Repository navigation
[P1] Intent vector store + similarity (ChromaDB) or amend proposal #11
Description
Problem
Proposal §3.2 / §5.2 specify a ChromaDB vector intent DB with similarity search. Implementation (storage/intent_db.py) is plain SQLite — no embeddings, no similarity.
Tasks (pick one direction)
A. Implement: add embedding generation + ChromaDB (or sqlite-vec) store; enable intent reuse via similarity lookup.
B. Amend: if vector search isn't needed for MVP, update proposal.md to reflect SQLite design and remove ChromaDB claims.
- Decide A vs B and note rationale in this issue.
- Implement chosen path.
Acceptance
Docs and code agree on the storage design.
Activity
pro-utkarshM commentedon Oct 15, 2026
Decision: option B, keep SQLite.
Rationale: nothing in the pipeline needs similarity search. Intents are looked up by component ID, reused across runs, and an edited intent is never overwritten. A vector store would add an embedding model and a dependency without a current use. If cross-component intent reuse becomes a feature, this can be revisited.
The current docs describe the SQLite design:
docs/architecture.md(storage module),docs/task.mdand the README's file table (.code2test/code2test.db).docs/proposal.mdopens with a note that it is the original January plan and differs from v1. Its ChromaDB sections are kept as the historical plan.Generated by Claude Code