Implement VectorStore (Local Vector Database)

[ustyuzhaninky]

Implement VectorStore (Local Vector Database)

Labels: stage-8, backend, retrieval, storage

Description

Create the foundational VectorStore component that stores embeddings, associated text chunks, and metadata.
This acts as the backend for all retrieval operations.

Requirements

• Choose initial storage backend:
  • Option A (simple): JSON file-based store
  • Option B (recommended): SQLite table with:
    • id
    • text
    • embedding vector (serialized)
    • metadata (JSON)
    • created_at
• Implement methods:
  • add(text, embedding, metadata={})
  • get(id)
  • all()
  • delete(id)
• Ensure the store persists across restarts
• Handle concurrent access safely

Acceptance Criteria

• VectorStore implemented with persistent storage
• Add/read/delete operations working and tested
• Metadata support included (filename, tags, etc.)
• Correct error handling for invalid records
• Unit tests for all CRUD methods

References

Parent: Stage 8 — Simple Retrieval Queries