src/pqc_rag_signing/adapters/base.py
| 1 | """Abstract vector store adapter interface.""" |
| 2 | |
| 3 | from __future__ import annotations |
| 4 | |
| 5 | from abc import ABC, abstractmethod |
| 6 | from typing import Iterable |
| 7 | |
| 8 | from pqc_rag_signing.chunk import SignedChunk |
| 9 | |
| 10 | |
| 11 | class VectorStoreAdapter(ABC): |
| 12 | """Base class for vector DB integrations. |
| 13 | |
| 14 | Implementations must preserve the full SignedChunk envelope (not just |
| 15 | text + embedding) so verification is possible at retrieval time. |
| 16 | Most vector DBs support arbitrary metadata - store ``chunk.to_dict()`` |
| 17 | as metadata alongside the embedding. |
| 18 | """ |
| 19 | |
| 20 | @abstractmethod |
| 21 | def upsert( |
| 22 | self, |
| 23 | chunks: Iterable[SignedChunk], |
| 24 | embeddings: list[list[float]], |
| 25 | ) -> None: |
| 26 | """Store signed chunks with their embeddings.""" |
| 27 | |
| 28 | @abstractmethod |
| 29 | def query(self, embedding: list[float], top_k: int = 5) -> list[SignedChunk]: |
| 30 | """Retrieve top-k chunks for a query embedding. |
| 31 | |
| 32 | Returns SignedChunks with full envelope intact so verification works |
| 33 | downstream. |
| 34 | """ |
| 35 | |
| 36 | @abstractmethod |
| 37 | def count(self) -> int: |
| 38 | """Return total number of chunks stored.""" |
| 39 | |