A self-hosted memory and knowledge layer for AI agents and humans. Carry project context across sessions, tools, and teams. Keep it in a wiki you can read, edit, and own.
Start in OpenClaw. Pick up in Codex. Keep the context.
One knowledge base
Scoped access
Readable by humans
01 / MEMORY THAT CARRIES FORWARD
Less starting over. More building on what you know.
Decisions, discoveries, and the details that make a project yours. Give them a home that your next agent can find.
Recall the words. Retrieve the meaning.
Search with PostgreSQL full-text retrieval out of the box. Add optional pgvector embeddings to find semantically related knowledge, even when the wording changes.
The memory layer ranks results, removes duplicates, handles conflicts, and fits useful context into your token budget.
A shared article store, a provider-aware recall layer, and integrations that bring knowledge into the tools you already use.
01
Your tools
Agent plugins, Codex / MCP, REST clients, and the human wiki.
CONNECT
02
Memory layer
Recall across providers. Rank, deduplicate, resolve conflicts, and budget context.
ORCHESTRATE
03
Hybrid retrieval
Full-text search Default
pgvector + RRF Opt-in
FIND
04
Durable knowledge
PostgreSQL holds your articles and metadata. Redis supports fast recall caching.
KEEP
Access stays scoped. Permissions apply to the knowledge each agent can retrieve.
Providers stay extensible. Recall supports Noosphere articles and Hindsight.
Vector retrieval combines lexical and semantic rankings with reciprocal rank fusion. It requires explicit activation, an embedding provider, and backfill. Keyword search remains the default.