// retrieval engine

Pathfinder

The router between the knowledge layer and everything that needs it. Domain-aware scoring, semantic plus keyword, sensitivity-filtered.

// for laypeople

Pathfinder is the search engine for Hive.

When something needs information from the knowledge layer — an agent making a decision, a Claude session looking for context, a Nova run prepping for a meeting — it asks Pathfinder, not Hive directly.

Pathfinder figures out what domain the question belongs to, ranks the most relevant notes, filters out anything inappropriate for the asker, and returns a clean answer.

It's the difference between dumping a database into a prompt and giving the AI exactly what it needs.

// for builders

GET /api/knowledge/pathfinder?domain={domain}&search={keywords} — a Next.js route in HPFHQ.

Composite scoring across recency, retrieval count, semantic similarity (when ?semantic=true), and keyword match. Domain routing supports exclude_domains for cross-domain de-noising; the Claude Code hook auto-detects task domain from branch prefix and keyword signals, then passes the exclusion list automatically.

Sensitivity filtering is enforced at the query level — sensitive entries never leak to read paths that aren't authorized. Returns the standard Pathfinder response shape: scored hits with file paths, snippets, and metadata.

Fire-and-forget hit instrumentation stamps last_retrieved_at and increments retrieval_count via record_memory_hits() after every read.