@alexkroman1/aai 0.10.2 → 0.10.3

package/dist/sqlite-vector.js

@@ -1,149 +0,0 @@
-import { DatabaseSync } from "node:sqlite";
-//#region sqlite-vector.ts
-/**
-* SQLite-backed vector store with local embeddings.
-*
-* Persists data across restarts using a local SQLite database file.
-* Uses brute-force cosine similarity over `node:sqlite` — no native
-* extensions required. Fast enough for local dev (sub-ms for <10k vectors).
-* Embeddings are computed locally via `all-MiniLM-L6-v2` (384 dims) —
-* no external API key required. The model is downloaded on first use
-* (~86 MB) and cached in `.aai/models/`.
-*/
-const DEFAULT_DIMENSIONS = 384;
-const DEFAULT_MODEL = "Xenova/all-MiniLM-L6-v2";
-/**
-* Create a local embedding function using `all-MiniLM-L6-v2`.
-*
-* The model is downloaded on first use (~86 MB) and cached locally.
-* Subsequent calls load from cache in ~90ms. Each embedding takes <2ms.
-*/
-function createLocalEmbedFn(cacheDir) {
-	let pipelinePromise = null;
-	async function getPipeline() {
-		if (!pipelinePromise) pipelinePromise = (async () => {
-			const { pipeline, env } = await import("@huggingface/transformers");
-			env.cacheDir = cacheDir;
-			return pipeline("feature-extraction", DEFAULT_MODEL);
-		})();
-		return pipelinePromise;
-	}
-	return async (text) => {
-		const output = await (await getPipeline())(text, {
-			pooling: "mean",
-			normalize: true
-		});
-		return Array.from(output.data);
-	};
-}
-/**
-* Create a deterministic hash-based embedding function for testing.
-*
-* Produces repeatable vectors where similar text yields similar embeddings.
-* Not suitable for production — use the default local model instead.
-*
-* @param dimensions - Vector dimensions (default: 384).
-*/
-function createTestEmbedFn(dimensions = DEFAULT_DIMENSIONS) {
-	return async (text) => {
-		const vec = new Float32Array(dimensions);
-		const words = text.toLowerCase().split(/\s+/).filter(Boolean);
-		for (const word of words) {
-			let hash = 0;
-			for (let i = 0; i < word.length; i++) hash = (hash << 5) - hash + word.charCodeAt(i) | 0;
-			for (let i = 0; i < 8; i++) {
-				const idx = Math.abs((hash + i * 31337) % dimensions);
-				vec[idx] = (vec[idx] ?? 0) + 1;
-			}
-		}
-		let norm = 0;
-		for (let i = 0; i < dimensions; i++) norm += (vec[i] ?? 0) * (vec[i] ?? 0);
-		norm = Math.sqrt(norm) || 1;
-		return Array.from(vec, (v) => v / norm);
-	};
-}
-/** Cosine similarity between two vectors stored as Buffers of float32. */
-function cosineSimilarity(a, b) {
-	const fa = new Float32Array(a.buffer, a.byteOffset, a.byteLength / 4);
-	const fb = new Float32Array(b.buffer, b.byteOffset, b.byteLength / 4);
-	let dot = 0;
-	let normA = 0;
-	let normB = 0;
-	for (let i = 0; i < fa.length; i++) {
-		const ai = fa[i] ?? 0;
-		const bi = fb[i] ?? 0;
-		dot += ai * bi;
-		normA += ai * ai;
-		normB += bi * bi;
-	}
-	const denom = Math.sqrt(normA) * Math.sqrt(normB);
-	return denom === 0 ? 0 : dot / denom;
-}
-/**
-* Create a SQLite-backed vector store with local embeddings.
-*
-* Data persists to a local SQLite file (default: `.aai/vectors.db`).
-* Embeddings are computed locally using `all-MiniLM-L6-v2` by default —
-* no API key required. The model auto-downloads on first use (~86 MB).
-*
-* Vector search uses brute-force cosine similarity over all stored
-* embeddings. This is fast for local dev workloads (<10k vectors).
-*
-* @param options - See {@link SqliteVecVectorStoreOptions}.
-* @returns A {@link VectorStore} instance.
-*
-* @example
-* ```ts
-* import { createSqliteVectorStore } from "@alexkroman1/aai/sqlite-vector";
-*
-* const vector = createSqliteVectorStore();
-* await vector.upsert("doc-1", "The capital of France is Paris.");
-* const results = await vector.query("France capital");
-* ```
-*/
-function createSqliteVectorStore(options) {
-	const dbPath = options?.path ?? ".aai/vectors.db";
-	const cacheDir = options?.modelCacheDir ?? ".aai/models";
-	const embedFn = options?.embedFn ?? createLocalEmbedFn(cacheDir);
-	const db = new DatabaseSync(dbPath);
-	db.exec("PRAGMA journal_mode = WAL");
-	db.exec(`
-    CREATE TABLE IF NOT EXISTS vectors (
-      id TEXT PRIMARY KEY,
-      data TEXT NOT NULL,
-      metadata TEXT NOT NULL DEFAULT '',
-      embedding BLOB NOT NULL
-    )
-  `);
-	const stmtUpsert = db.prepare("INSERT OR REPLACE INTO vectors (id, data, metadata, embedding) VALUES (?, ?, ?, ?)");
-	const stmtDelete = db.prepare("DELETE FROM vectors WHERE id = ?");
-	const stmtAll = db.prepare("SELECT id, data, metadata, embedding FROM vectors");
-	return {
-		async upsert(id, data, metadata) {
-			const vector = await embedFn(data);
-			const embedding = Buffer.from(new Float32Array(vector).buffer);
-			const metaJson = metadata ? JSON.stringify(metadata) : "";
-			stmtUpsert.run(id, data, metaJson, embedding);
-		},
-		async query(text, queryOptions) {
-			const topK = queryOptions?.topK ?? 10;
-			if (!text.trim()) return [];
-			const queryVec = await embedFn(text);
-			const queryBuf = Buffer.from(new Float32Array(queryVec).buffer);
-			const scored = stmtAll.all().map((row) => ({
-				id: row.id,
-				score: cosineSimilarity(queryBuf, row.embedding),
-				data: row.data,
-				metadata: row.metadata && row.metadata !== "" ? JSON.parse(row.metadata) : void 0
-			}));
-			scored.sort((a, b) => b.score - a.score);
-			return scored.slice(0, topK);
-		},
-		async delete(ids) {
-			const idArray = Array.isArray(ids) ? ids : [ids];
-			for (const id of idArray) stmtDelete.run(id);
-		}
-	};
-}
-//#endregion
-export { createSqliteVectorStore, createTestEmbedFn };