OKP vs mem0 / claude-mem / basicmemory: knowledge, not memory

[markbang]

OKP vs mem0 / claude-mem / basicmemory: knowledge, not memory

"Memory" tools have exploded over the last year — mem0, claude-mem, basicmemory, Zep, Letta and friends. They solve a real problem: agents forget. But OKP deliberately takes a different position: agents don't need more memory, they need better context.

The distinction

	Memory tools	OKP
Unit	Session recall, snippets	Domain-shaped knowledge
Structure	Flat or timeline	README + schema + typed concepts + links
Visibility	Private by default	Public by default (opt-out)
Writing	One agent, one user	Shared writing with invite codes, one accountable host
Integration	SDK / MCP server	Skills (okp-search, okp-import)
Protocol	Proprietary per tool	Open, self-hostable, MIT

When to use which

• Use mem0 / claude-mem / basicmemory when you want your coding agent to remember your preferences across sessions. They do that well.
• Use OKP when knowledge is a shared asset: a domain README that everyone (and every agent) reads, concepts with provenance, links between them.

Why "memory" is not the endgame

Memory is per-agent state; context is shared understanding. The AI-tool sites growing fastest right now are winning on context — up-to-date, structured, verifiable knowledge — not on remembering yesterday's conversation.

OKP is for the moment you realize your agent keeps asking questions your team already answered, and that the answer belongs in an open, structured, searchable place — not in one agent's private memory.

Try it

npm install -g @markbangwu/okp
npx skills add https://github.com/talesofai/okp --skill okp-search --agent codex

Disagree with the memory/context framing? That is exactly what this thread is for.