@fenecdb/langchain

A LangChain.js vector store over fenecdb: the database in the page (a Fenec from @fenecdb/web), or behind fenec-server's HTTP endpoint (a FenecHttp) -- anything with run(sql, params).

import { Fenec } from '@fenecdb/web';
import { FenecVectorStore } from '@fenecdb/langchain';

const db = await Fenec.open('./fenec.wasm');
const store = new FenecVectorStore(embeddings, {
  client: db,
  collection: 'handbook',
  metadataFields: { source: 'text' },
  fullText: true,
});
await store.addDocuments(documents);
await store.similaritySearch('how do I compact', 4, { source: 'handbook' });
await store.search('compact', 4, { mode: 'hybrid' });

It is the Python store's collection (pip install "fenecdb[langchain]"), from JavaScript:

• A document is a row: its id, text, metadata as JSON and vector under @hnsw, over int8 or bit codes with quant. The collection is made on the first write, once the embedding's size is known.
• A field named in metadataFields gets a column of its own under @hash. filter takes equality and $in over those, answered by the index before the search ranks what passed.
• With fullText, the text is indexed for BM25 as well. search(query, k, {mode}) ranks by the words ('text') or by the words and the vector fused ('hybrid').
• Writing an id that is already there replaces it. On a Fenec both statements run as one block; over HTTP they are two requests.

npm test holds it to what LangChain.js's own vector store integrations are tested for, over a database in the page and over a fenec-server. It needs web/fenec.wasm and the fenec-server binary.