Hive

Hive is the knowledge layer behind everything I run.

When something is worth keeping — a decision, an SOP, a campaign that worked, a bug we solved, a constraint we hit — it goes into Hive as a structured note tagged by domain. It's not free-floating Notion docs. Each entry has shape.

The agents that operate my businesses query Hive before they act. They never start from amnesia. Today's revenue, last week's decisions, every SOP filed, the last seven end-of-day logs — all of it is searchable in every conversation.

The bet is simple: an AI is only as good as the memory you give it. Hive is the memory.

A Postgres-backed knowledge store inside HPFHQ with semantic embeddings, an 8-domain taxonomy, context_type classification, sensitivity flags, and access metrics (last_retrieved_at, retrieval_count).

Notes are hand-curated by operators and dream-extracted from session transcripts. They live as rows but render as markdown for read paths. Every retrieval fires a record_memory_hits() RPC that stamps access time and increments count — feeds back into recency scoring.

Hive is the substrate. Pathfinder is the read API. Nova is one of its consumers. The Claude Code memory hook is another. Every active agent in HPFHQ resolves through Pathfinder against Hive.