval scopes orthogonally by user, agent, app, project and session, and reflection consolidates episode clusters between sessions instead of only recalling. Trade-offs: it runs as a server with its own TOML config, full capability needs embedding and rerank models configured beyond the one-key start, Markdown-as-truth means large vaults get large, and `/api/v1` is already a legacy alias.
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.