Skip to content
KiomonDocs

Search

Kiomon search is hybrid: it combines an exact keyword index with semantic matching, so you find things by phrase and by meaning.

How it works

Section titled “How it works”

Search runs across two complementary legs and their results are fused into one ranked list:

Leg Strengths
Keyword Exact phrases, technical terms, code identifiers, quotes
Semantic Meaning and paraphrase — matching even when your wording differs from the saved text

Documents that match strongly in both legs rank highest; something only the semantic leg understands still surfaces. If the semantic leg isn’t available for a deployment, search automatically degrades to keyword-only — you always get results, never a hard failure.

Search behavior

Section titled “Search behavior”
  • Clean queries — searches are scrubbed server-side; malformed input can’t break the index
  • AND-first, OR fallback — matches on all your terms first; if nothing matches, the query loosens
  • Prefix matching — partial terms like stripe* work for technical content
  • Snippets — results come back with clean snippets of the matching passage
  • Kind filter — narrow to a memory kind (semantic, procedural, episodic, reference) — e.g. search only Facts or only Routines
  • Workspace scoping — restrict a search to a specific workspace

Where search appears

Section titled “Where search appears”
  1. Dashboard library — type-ahead, live search over the active workspace
  2. MCP search_memory — how your agents search your brain (hybrid, with kind filters)
  3. REST GET /api/search?q=... — scripted access for any tool

Index freshness

Section titled “Index freshness”

The search index stays in sync with your library automatically:

  • New saves and approved memories are searchable immediately
  • Archived items are hidden from results
  • Deleted or forgotten items are removed from search entirely

No manual indexing, no rebuilds.

Rate limits

Section titled “Rate limits”

Search is rate-limited per plan with a per-day quota on each API key; exceeding it returns a clear message. Pro gets a 25× higher quota — see Plans & Quotas.

Tips

Section titled “Tips”
  • Search reads naturally: both chat webhook docs and how do I receive webhooks find the same content (that’s the semantic leg at work).
  • For code or library names, rely on keyword matching — exact tokens win.
  • Agents: always pass kind when you know what you’re looking for, and prefer get_workspace_context(topic) for open-ended orienting over a bare search.