Skip to content

GEMINI.md

Latest commit

 

History

History
26 lines (20 loc) · 1.22 KB

GEMINI.md

File metadata and controls

26 lines (20 loc) · 1.22 KB

Tana Semantic Search Guidelines (Notion AI Style)

Infrastructure

  • Vector Store: LanceDB (~/TanaEmbeddings/vector_store)
  • Embedding Model: models/gemini-embedding-001 (Google AI)
  • Sync Schedule: Hourly via launchd (com.kavia.tanaembeddings.sync)

Search Strategy (The Hybrid Approach)

When asked to "Search Tana" or "Find information":

  1. Step 1: Semantic Lookup

    • Run python3 ~/TanaEmbeddings/search_tana.py "[query]"
    • This finds conceptual matches, even if words don't match exactly.

  2. Step 2: Native Keyword Lookup (Fallback/Augment)

    • Simultaneously use tana.search_nodes with textContains.
    • This is essential for specific names, dates, or technical IDs.

  3. Step 3: Arbitration

    • If Semantic search returns highly relevant nodes, prioritize them.
    • If the user query is very specific (e.g., "Note from March 7"), prioritize Native search.
    • Combine both into a single, clean response. Do not show "Technical IDs" unless asked.

Energy & Credit Efficiency

  • Polling: The background sync is hourly. Do not trigger manual syncs unless the user explicitly says "I just added something, find it now."
  • Model: Stick to Google AI to stay within the user's Pro quota.