@agentuity/runtime 2.0.11 → 3.0.0-alpha.1

package/src/services/local/vector.ts

@@ -1,438 +0,0 @@
-import type { Database } from 'bun:sqlite';
-import type {
-	VectorStorage,
-	VectorUpsertParams,
-	VectorUpsertResult,
-	VectorResult,
-	VectorResultNotFound,
-	VectorSearchResultWithDocument,
-	VectorSearchParams,
-	VectorSearchResult,
-	VectorNamespaceStats,
-	VectorNamespaceStatsWithSamples,
-	VectorGetAllStatsParams,
-	VectorStatsPaginated,
-} from '@agentuity/core';
-import { randomUUID } from 'node:crypto';
-import { simpleEmbedding, cosineSimilarity, now } from './_util';
-
-export class LocalVectorStorage implements VectorStorage {
-	#db: Database;
-	#projectPath: string;
-
-	constructor(db: Database, projectPath: string) {
-		this.#db = db;
-		this.#projectPath = projectPath;
-	}
-
-	async upsert(name: string, ...documents: VectorUpsertParams[]): Promise<VectorUpsertResult[]> {
-		if (!name?.trim()) {
-			throw new Error('Vector storage name is required');
-		}
-		if (documents.length === 0) {
-			throw new Error('At least one document is required');
-		}
-
-		const results: VectorUpsertResult[] = [];
-		const stmt = this.#db.prepare(`
-			INSERT INTO vector_storage (
-				project_path, name, id, key, embedding, document, metadata, created_at, updated_at
-			)
-			VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
-			ON CONFLICT(project_path, name, key) 
-			DO UPDATE SET 
-				embedding = excluded.embedding,
-				document = excluded.document,
-				metadata = excluded.metadata,
-				updated_at = excluded.updated_at
-		`);
-
-		for (const doc of documents) {
-			if (!doc.key?.trim()) {
-				throw new Error('Each document must have a non-empty key');
-			}
-
-			// Generate or use provided embeddings
-			let embedding: number[];
-			if ('embeddings' in doc && doc.embeddings) {
-				if (!Array.isArray(doc.embeddings) || doc.embeddings.length === 0) {
-					throw new Error('Embeddings must be a non-empty array');
-				}
-				embedding = doc.embeddings;
-			} else if ('document' in doc && doc.document) {
-				if (!doc.document?.trim()) {
-					throw new Error('Document text must be non-empty');
-				}
-				embedding = simpleEmbedding(doc.document);
-			} else {
-				throw new Error('Each document must have either embeddings or document text');
-			}
-
-			const id = randomUUID();
-			const timestamp = now();
-			const embeddingJson = JSON.stringify(embedding);
-			const documentText = 'document' in doc ? doc.document : null;
-			const metadata = doc.metadata ? JSON.stringify(doc.metadata) : null;
-
-			stmt.run(
-				this.#projectPath,
-				name,
-				id,
-				doc.key,
-				embeddingJson,
-				documentText ?? null,
-				metadata ?? null,
-				timestamp,
-				timestamp
-			);
-
-			const row = this.#db
-				.prepare(
-					'SELECT id FROM vector_storage WHERE project_path = ? AND name = ? AND key = ?'
-				)
-				.get(this.#projectPath, name, doc.key) as { id: string } | undefined;
-
-			const actualId = row?.id ?? id;
-			results.push({ key: doc.key, id: actualId });
-		}
-
-		return results;
-	}
-
-	async get<T extends Record<string, unknown> = Record<string, unknown>>(
-		name: string,
-		key: string
-	): Promise<VectorResult<T>> {
-		if (!name?.trim() || !key?.trim()) {
-			throw new Error('Vector storage name and key are required');
-		}
-
-		const query = this.#db.query(`
-			SELECT id, key, embedding, document, metadata 
-			FROM vector_storage 
-			WHERE project_path = ? AND name = ? AND key = ?
-		`);
-
-		const row = query.get(this.#projectPath, name, key) as {
-			id: string;
-			key: string;
-			embedding: string;
-			document: string | null;
-			metadata: string | null;
-		} | null;
-
-		if (!row) {
-			return { exists: false } as VectorResultNotFound;
-		}
-
-		return {
-			exists: true,
-			data: {
-				id: row.id,
-				key: row.key,
-				embeddings: JSON.parse(row.embedding),
-				document: row.document || undefined,
-				metadata: row.metadata ? JSON.parse(row.metadata) : undefined,
-				similarity: 1.0, // Perfect match for direct get
-			} as VectorSearchResultWithDocument<T>,
-		};
-	}
-
-	async getMany<T extends Record<string, unknown> = Record<string, unknown>>(
-		name: string,
-		...keys: string[]
-	): Promise<Map<string, VectorSearchResultWithDocument<T>>> {
-		if (!name?.trim()) {
-			throw new Error('Vector storage name is required');
-		}
-		if (keys.length === 0) {
-			return new Map();
-		}
-
-		const results = await Promise.all(
-			keys.map(async (key) => {
-				const result = await this.get<T>(name, key);
-				return { key, result };
-			})
-		);
-
-		const map = new Map<string, VectorSearchResultWithDocument<T>>();
-		for (const { key, result } of results) {
-			if (result.exists) {
-				map.set(key, result.data);
-			}
-		}
-
-		return map;
-	}
-
-	async search<T extends Record<string, unknown> = Record<string, unknown>>(
-		name: string,
-		params: VectorSearchParams<T>
-	): Promise<VectorSearchResult<T>[]> {
-		if (!name?.trim()) {
-			throw new Error('Vector storage name is required');
-		}
-		if (!params.query?.trim()) {
-			throw new Error('Query is required');
-		}
-
-		// Fetch all vectors for this name
-		const query = this.#db.query(`
-			SELECT id, key, embedding, metadata 
-			FROM vector_storage 
-			WHERE project_path = ? AND name = ?
-		`);
-
-		const rows = query.all(this.#projectPath, name) as Array<{
-			id: string;
-			key: string;
-			embedding: string;
-			metadata: string | null;
-		}>;
-
-		// If no vectors exist, return empty results
-		const row = rows[0];
-		if (!row) {
-			return [];
-		}
-
-		// Detect dimensionality from first stored vector
-		const firstEmbedding = JSON.parse(row.embedding);
-		const dimensions = firstEmbedding.length;
-
-		// Generate query embedding with matching dimensions
-		const queryEmbedding = simpleEmbedding(params.query, dimensions);
-
-		// Calculate similarities
-		const results: Array<VectorSearchResult<T> & { similarity: number }> = [];
-
-		for (const row of rows) {
-			const embedding = JSON.parse(row.embedding);
-			const similarity = cosineSimilarity(queryEmbedding, embedding);
-
-			// Apply similarity threshold
-			if (params.similarity !== undefined && similarity < params.similarity) {
-				continue;
-			}
-
-			// Apply metadata filter
-			if (params.metadata) {
-				const rowMetadata = row.metadata ? JSON.parse(row.metadata) : {};
-				const matches = Object.entries(params.metadata).every(
-					([key, value]) => rowMetadata[key] === value
-				);
-				if (!matches) {
-					continue;
-				}
-			}
-
-			results.push({
-				id: row.id,
-				key: row.key,
-				metadata: row.metadata ? JSON.parse(row.metadata) : undefined,
-				similarity,
-			} as VectorSearchResult<T> & { similarity: number });
-		}
-
-		// Sort by similarity descending
-		results.sort((a, b) => b.similarity - a.similarity);
-
-		// Apply limit
-		const limit = params.limit || 10;
-		return results.slice(0, limit);
-	}
-
-	async delete(name: string, ...keys: string[]): Promise<number> {
-		if (!name?.trim()) {
-			throw new Error('Vector storage name is required');
-		}
-		if (keys.length === 0) {
-			return 0;
-		}
-
-		const placeholders = keys.map(() => '?').join(', ');
-		const stmt = this.#db.prepare(`
-			DELETE FROM vector_storage 
-			WHERE project_path = ? AND name = ? AND key IN (${placeholders})
-		`);
-
-		const result = stmt.run(this.#projectPath, name, ...keys);
-		return result.changes;
-	}
-
-	async exists(name: string): Promise<boolean> {
-		if (!name?.trim()) {
-			throw new Error('Vector storage name is required');
-		}
-
-		const query = this.#db.query(`
-			SELECT COUNT(*) as count 
-			FROM vector_storage 
-			WHERE project_path = ? AND name = ?
-		`);
-
-		const { count } = query.get(this.#projectPath, name) as { count: number };
-		return count > 0;
-	}
-
-	async getStats(name: string): Promise<VectorNamespaceStatsWithSamples> {
-		if (!name?.trim()) {
-			throw new Error('Vector storage name is required');
-		}
-
-		const countQuery = this.#db.query(`
-			SELECT COUNT(*) as count,
-			MIN(created_at) as created_at, MAX(updated_at) as last_used
-			FROM vector_storage 
-			WHERE project_path = ? AND name = ?
-		`);
-
-		const stats = countQuery.get(this.#projectPath, name) as {
-			count: number;
-			created_at: number | null;
-			last_used: number | null;
-		};
-
-		if (stats.count === 0) {
-			return { sum: 0, count: 0 };
-		}
-
-		const sampleQuery = this.#db.query(`
-			SELECT key, embedding, document, metadata, created_at, updated_at
-			FROM vector_storage 
-			WHERE project_path = ? AND name = ?
-			LIMIT 20
-		`);
-
-		const samples = sampleQuery.all(this.#projectPath, name) as Array<{
-			key: string;
-			embedding: string;
-			document: string | null;
-			metadata: string | null;
-			created_at: number;
-			updated_at: number;
-		}>;
-
-		const encoder = new TextEncoder();
-		let totalSum = 0;
-		const sampledResults: VectorNamespaceStatsWithSamples['sampledResults'] = {};
-		for (const sample of samples) {
-			const embeddingBytes = encoder.encode(sample.embedding).length;
-			const documentBytes = sample.document ? encoder.encode(sample.document).length : 0;
-			const size = embeddingBytes + documentBytes;
-			totalSum += size;
-			sampledResults![sample.key] = {
-				embedding: JSON.parse(sample.embedding),
-				document: sample.document || undefined,
-				size,
-				metadata: sample.metadata ? JSON.parse(sample.metadata) : undefined,
-				firstUsed: sample.created_at,
-				lastUsed: sample.updated_at,
-			};
-		}
-
-		// Estimate total size based on sampled average if we have more records than samples
-		const estimatedSum =
-			stats.count <= samples.length
-				? totalSum
-				: Math.round((totalSum / samples.length) * stats.count);
-
-		return {
-			sum: estimatedSum,
-			count: stats.count,
-			createdAt: stats.created_at || undefined,
-			lastUsed: stats.last_used || undefined,
-			sampledResults,
-		};
-	}
-
-	async getAllStats(
-		_params?: VectorGetAllStatsParams
-	): Promise<Record<string, VectorNamespaceStats> | VectorStatsPaginated> {
-		const query = this.#db.query(`
-			SELECT name, embedding, document
-			FROM vector_storage 
-			WHERE project_path = ?
-		`);
-
-		const rows = query.all(this.#projectPath) as Array<{
-			name: string;
-			embedding: string;
-			document: string | null;
-		}>;
-
-		const encoder = new TextEncoder();
-		const namespaceStats = new Map<
-			string,
-			{ sum: number; count: number; createdAt?: number; lastUsed?: number }
-		>();
-
-		for (const row of rows) {
-			const embeddingBytes = encoder.encode(row.embedding).length;
-			const documentBytes = row.document ? encoder.encode(row.document).length : 0;
-			const size = embeddingBytes + documentBytes;
-
-			const existing = namespaceStats.get(row.name);
-			if (existing) {
-				existing.sum += size;
-				existing.count += 1;
-			} else {
-				namespaceStats.set(row.name, { sum: size, count: 1 });
-			}
-		}
-
-		// Get timestamps in a separate query
-		const timestampQuery = this.#db.query(`
-			SELECT name, MIN(created_at) as created_at, MAX(updated_at) as last_used
-			FROM vector_storage 
-			WHERE project_path = ?
-			GROUP BY name
-		`);
-
-		const timestamps = timestampQuery.all(this.#projectPath) as Array<{
-			name: string;
-			created_at: number | null;
-			last_used: number | null;
-		}>;
-
-		for (const ts of timestamps) {
-			const stats = namespaceStats.get(ts.name);
-			if (stats) {
-				stats.createdAt = ts.created_at || undefined;
-				stats.lastUsed = ts.last_used || undefined;
-			}
-		}
-
-		const results: Record<string, VectorNamespaceStats> = {};
-		for (const [name, stats] of namespaceStats) {
-			results[name] = stats;
-		}
-
-		return results;
-	}
-
-	async getNamespaces(): Promise<string[]> {
-		const query = this.#db.query(`
-			SELECT DISTINCT name 
-			FROM vector_storage 
-			WHERE project_path = ?
-		`);
-
-		const rows = query.all(this.#projectPath) as Array<{ name: string }>;
-		return rows.map((row) => row.name);
-	}
-
-	async deleteNamespace(name: string): Promise<void> {
-		if (!name?.trim()) {
-			throw new Error('Vector storage name is required');
-		}
-
-		const stmt = this.#db.prepare(`
-			DELETE FROM vector_storage 
-			WHERE project_path = ? AND name = ?
-		`);
-
-		stmt.run(this.#projectPath, name);
-	}
-}