Hybrid SQLite Storage & Local Vector Search

How Dori combines relational database models with local semantic search.

Dori bridges the gap between structured relational data (projects, tasks, meetings) and unstructured knowledge (notes, transcripts, documents).

Storage & Search Architecture

1. Embedded Relational SQLite DB

Relational entities, workflow states, and system projections live in an embedded SQLite database running locally within the desktop process space:

• Low latency queries for UI rendering.
• Foreign key constraints and transactional integrity.

2. SQLite FTS5 + Local Vector Search

Searching personal data requires precision for exact names as well as recall for conceptual topics:

• BM25 Exact Match (FTS5): Instant keyword lookup for specific names, tags, and titles.
• Semantic Vector Indexing: Computes vector embeddings locally to match related concepts even when keywords differ.
• Hybrid Ranker: Combines keyword score and vector similarity to return highly relevant results.

3. Content Fingerprinting

Files are tracked via SHA-256 content hashes. Idle editing overhead is minimal because unchanged files bypass re-indexing and vector re-embedding.