@semiont/vectors 0.4.12

package/README.md

@@ -0,0 +1,148 @@
+# @semiont/vectors
+
+Vector storage, embedding, and semantic search for Semiont.
+
+Provides a pluggable abstraction over vector databases and embedding providers, with text chunking utilities. Used by the Smelter actor to index content and by Gatherer/Matcher to retrieve semantically similar resources and annotations.
+
+## Architecture
+
+Two separate vector collections:
+
+- **resources** — chunked full-text content from stored files
+- **annotations** — W3C Web Annotation entities with motivation, entity types, and exact text
+
+Both collections support filtered similarity search with configurable score thresholds.
+
+## Vector Stores
+
+### Qdrant (production)
+
+```typescript
+import { createVectorStore } from '@semiont/vectors';
+
+const store = await createVectorStore({
+  type: 'qdrant',
+  host: 'localhost',
+  port: 6333,
+  dimensions: 1024,
+});
+```
+
+Requires a running [Qdrant](https://qdrant.tech) instance. The `@qdrant/js-client-rest` peer dependency is lazy-loaded on `connect()`. Collections are auto-created if they don't exist.
+
+### Memory (testing)
+
+```typescript
+const store = await createVectorStore({
+  type: 'memory',
+  dimensions: 768,
+});
+```
+
+Brute-force cosine similarity. No external dependencies.
+
+## Embedding Providers
+
+### Voyage AI (cloud)
+
+```typescript
+import { createEmbeddingProvider } from '@semiont/vectors';
+
+const provider = await createEmbeddingProvider({
+  type: 'voyage',
+  model: 'voyage-3',       // 1024 dimensions
+  apiKey: '...',
+});
+```
+
+Models: `voyage-3` (1024), `voyage-3-lite` (512), `voyage-code-3`, `voyage-finance-2`, `voyage-law-2`.
+
+### Ollama (local)
+
+```typescript
+const provider = await createEmbeddingProvider({
+  type: 'ollama',
+  model: 'nomic-embed-text',  // 768 dimensions
+  baseURL: 'http://localhost:11434',
+});
+```
+
+Models: `nomic-embed-text` (768), `all-minilm` (384), `mxbai-embed-large` (1024), `snowflake-arctic-embed` (1024).
+
+## Text Chunking
+
+```typescript
+import { chunkText, DEFAULT_CHUNKING_CONFIG } from '@semiont/vectors';
+
+const chunks = chunkText(longDocument, { chunkSize: 512, overlap: 50 });
+// => string[]
+```
+
+Splits on paragraph boundaries, then sentence boundaries, then word boundaries. `chunkSize` and `overlap` are in tokens (~4 characters per token).
+
+## Search
+
+```typescript
+const embedding = await provider.embed('quantum computing');
+
+// Search resources
+const resources = await store.searchResources(embedding, {
+  limit: 10,
+  scoreThreshold: 0.7,
+  filter: { excludeResourceId: 'res-already-open' },
+});
+
+// Search annotations
+const annotations = await store.searchAnnotations(embedding, {
+  limit: 5,
+  filter: { entityTypes: ['Person', 'Organization'], motivation: 'describing' },
+});
+```
+
+Each result includes `id`, `score`, `resourceId`, `text`, and optionally `annotationId` and `entityTypes`.
+
+## Writing Vectors
+
+```typescript
+// Index a resource's content
+const chunks = chunkText(content, DEFAULT_CHUNKING_CONFIG);
+const embeddings = await provider.embedBatch(chunks);
+await store.upsertResourceVectors(resourceId, chunks.map((text, i) => ({
+  chunkIndex: i,
+  text,
+  embedding: embeddings[i],
+})));
+
+// Index an annotation
+const vec = await provider.embed(annotation.exactText);
+await store.upsertAnnotationVector(annotationId, {
+  annotationId,
+  resourceId,
+  motivation: 'describing',
+  entityTypes: ['Person'],
+  exactText: 'Marie Curie',
+}, vec);
+```
+
+## Configuration
+
+In `semiont.toml`:
+
+```toml
+[environments.local.services.vectors]
+type = "qdrant"
+host = "localhost"
+port = 6333
+
+[environments.local.services.vectors.embedding]
+type = "voyage"
+model = "voyage-3"
+
+[environments.local.services.vectors.chunking]
+chunkSize = 512
+overlap = 50
+```
+
+## License
+
+Apache-2.0

package/dist/chunk-GZRNY7LY.js

@@ -0,0 +1,128 @@
+// src/store/qdrant.ts
+var QdrantVectorStore = class {
+  client = null;
+  connected = false;
+  config;
+  constructor(config) {
+    this.config = config;
+  }
+  async connect() {
+    const { QdrantClient } = await import("@qdrant/js-client-rest");
+    this.client = new QdrantClient({
+      host: this.config.host,
+      port: this.config.port
+    });
+    await this.ensureCollection("resources", this.config.dimensions);
+    await this.ensureCollection("annotations", this.config.dimensions);
+    this.connected = true;
+  }
+  async disconnect() {
+    this.client = null;
+    this.connected = false;
+  }
+  isConnected() {
+    return this.connected;
+  }
+  async ensureCollection(name, dimensions) {
+    try {
+      await this.client.getCollection(name);
+    } catch {
+      await this.client.createCollection(name, {
+        vectors: { size: dimensions, distance: "Cosine" }
+      });
+    }
+  }
+  async upsertResourceVectors(resourceId, chunks) {
+    if (chunks.length === 0) return;
+    const points = chunks.map((chunk) => ({
+      id: `${resourceId}-${chunk.chunkIndex}`,
+      vector: chunk.embedding,
+      payload: {
+        resourceId: String(resourceId),
+        chunkIndex: chunk.chunkIndex,
+        text: chunk.text
+      }
+    }));
+    await this.client.upsert("resources", { points });
+  }
+  async upsertAnnotationVector(annotationId, embedding, payload) {
+    await this.client.upsert("annotations", {
+      points: [{
+        id: String(annotationId),
+        vector: embedding,
+        payload: {
+          annotationId: String(payload.annotationId),
+          resourceId: String(payload.resourceId),
+          motivation: payload.motivation,
+          entityTypes: payload.entityTypes,
+          text: payload.exactText
+        }
+      }]
+    });
+  }
+  async deleteResourceVectors(resourceId) {
+    await this.client.delete("resources", {
+      filter: {
+        must: [{ key: "resourceId", match: { value: String(resourceId) } }]
+      }
+    });
+  }
+  async deleteAnnotationVector(annotationId) {
+    await this.client.delete("annotations", {
+      points: [String(annotationId)]
+    });
+  }
+  async searchResources(embedding, opts) {
+    return this.search("resources", embedding, opts);
+  }
+  async searchAnnotations(embedding, opts) {
+    return this.search("annotations", embedding, opts);
+  }
+  async search(collection, embedding, opts) {
+    const filter = this.buildFilter(opts.filter);
+    const results = await this.client.search(collection, {
+      vector: embedding,
+      limit: opts.limit,
+      score_threshold: opts.scoreThreshold,
+      filter: filter || void 0,
+      with_payload: true
+    });
+    return results.map((r) => ({
+      id: String(r.id),
+      score: r.score,
+      resourceId: r.payload.resourceId,
+      annotationId: r.payload.annotationId,
+      text: r.payload.text,
+      entityTypes: r.payload.entityTypes
+    }));
+  }
+  buildFilter(filter) {
+    if (!filter) return null;
+    const must = [];
+    if (filter.entityTypes && filter.entityTypes.length > 0) {
+      for (const et of filter.entityTypes) {
+        must.push({ key: "entityTypes", match: { value: et } });
+      }
+    }
+    if (filter.resourceId) {
+      must.push({ key: "resourceId", match: { value: String(filter.resourceId) } });
+    }
+    if (filter.motivation) {
+      must.push({ key: "motivation", match: { value: filter.motivation } });
+    }
+    const must_not = [];
+    if (filter.excludeResourceId) {
+      must_not.push({ key: "resourceId", match: { value: String(filter.excludeResourceId) } });
+    }
+    if (must.length === 0 && must_not.length === 0) return null;
+    return {
+      ...must.length > 0 ? { must } : {},
+      ...must_not.length > 0 ? { must_not } : {}
+    };
+  }
+};
+
+export {
+  QdrantVectorStore
+};
+//# sourceMappingURL=chunk-GZRNY7LY.js.map

package/dist/chunk-GZRNY7LY.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/store/qdrant.ts"],"sourcesContent":["/**\n * Qdrant VectorStore Implementation\n *\n * Uses the Qdrant REST API via @qdrant/js-client-rest.\n * Manages two collections: 'resources' and 'annotations'.\n */\n\nimport type { ResourceId, AnnotationId } from '@semiont/core';\nimport type { VectorStore, EmbeddingChunk, AnnotationPayload, VectorSearchResult, SearchOptions } from './interface';\n\nexport interface QdrantConfig {\n  host: string;\n  port: number;\n  dimensions: number;\n}\n\nexport class QdrantVectorStore implements VectorStore {\n  private client: any = null;\n  private connected = false;\n  private config: QdrantConfig;\n\n  constructor(config: QdrantConfig) {\n    this.config = config;\n  }\n\n  async connect(): Promise<void> {\n    const { QdrantClient } = await import('@qdrant/js-client-rest');\n    this.client = new QdrantClient({\n      host: this.config.host,\n      port: this.config.port,\n    });\n\n    // Ensure collections exist\n    await this.ensureCollection('resources', this.config.dimensions);\n    await this.ensureCollection('annotations', this.config.dimensions);\n    this.connected = true;\n  }\n\n  async disconnect(): Promise<void> {\n    this.client = null;\n    this.connected = false;\n  }\n\n  isConnected(): boolean {\n    return this.connected;\n  }\n\n  private async ensureCollection(name: string, dimensions: number): Promise<void> {\n    try {\n      await this.client.getCollection(name);\n    } catch {\n      await this.client.createCollection(name, {\n        vectors: { size: dimensions, distance: 'Cosine' },\n      });\n    }\n  }\n\n  async upsertResourceVectors(resourceId: ResourceId, chunks: EmbeddingChunk[]): Promise<void> {\n    if (chunks.length === 0) return;\n\n    const points = chunks.map((chunk) => ({\n      id: `${resourceId}-${chunk.chunkIndex}`,\n      vector: chunk.embedding,\n      payload: {\n        resourceId: String(resourceId),\n        chunkIndex: chunk.chunkIndex,\n        text: chunk.text,\n      },\n    }));\n\n    await this.client.upsert('resources', { points });\n  }\n\n  async upsertAnnotationVector(\n    annotationId: AnnotationId,\n    embedding: number[],\n    payload: AnnotationPayload\n  ): Promise<void> {\n    await this.client.upsert('annotations', {\n      points: [{\n        id: String(annotationId),\n        vector: embedding,\n        payload: {\n          annotationId: String(payload.annotationId),\n          resourceId: String(payload.resourceId),\n          motivation: payload.motivation,\n          entityTypes: payload.entityTypes,\n          text: payload.exactText,\n        },\n      }],\n    });\n  }\n\n  async deleteResourceVectors(resourceId: ResourceId): Promise<void> {\n    await this.client.delete('resources', {\n      filter: {\n        must: [{ key: 'resourceId', match: { value: String(resourceId) } }],\n      },\n    });\n  }\n\n  async deleteAnnotationVector(annotationId: AnnotationId): Promise<void> {\n    await this.client.delete('annotations', {\n      points: [String(annotationId)],\n    });\n  }\n\n  async searchResources(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]> {\n    return this.search('resources', embedding, opts);\n  }\n\n  async searchAnnotations(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]> {\n    return this.search('annotations', embedding, opts);\n  }\n\n  private async search(collection: string, embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]> {\n    const filter = this.buildFilter(opts.filter);\n\n    const results = await this.client.search(collection, {\n      vector: embedding,\n      limit: opts.limit,\n      score_threshold: opts.scoreThreshold,\n      filter: filter || undefined,\n      with_payload: true,\n    });\n\n    return results.map((r: any) => ({\n      id: String(r.id),\n      score: r.score,\n      resourceId: r.payload.resourceId as ResourceId,\n      annotationId: r.payload.annotationId as AnnotationId | undefined,\n      text: r.payload.text as string,\n      entityTypes: r.payload.entityTypes as string[] | undefined,\n    }));\n  }\n\n  private buildFilter(filter?: SearchOptions['filter']): any | null {\n    if (!filter) return null;\n\n    const must: any[] = [];\n\n    if (filter.entityTypes && filter.entityTypes.length > 0) {\n      for (const et of filter.entityTypes) {\n        must.push({ key: 'entityTypes', match: { value: et } });\n      }\n    }\n\n    if (filter.resourceId) {\n      must.push({ key: 'resourceId', match: { value: String(filter.resourceId) } });\n    }\n\n    if (filter.motivation) {\n      must.push({ key: 'motivation', match: { value: filter.motivation } });\n    }\n\n    const must_not: any[] = [];\n\n    if (filter.excludeResourceId) {\n      must_not.push({ key: 'resourceId', match: { value: String(filter.excludeResourceId) } });\n    }\n\n    if (must.length === 0 && must_not.length === 0) return null;\n\n    return {\n      ...(must.length > 0 ? { must } : {}),\n      ...(must_not.length > 0 ? { must_not } : {}),\n    };\n  }\n}\n"],"mappings":";AAgBO,IAAM,oBAAN,MAA+C;AAAA,EAC5C,SAAc;AAAA,EACd,YAAY;AAAA,EACZ;AAAA,EAER,YAAY,QAAsB;AAChC,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,MAAM,UAAyB;AAC7B,UAAM,EAAE,aAAa,IAAI,MAAM,OAAO,wBAAwB;AAC9D,SAAK,SAAS,IAAI,aAAa;AAAA,MAC7B,MAAM,KAAK,OAAO;AAAA,MAClB,MAAM,KAAK,OAAO;AAAA,IACpB,CAAC;AAGD,UAAM,KAAK,iBAAiB,aAAa,KAAK,OAAO,UAAU;AAC/D,UAAM,KAAK,iBAAiB,eAAe,KAAK,OAAO,UAAU;AACjE,SAAK,YAAY;AAAA,EACnB;AAAA,EAEA,MAAM,aAA4B;AAChC,SAAK,SAAS;AACd,SAAK,YAAY;AAAA,EACnB;AAAA,EAEA,cAAuB;AACrB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAc,iBAAiB,MAAc,YAAmC;AAC9E,QAAI;AACF,YAAM,KAAK,OAAO,cAAc,IAAI;AAAA,IACtC,QAAQ;AACN,YAAM,KAAK,OAAO,iBAAiB,MAAM;AAAA,QACvC,SAAS,EAAE,MAAM,YAAY,UAAU,SAAS;AAAA,MAClD,CAAC;AAAA,IACH;AAAA,EACF;AAAA,EAEA,MAAM,sBAAsB,YAAwB,QAAyC;AAC3F,QAAI,OAAO,WAAW,EAAG;AAEzB,UAAM,SAAS,OAAO,IAAI,CAAC,WAAW;AAAA,MACpC,IAAI,GAAG,UAAU,IAAI,MAAM,UAAU;AAAA,MACrC,QAAQ,MAAM;AAAA,MACd,SAAS;AAAA,QACP,YAAY,OAAO,UAAU;AAAA,QAC7B,YAAY,MAAM;AAAA,QAClB,MAAM,MAAM;AAAA,MACd;AAAA,IACF,EAAE;AAEF,UAAM,KAAK,OAAO,OAAO,aAAa,EAAE,OAAO,CAAC;AAAA,EAClD;AAAA,EAEA,MAAM,uBACJ,cACA,WACA,SACe;AACf,UAAM,KAAK,OAAO,OAAO,eAAe;AAAA,MACtC,QAAQ,CAAC;AAAA,QACP,IAAI,OAAO,YAAY;AAAA,QACvB,QAAQ;AAAA,QACR,SAAS;AAAA,UACP,cAAc,OAAO,QAAQ,YAAY;AAAA,UACzC,YAAY,OAAO,QAAQ,UAAU;AAAA,UACrC,YAAY,QAAQ;AAAA,UACpB,aAAa,QAAQ;AAAA,UACrB,MAAM,QAAQ;AAAA,QAChB;AAAA,MACF,CAAC;AAAA,IACH,CAAC;AAAA,EACH;AAAA,EAEA,MAAM,sBAAsB,YAAuC;AACjE,UAAM,KAAK,OAAO,OAAO,aAAa;AAAA,MACpC,QAAQ;AAAA,QACN,MAAM,CAAC,EAAE,KAAK,cAAc,OAAO,EAAE,OAAO,OAAO,UAAU,EAAE,EAAE,CAAC;AAAA,MACpE;AAAA,IACF,CAAC;AAAA,EACH;AAAA,EAEA,MAAM,uBAAuB,cAA2C;AACtE,UAAM,KAAK,OAAO,OAAO,eAAe;AAAA,MACtC,QAAQ,CAAC,OAAO,YAAY,CAAC;AAAA,IAC/B,CAAC;AAAA,EACH;AAAA,EAEA,MAAM,gBAAgB,WAAqB,MAAoD;AAC7F,WAAO,KAAK,OAAO,aAAa,WAAW,IAAI;AAAA,EACjD;AAAA,EAEA,MAAM,kBAAkB,WAAqB,MAAoD;AAC/F,WAAO,KAAK,OAAO,eAAe,WAAW,IAAI;AAAA,EACnD;AAAA,EAEA,MAAc,OAAO,YAAoB,WAAqB,MAAoD;AAChH,UAAM,SAAS,KAAK,YAAY,KAAK,MAAM;AAE3C,UAAM,UAAU,MAAM,KAAK,OAAO,OAAO,YAAY;AAAA,MACnD,QAAQ;AAAA,MACR,OAAO,KAAK;AAAA,MACZ,iBAAiB,KAAK;AAAA,MACtB,QAAQ,UAAU;AAAA,MAClB,cAAc;AAAA,IAChB,CAAC;AAED,WAAO,QAAQ,IAAI,CAAC,OAAY;AAAA,MAC9B,IAAI,OAAO,EAAE,EAAE;AAAA,MACf,OAAO,EAAE;AAAA,MACT,YAAY,EAAE,QAAQ;AAAA,MACtB,cAAc,EAAE,QAAQ;AAAA,MACxB,MAAM,EAAE,QAAQ;AAAA,MAChB,aAAa,EAAE,QAAQ;AAAA,IACzB,EAAE;AAAA,EACJ;AAAA,EAEQ,YAAY,QAA8C;AAChE,QAAI,CAAC,OAAQ,QAAO;AAEpB,UAAM,OAAc,CAAC;AAErB,QAAI,OAAO,eAAe,OAAO,YAAY,SAAS,GAAG;AACvD,iBAAW,MAAM,OAAO,aAAa;AACnC,aAAK,KAAK,EAAE,KAAK,eAAe,OAAO,EAAE,OAAO,GAAG,EAAE,CAAC;AAAA,MACxD;AAAA,IACF;AAEA,QAAI,OAAO,YAAY;AACrB,WAAK,KAAK,EAAE,KAAK,cAAc,OAAO,EAAE,OAAO,OAAO,OAAO,UAAU,EAAE,EAAE,CAAC;AAAA,IAC9E;AAEA,QAAI,OAAO,YAAY;AACrB,WAAK,KAAK,EAAE,KAAK,cAAc,OAAO,EAAE,OAAO,OAAO,WAAW,EAAE,CAAC;AAAA,IACtE;AAEA,UAAM,WAAkB,CAAC;AAEzB,QAAI,OAAO,mBAAmB;AAC5B,eAAS,KAAK,EAAE,KAAK,cAAc,OAAO,EAAE,OAAO,OAAO,OAAO,iBAAiB,EAAE,EAAE,CAAC;AAAA,IACzF;AAEA,QAAI,KAAK,WAAW,KAAK,SAAS,WAAW,EAAG,QAAO;AAEvD,WAAO;AAAA,MACL,GAAI,KAAK,SAAS,IAAI,EAAE,KAAK,IAAI,CAAC;AAAA,MAClC,GAAI,SAAS,SAAS,IAAI,EAAE,SAAS,IAAI,CAAC;AAAA,IAC5C;AAAA,EACF;AACF;","names":[]}

package/dist/chunk-INCF7JMV.js

@@ -0,0 +1,49 @@
+// src/embedding/voyage.ts
+var VOYAGE_DIMENSIONS = {
+  "voyage-3": 1024,
+  "voyage-3-lite": 512,
+  "voyage-code-3": 1024,
+  "voyage-finance-2": 1024,
+  "voyage-law-2": 1024
+};
+var VoyageEmbeddingProvider = class {
+  config;
+  constructor(config) {
+    this.config = config;
+  }
+  async embed(text) {
+    const results = await this.embedBatch([text]);
+    return results[0];
+  }
+  async embedBatch(texts) {
+    const endpoint = this.config.endpoint ?? "https://api.voyageai.com/v1/embeddings";
+    const response = await fetch(endpoint, {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+        "Authorization": `Bearer ${this.config.apiKey}`
+      },
+      body: JSON.stringify({
+        model: this.config.model,
+        input: texts
+      })
+    });
+    if (!response.ok) {
+      const body = await response.text();
+      throw new Error(`Voyage API error ${response.status}: ${body}`);
+    }
+    const json = await response.json();
+    return json.data.map((d) => d.embedding);
+  }
+  dimensions() {
+    return VOYAGE_DIMENSIONS[this.config.model] ?? 1024;
+  }
+  model() {
+    return this.config.model;
+  }
+};
+
+export {
+  VoyageEmbeddingProvider
+};
+//# sourceMappingURL=chunk-INCF7JMV.js.map

package/dist/chunk-INCF7JMV.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/embedding/voyage.ts"],"sourcesContent":["/**\n * Voyage AI Embedding Provider\n *\n * Cloud embedding via the Voyage AI API (partner of Anthropic).\n * Uses the same API key as Anthropic inference.\n */\n\nimport type { EmbeddingProvider } from './interface';\n\nexport interface VoyageConfig {\n  apiKey: string;\n  model: string;\n  endpoint?: string;\n}\n\nconst VOYAGE_DIMENSIONS: Record<string, number> = {\n  'voyage-3': 1024,\n  'voyage-3-lite': 512,\n  'voyage-code-3': 1024,\n  'voyage-finance-2': 1024,\n  'voyage-law-2': 1024,\n};\n\nexport class VoyageEmbeddingProvider implements EmbeddingProvider {\n  private config: VoyageConfig;\n\n  constructor(config: VoyageConfig) {\n    this.config = config;\n  }\n\n  async embed(text: string): Promise<number[]> {\n    const results = await this.embedBatch([text]);\n    return results[0];\n  }\n\n  async embedBatch(texts: string[]): Promise<number[][]> {\n    const endpoint = this.config.endpoint ?? 'https://api.voyageai.com/v1/embeddings';\n\n    const response = await fetch(endpoint, {\n      method: 'POST',\n      headers: {\n        'Content-Type': 'application/json',\n        'Authorization': `Bearer ${this.config.apiKey}`,\n      },\n      body: JSON.stringify({\n        model: this.config.model,\n        input: texts,\n      }),\n    });\n\n    if (!response.ok) {\n      const body = await response.text();\n      throw new Error(`Voyage API error ${response.status}: ${body}`);\n    }\n\n    const json = await response.json() as { data: Array<{ embedding: number[] }> };\n    return json.data.map(d => d.embedding);\n  }\n\n  dimensions(): number {\n    return VOYAGE_DIMENSIONS[this.config.model] ?? 1024;\n  }\n\n  model(): string {\n    return this.config.model;\n  }\n}\n"],"mappings":";AAeA,IAAM,oBAA4C;AAAA,EAChD,YAAY;AAAA,EACZ,iBAAiB;AAAA,EACjB,iBAAiB;AAAA,EACjB,oBAAoB;AAAA,EACpB,gBAAgB;AAClB;AAEO,IAAM,0BAAN,MAA2D;AAAA,EACxD;AAAA,EAER,YAAY,QAAsB;AAChC,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,MAAM,MAAM,MAAiC;AAC3C,UAAM,UAAU,MAAM,KAAK,WAAW,CAAC,IAAI,CAAC;AAC5C,WAAO,QAAQ,CAAC;AAAA,EAClB;AAAA,EAEA,MAAM,WAAW,OAAsC;AACrD,UAAM,WAAW,KAAK,OAAO,YAAY;AAEzC,UAAM,WAAW,MAAM,MAAM,UAAU;AAAA,MACrC,QAAQ;AAAA,MACR,SAAS;AAAA,QACP,gBAAgB;AAAA,QAChB,iBAAiB,UAAU,KAAK,OAAO,MAAM;AAAA,MAC/C;AAAA,MACA,MAAM,KAAK,UAAU;AAAA,QACnB,OAAO,KAAK,OAAO;AAAA,QACnB,OAAO;AAAA,MACT,CAAC;AAAA,IACH,CAAC;AAED,QAAI,CAAC,SAAS,IAAI;AAChB,YAAM,OAAO,MAAM,SAAS,KAAK;AACjC,YAAM,IAAI,MAAM,oBAAoB,SAAS,MAAM,KAAK,IAAI,EAAE;AAAA,IAChE;AAEA,UAAM,OAAO,MAAM,SAAS,KAAK;AACjC,WAAO,KAAK,KAAK,IAAI,OAAK,EAAE,SAAS;AAAA,EACvC;AAAA,EAEA,aAAqB;AACnB,WAAO,kBAAkB,KAAK,OAAO,KAAK,KAAK;AAAA,EACjD;AAAA,EAEA,QAAgB;AACd,WAAO,KAAK,OAAO;AAAA,EACrB;AACF;","names":[]}

package/dist/chunk-LBAPGZOW.js

@@ -0,0 +1,58 @@
+// src/embedding/ollama.ts
+var OLLAMA_DIMENSIONS = {
+  "nomic-embed-text": 768,
+  "all-minilm": 384,
+  "mxbai-embed-large": 1024,
+  "snowflake-arctic-embed": 1024
+};
+var OllamaEmbeddingProvider = class {
+  config;
+  constructor(config) {
+    this.config = config;
+  }
+  async embed(text) {
+    const baseURL = this.config.baseURL ?? "http://localhost:11434";
+    const response = await fetch(`${baseURL}/api/embed`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({
+        model: this.config.model,
+        input: text
+      })
+    });
+    if (!response.ok) {
+      const body = await response.text();
+      throw new Error(`Ollama embed error ${response.status}: ${body}`);
+    }
+    const json = await response.json();
+    return json.embeddings[0];
+  }
+  async embedBatch(texts) {
+    const baseURL = this.config.baseURL ?? "http://localhost:11434";
+    const response = await fetch(`${baseURL}/api/embed`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({
+        model: this.config.model,
+        input: texts
+      })
+    });
+    if (!response.ok) {
+      const body = await response.text();
+      throw new Error(`Ollama embed error ${response.status}: ${body}`);
+    }
+    const json = await response.json();
+    return json.embeddings;
+  }
+  dimensions() {
+    return OLLAMA_DIMENSIONS[this.config.model] ?? 768;
+  }
+  model() {
+    return this.config.model;
+  }
+};
+
+export {
+  OllamaEmbeddingProvider
+};
+//# sourceMappingURL=chunk-LBAPGZOW.js.map

package/dist/chunk-LBAPGZOW.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/embedding/ollama.ts"],"sourcesContent":["/**\n * Ollama Embedding Provider\n *\n * Local embedding via the Ollama API.\n * Uses models like nomic-embed-text, all-minilm, etc.\n */\n\nimport type { EmbeddingProvider } from './interface';\n\nexport interface OllamaEmbeddingConfig {\n  model: string;\n  baseURL?: string;\n}\n\nconst OLLAMA_DIMENSIONS: Record<string, number> = {\n  'nomic-embed-text': 768,\n  'all-minilm': 384,\n  'mxbai-embed-large': 1024,\n  'snowflake-arctic-embed': 1024,\n};\n\nexport class OllamaEmbeddingProvider implements EmbeddingProvider {\n  private config: OllamaEmbeddingConfig;\n\n  constructor(config: OllamaEmbeddingConfig) {\n    this.config = config;\n  }\n\n  async embed(text: string): Promise<number[]> {\n    const baseURL = this.config.baseURL ?? 'http://localhost:11434';\n\n    const response = await fetch(`${baseURL}/api/embed`, {\n      method: 'POST',\n      headers: { 'Content-Type': 'application/json' },\n      body: JSON.stringify({\n        model: this.config.model,\n        input: text,\n      }),\n    });\n\n    if (!response.ok) {\n      const body = await response.text();\n      throw new Error(`Ollama embed error ${response.status}: ${body}`);\n    }\n\n    const json = await response.json() as { embeddings: number[][] };\n    return json.embeddings[0];\n  }\n\n  async embedBatch(texts: string[]): Promise<number[][]> {\n    // Ollama's /api/embed supports batch input\n    const baseURL = this.config.baseURL ?? 'http://localhost:11434';\n\n    const response = await fetch(`${baseURL}/api/embed`, {\n      method: 'POST',\n      headers: { 'Content-Type': 'application/json' },\n      body: JSON.stringify({\n        model: this.config.model,\n        input: texts,\n      }),\n    });\n\n    if (!response.ok) {\n      const body = await response.text();\n      throw new Error(`Ollama embed error ${response.status}: ${body}`);\n    }\n\n    const json = await response.json() as { embeddings: number[][] };\n    return json.embeddings;\n  }\n\n  dimensions(): number {\n    return OLLAMA_DIMENSIONS[this.config.model] ?? 768;\n  }\n\n  model(): string {\n    return this.config.model;\n  }\n}\n"],"mappings":";AAcA,IAAM,oBAA4C;AAAA,EAChD,oBAAoB;AAAA,EACpB,cAAc;AAAA,EACd,qBAAqB;AAAA,EACrB,0BAA0B;AAC5B;AAEO,IAAM,0BAAN,MAA2D;AAAA,EACxD;AAAA,EAER,YAAY,QAA+B;AACzC,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,MAAM,MAAM,MAAiC;AAC3C,UAAM,UAAU,KAAK,OAAO,WAAW;AAEvC,UAAM,WAAW,MAAM,MAAM,GAAG,OAAO,cAAc;AAAA,MACnD,QAAQ;AAAA,MACR,SAAS,EAAE,gBAAgB,mBAAmB;AAAA,MAC9C,MAAM,KAAK,UAAU;AAAA,QACnB,OAAO,KAAK,OAAO;AAAA,QACnB,OAAO;AAAA,MACT,CAAC;AAAA,IACH,CAAC;AAED,QAAI,CAAC,SAAS,IAAI;AAChB,YAAM,OAAO,MAAM,SAAS,KAAK;AACjC,YAAM,IAAI,MAAM,sBAAsB,SAAS,MAAM,KAAK,IAAI,EAAE;AAAA,IAClE;AAEA,UAAM,OAAO,MAAM,SAAS,KAAK;AACjC,WAAO,KAAK,WAAW,CAAC;AAAA,EAC1B;AAAA,EAEA,MAAM,WAAW,OAAsC;AAErD,UAAM,UAAU,KAAK,OAAO,WAAW;AAEvC,UAAM,WAAW,MAAM,MAAM,GAAG,OAAO,cAAc;AAAA,MACnD,QAAQ;AAAA,MACR,SAAS,EAAE,gBAAgB,mBAAmB;AAAA,MAC9C,MAAM,KAAK,UAAU;AAAA,QACnB,OAAO,KAAK,OAAO;AAAA,QACnB,OAAO;AAAA,MACT,CAAC;AAAA,IACH,CAAC;AAED,QAAI,CAAC,SAAS,IAAI;AAChB,YAAM,OAAO,MAAM,SAAS,KAAK;AACjC,YAAM,IAAI,MAAM,sBAAsB,SAAS,MAAM,KAAK,IAAI,EAAE;AAAA,IAClE;AAEA,UAAM,OAAO,MAAM,SAAS,KAAK;AACjC,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,aAAqB;AACnB,WAAO,kBAAkB,KAAK,OAAO,KAAK,KAAK;AAAA,EACjD;AAAA,EAEA,QAAgB;AACd,WAAO,KAAK,OAAO;AAAA,EACrB;AACF;","names":[]}

package/dist/index.d.ts

@@ -0,0 +1,214 @@
+import { AnnotationId, ResourceId } from '@semiont/core';
+
+/**
+ * VectorStore Interface
+ *
+ * Abstraction over vector database backends (Qdrant, memory).
+ * Stores pre-computed embedding vectors with metadata payloads
+ * and provides similarity search with payload filtering.
+ */
+
+interface EmbeddingChunk {
+    chunkIndex: number;
+    text: string;
+    embedding: number[];
+}
+interface AnnotationPayload {
+    annotationId: AnnotationId;
+    resourceId: ResourceId;
+    motivation: string;
+    entityTypes: string[];
+    exactText: string;
+}
+interface VectorSearchResult {
+    id: string;
+    score: number;
+    resourceId: ResourceId;
+    annotationId?: AnnotationId;
+    text: string;
+    entityTypes?: string[];
+}
+interface SearchOptions {
+    limit: number;
+    scoreThreshold?: number;
+    filter?: {
+        entityTypes?: string[];
+        resourceId?: ResourceId;
+        motivation?: string;
+        excludeResourceId?: ResourceId;
+    };
+}
+interface VectorStore {
+    connect(): Promise<void>;
+    disconnect(): Promise<void>;
+    isConnected(): boolean;
+    upsertResourceVectors(resourceId: ResourceId, chunks: EmbeddingChunk[]): Promise<void>;
+    upsertAnnotationVector(annotationId: AnnotationId, embedding: number[], payload: AnnotationPayload): Promise<void>;
+    deleteResourceVectors(resourceId: ResourceId): Promise<void>;
+    deleteAnnotationVector(annotationId: AnnotationId): Promise<void>;
+    searchResources(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]>;
+    searchAnnotations(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]>;
+}
+
+/**
+ * Qdrant VectorStore Implementation
+ *
+ * Uses the Qdrant REST API via @qdrant/js-client-rest.
+ * Manages two collections: 'resources' and 'annotations'.
+ */
+
+interface QdrantConfig {
+    host: string;
+    port: number;
+    dimensions: number;
+}
+declare class QdrantVectorStore implements VectorStore {
+    private client;
+    private connected;
+    private config;
+    constructor(config: QdrantConfig);
+    connect(): Promise<void>;
+    disconnect(): Promise<void>;
+    isConnected(): boolean;
+    private ensureCollection;
+    upsertResourceVectors(resourceId: ResourceId, chunks: EmbeddingChunk[]): Promise<void>;
+    upsertAnnotationVector(annotationId: AnnotationId, embedding: number[], payload: AnnotationPayload): Promise<void>;
+    deleteResourceVectors(resourceId: ResourceId): Promise<void>;
+    deleteAnnotationVector(annotationId: AnnotationId): Promise<void>;
+    searchResources(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]>;
+    searchAnnotations(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]>;
+    private search;
+    private buildFilter;
+}
+
+/**
+ * In-Memory VectorStore Implementation
+ *
+ * For testing and development without a running Qdrant instance.
+ * Uses brute-force cosine similarity search.
+ */
+
+declare class MemoryVectorStore implements VectorStore {
+    private resources;
+    private annotations;
+    private connected;
+    connect(): Promise<void>;
+    disconnect(): Promise<void>;
+    isConnected(): boolean;
+    upsertResourceVectors(resourceId: ResourceId, chunks: EmbeddingChunk[]): Promise<void>;
+    upsertAnnotationVector(annotationId: AnnotationId, embedding: number[], payload: AnnotationPayload): Promise<void>;
+    deleteResourceVectors(resourceId: ResourceId): Promise<void>;
+    deleteAnnotationVector(annotationId: AnnotationId): Promise<void>;
+    searchResources(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]>;
+    searchAnnotations(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]>;
+    private search;
+    private toResult;
+}
+
+/**
+ * VectorStore Factory
+ *
+ * Creates a VectorStore instance based on configuration.
+ */
+
+interface VectorStoreConfig {
+    type: 'qdrant' | 'memory';
+    host?: string;
+    port?: number;
+    dimensions: number;
+}
+declare function createVectorStore(config: VectorStoreConfig): Promise<VectorStore>;
+declare function getVectorStore(): VectorStore | null;
+
+/**
+ * EmbeddingProvider Interface
+ *
+ * Abstraction over embedding model providers (Voyage AI, Ollama).
+ * Converts text into dense vector representations for similarity search.
+ */
+interface EmbeddingProvider {
+    /** Embed a single text string. */
+    embed(text: string): Promise<number[]>;
+    /** Embed multiple texts in a single batch call. */
+    embedBatch(texts: string[]): Promise<number[][]>;
+    /** The dimensionality of vectors produced by this provider. */
+    dimensions(): number;
+    /** The model identifier (e.g. "voyage-3", "nomic-embed-text"). */
+    model(): string;
+}
+
+/**
+ * Voyage AI Embedding Provider
+ *
+ * Cloud embedding via the Voyage AI API (partner of Anthropic).
+ * Uses the same API key as Anthropic inference.
+ */
+
+interface VoyageConfig {
+    apiKey: string;
+    model: string;
+    endpoint?: string;
+}
+declare class VoyageEmbeddingProvider implements EmbeddingProvider {
+    private config;
+    constructor(config: VoyageConfig);
+    embed(text: string): Promise<number[]>;
+    embedBatch(texts: string[]): Promise<number[][]>;
+    dimensions(): number;
+    model(): string;
+}
+
+/**
+ * Ollama Embedding Provider
+ *
+ * Local embedding via the Ollama API.
+ * Uses models like nomic-embed-text, all-minilm, etc.
+ */
+
+interface OllamaEmbeddingConfig {
+    model: string;
+    baseURL?: string;
+}
+declare class OllamaEmbeddingProvider implements EmbeddingProvider {
+    private config;
+    constructor(config: OllamaEmbeddingConfig);
+    embed(text: string): Promise<number[]>;
+    embedBatch(texts: string[]): Promise<number[][]>;
+    dimensions(): number;
+    model(): string;
+}
+
+/**
+ * EmbeddingProvider Factory
+ */
+
+interface EmbeddingConfig {
+    type: 'voyage' | 'ollama';
+    model: string;
+    apiKey?: string;
+    baseURL?: string;
+    endpoint?: string;
+}
+declare function createEmbeddingProvider(config: EmbeddingConfig): Promise<EmbeddingProvider>;
+
+/**
+ * Text Chunking Utilities
+ *
+ * Splits long text into overlapping chunks for embedding.
+ * Each chunk is a passage that fits within the embedding model's context window.
+ */
+interface ChunkingConfig {
+    chunkSize: number;
+    overlap: number;
+}
+declare const DEFAULT_CHUNKING_CONFIG: ChunkingConfig;
+/**
+ * Split text into overlapping chunks.
+ *
+ * Splits on paragraph boundaries when possible, falling back to sentence
+ * boundaries, then word boundaries. Each chunk overlaps with the previous
+ * by `overlap` tokens worth of text.
+ */
+declare function chunkText(text: string, config?: ChunkingConfig): string[];
+
+export { type AnnotationPayload, type ChunkingConfig, DEFAULT_CHUNKING_CONFIG, type EmbeddingChunk, type EmbeddingConfig, type EmbeddingProvider, MemoryVectorStore, type OllamaEmbeddingConfig, OllamaEmbeddingProvider, type QdrantConfig, QdrantVectorStore, type SearchOptions, type VectorSearchResult, type VectorStore, type VectorStoreConfig, type VoyageConfig, VoyageEmbeddingProvider, chunkText, createEmbeddingProvider, createVectorStore, getVectorStore };

package/dist/index.js

@@ -0,0 +1,209 @@
+import {
+  QdrantVectorStore
+} from "./chunk-GZRNY7LY.js";
+import {
+  VoyageEmbeddingProvider
+} from "./chunk-INCF7JMV.js";
+import {
+  OllamaEmbeddingProvider
+} from "./chunk-LBAPGZOW.js";
+
+// src/store/memory.ts
+function cosineSimilarity(a, b) {
+  let dotProduct = 0;
+  let normA = 0;
+  let normB = 0;
+  for (let i = 0; i < a.length; i++) {
+    dotProduct += a[i] * b[i];
+    normA += a[i] * a[i];
+    normB += b[i] * b[i];
+  }
+  const denom = Math.sqrt(normA) * Math.sqrt(normB);
+  return denom === 0 ? 0 : dotProduct / denom;
+}
+var MemoryVectorStore = class {
+  resources = [];
+  annotations = [];
+  connected = false;
+  async connect() {
+    this.connected = true;
+  }
+  async disconnect() {
+    this.connected = false;
+  }
+  isConnected() {
+    return this.connected;
+  }
+  async upsertResourceVectors(resourceId, chunks) {
+    this.resources = this.resources.filter((p) => p.payload.resourceId !== String(resourceId));
+    for (const chunk of chunks) {
+      this.resources.push({
+        id: `${resourceId}-${chunk.chunkIndex}`,
+        vector: chunk.embedding,
+        payload: {
+          resourceId: String(resourceId),
+          chunkIndex: chunk.chunkIndex,
+          text: chunk.text
+        }
+      });
+    }
+  }
+  async upsertAnnotationVector(annotationId, embedding, payload) {
+    this.annotations = this.annotations.filter((p) => p.id !== String(annotationId));
+    this.annotations.push({
+      id: String(annotationId),
+      vector: embedding,
+      payload: {
+        annotationId: String(payload.annotationId),
+        resourceId: String(payload.resourceId),
+        motivation: payload.motivation,
+        entityTypes: payload.entityTypes,
+        text: payload.exactText
+      }
+    });
+  }
+  async deleteResourceVectors(resourceId) {
+    this.resources = this.resources.filter((p) => p.payload.resourceId !== String(resourceId));
+  }
+  async deleteAnnotationVector(annotationId) {
+    this.annotations = this.annotations.filter((p) => p.id !== String(annotationId));
+  }
+  async searchResources(embedding, opts) {
+    return this.search(this.resources, embedding, opts);
+  }
+  async searchAnnotations(embedding, opts) {
+    return this.search(this.annotations, embedding, opts);
+  }
+  search(points, embedding, opts) {
+    let filtered = points;
+    if (opts.filter) {
+      const f = opts.filter;
+      filtered = points.filter((p) => {
+        if (f.resourceId && p.payload.resourceId !== String(f.resourceId)) return false;
+        if (f.excludeResourceId && p.payload.resourceId === String(f.excludeResourceId)) return false;
+        if (f.motivation && p.payload.motivation !== f.motivation) return false;
+        if (f.entityTypes && f.entityTypes.length > 0) {
+          const pTypes = p.payload.entityTypes ?? [];
+          if (!f.entityTypes.some((t) => pTypes.includes(t))) return false;
+        }
+        return true;
+      });
+    }
+    const scored = filtered.map((p) => ({
+      ...p,
+      score: cosineSimilarity(embedding, p.vector)
+    }));
+    scored.sort((a, b) => b.score - a.score);
+    if (opts.scoreThreshold) {
+      const threshold = opts.scoreThreshold;
+      return scored.filter((s) => s.score >= threshold).slice(0, opts.limit).map((s) => this.toResult(s));
+    }
+    return scored.slice(0, opts.limit).map((s) => this.toResult(s));
+  }
+  toResult(s) {
+    return {
+      id: s.id,
+      score: s.score,
+      resourceId: s.payload.resourceId,
+      annotationId: s.payload.annotationId,
+      text: s.payload.text,
+      entityTypes: s.payload.entityTypes
+    };
+  }
+};
+
+// src/store/factory.ts
+var instance = null;
+async function createVectorStore(config) {
+  if (instance) return instance;
+  if (config.type === "qdrant") {
+    const { QdrantVectorStore: QdrantVectorStore2 } = await import("./qdrant-JXH3K3RP.js");
+    instance = new QdrantVectorStore2({
+      host: config.host ?? "localhost",
+      port: config.port ?? 6333,
+      dimensions: config.dimensions
+    });
+  } else {
+    instance = new MemoryVectorStore();
+  }
+  await instance.connect();
+  return instance;
+}
+function getVectorStore() {
+  return instance;
+}
+
+// src/embedding/factory.ts
+async function createEmbeddingProvider(config) {
+  if (config.type === "voyage") {
+    const { VoyageEmbeddingProvider: VoyageEmbeddingProvider2 } = await import("./voyage-UWOFVIYM.js");
+    if (!config.apiKey) throw new Error("apiKey is required for Voyage embedding provider");
+    return new VoyageEmbeddingProvider2({
+      apiKey: config.apiKey,
+      model: config.model,
+      endpoint: config.endpoint
+    });
+  }
+  if (config.type === "ollama") {
+    const { OllamaEmbeddingProvider: OllamaEmbeddingProvider2 } = await import("./ollama-H62PXJ5D.js");
+    return new OllamaEmbeddingProvider2({
+      model: config.model,
+      baseURL: config.baseURL
+    });
+  }
+  throw new Error(`Unknown embedding provider type: ${config.type}`);
+}
+
+// src/chunking.ts
+var DEFAULT_CHUNKING_CONFIG = {
+  chunkSize: 512,
+  overlap: 64
+};
+function estimateTokens(text) {
+  return Math.ceil(text.length / 4);
+}
+function chunkText(text, config = DEFAULT_CHUNKING_CONFIG) {
+  const totalTokens = estimateTokens(text);
+  if (totalTokens <= config.chunkSize) {
+    return [text];
+  }
+  const chunkChars = config.chunkSize * 4;
+  const overlapChars = config.overlap * 4;
+  const chunks = [];
+  let start = 0;
+  while (start < text.length) {
+    let end = Math.min(start + chunkChars, text.length);
+    if (end < text.length) {
+      const paraBreak = text.lastIndexOf("\n\n", end);
+      if (paraBreak > start + chunkChars / 2) {
+        end = paraBreak;
+      } else {
+        const sentenceBreak = text.lastIndexOf(". ", end);
+        if (sentenceBreak > start + chunkChars / 2) {
+          end = sentenceBreak + 1;
+        } else {
+          const wordBreak = text.lastIndexOf(" ", end);
+          if (wordBreak > start + chunkChars / 2) {
+            end = wordBreak;
+          }
+        }
+      }
+    }
+    chunks.push(text.slice(start, end).trim());
+    start = end - overlapChars;
+    if (start >= text.length) break;
+  }
+  return chunks.filter((c) => c.length > 0);
+}
+export {
+  DEFAULT_CHUNKING_CONFIG,
+  MemoryVectorStore,
+  OllamaEmbeddingProvider,
+  QdrantVectorStore,
+  VoyageEmbeddingProvider,
+  chunkText,
+  createEmbeddingProvider,
+  createVectorStore,
+  getVectorStore
+};
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/store/memory.ts","../src/store/factory.ts","../src/embedding/factory.ts","../src/chunking.ts"],"sourcesContent":["/**\n * In-Memory VectorStore Implementation\n *\n * For testing and development without a running Qdrant instance.\n * Uses brute-force cosine similarity search.\n */\n\nimport type { ResourceId, AnnotationId } from '@semiont/core';\nimport type { VectorStore, EmbeddingChunk, AnnotationPayload, VectorSearchResult, SearchOptions } from './interface';\n\ninterface StoredPoint {\n  id: string;\n  vector: number[];\n  payload: {\n    resourceId: string;\n    annotationId?: string;\n    chunkIndex?: number;\n    text: string;\n    motivation?: string;\n    entityTypes?: string[];\n  };\n}\n\nfunction cosineSimilarity(a: number[], b: number[]): number {\n  let dotProduct = 0;\n  let normA = 0;\n  let normB = 0;\n  for (let i = 0; i < a.length; i++) {\n    dotProduct += a[i] * b[i];\n    normA += a[i] * a[i];\n    normB += b[i] * b[i];\n  }\n  const denom = Math.sqrt(normA) * Math.sqrt(normB);\n  return denom === 0 ? 0 : dotProduct / denom;\n}\n\nexport class MemoryVectorStore implements VectorStore {\n  private resources: StoredPoint[] = [];\n  private annotations: StoredPoint[] = [];\n  private connected = false;\n\n  async connect(): Promise<void> {\n    this.connected = true;\n  }\n\n  async disconnect(): Promise<void> {\n    this.connected = false;\n  }\n\n  isConnected(): boolean {\n    return this.connected;\n  }\n\n  async upsertResourceVectors(resourceId: ResourceId, chunks: EmbeddingChunk[]): Promise<void> {\n    // Remove existing vectors for this resource\n    this.resources = this.resources.filter(p => p.payload.resourceId !== String(resourceId));\n\n    for (const chunk of chunks) {\n      this.resources.push({\n        id: `${resourceId}-${chunk.chunkIndex}`,\n        vector: chunk.embedding,\n        payload: {\n          resourceId: String(resourceId),\n          chunkIndex: chunk.chunkIndex,\n          text: chunk.text,\n        },\n      });\n    }\n  }\n\n  async upsertAnnotationVector(\n    annotationId: AnnotationId,\n    embedding: number[],\n    payload: AnnotationPayload\n  ): Promise<void> {\n    this.annotations = this.annotations.filter(p => p.id !== String(annotationId));\n    this.annotations.push({\n      id: String(annotationId),\n      vector: embedding,\n      payload: {\n        annotationId: String(payload.annotationId),\n        resourceId: String(payload.resourceId),\n        motivation: payload.motivation,\n        entityTypes: payload.entityTypes,\n        text: payload.exactText,\n      },\n    });\n  }\n\n  async deleteResourceVectors(resourceId: ResourceId): Promise<void> {\n    this.resources = this.resources.filter(p => p.payload.resourceId !== String(resourceId));\n  }\n\n  async deleteAnnotationVector(annotationId: AnnotationId): Promise<void> {\n    this.annotations = this.annotations.filter(p => p.id !== String(annotationId));\n  }\n\n  async searchResources(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]> {\n    return this.search(this.resources, embedding, opts);\n  }\n\n  async searchAnnotations(embedding: number[], opts: SearchOptions): Promise<VectorSearchResult[]> {\n    return this.search(this.annotations, embedding, opts);\n  }\n\n  private search(points: StoredPoint[], embedding: number[], opts: SearchOptions): VectorSearchResult[] {\n    let filtered = points;\n\n    if (opts.filter) {\n      const f = opts.filter;\n      filtered = points.filter(p => {\n        if (f.resourceId && p.payload.resourceId !== String(f.resourceId)) return false;\n        if (f.excludeResourceId && p.payload.resourceId === String(f.excludeResourceId)) return false;\n        if (f.motivation && p.payload.motivation !== f.motivation) return false;\n        if (f.entityTypes && f.entityTypes.length > 0) {\n          const pTypes = p.payload.entityTypes ?? [];\n          if (!f.entityTypes.some(t => pTypes.includes(t))) return false;\n        }\n        return true;\n      });\n    }\n\n    const scored = filtered.map(p => ({\n      ...p,\n      score: cosineSimilarity(embedding, p.vector),\n    }));\n\n    scored.sort((a, b) => b.score - a.score);\n\n    if (opts.scoreThreshold) {\n      const threshold = opts.scoreThreshold;\n      return scored\n        .filter(s => s.score >= threshold)\n        .slice(0, opts.limit)\n        .map(s => this.toResult(s));\n    }\n\n    return scored.slice(0, opts.limit).map(s => this.toResult(s));\n  }\n\n  private toResult(s: StoredPoint & { score: number }): VectorSearchResult {\n    return {\n      id: s.id,\n      score: s.score,\n      resourceId: s.payload.resourceId as ResourceId,\n      annotationId: s.payload.annotationId as AnnotationId | undefined,\n      text: s.payload.text,\n      entityTypes: s.payload.entityTypes,\n    };\n  }\n}\n","/**\n * VectorStore Factory\n *\n * Creates a VectorStore instance based on configuration.\n */\n\nimport type { VectorStore } from './interface';\nimport { MemoryVectorStore } from './memory';\n\nexport interface VectorStoreConfig {\n  type: 'qdrant' | 'memory';\n  host?: string;\n  port?: number;\n  dimensions: number;\n}\n\nlet instance: VectorStore | null = null;\n\nexport async function createVectorStore(config: VectorStoreConfig): Promise<VectorStore> {\n  if (instance) return instance;\n\n  if (config.type === 'qdrant') {\n    const { QdrantVectorStore } = await import('./qdrant');\n    instance = new QdrantVectorStore({\n      host: config.host ?? 'localhost',\n      port: config.port ?? 6333,\n      dimensions: config.dimensions,\n    });\n  } else {\n    instance = new MemoryVectorStore();\n  }\n\n  await instance.connect();\n  return instance;\n}\n\nexport function getVectorStore(): VectorStore | null {\n  return instance;\n}\n","/**\n * EmbeddingProvider Factory\n */\n\nimport type { EmbeddingProvider } from './interface';\n\nexport interface EmbeddingConfig {\n  type: 'voyage' | 'ollama';\n  model: string;\n  apiKey?: string;\n  baseURL?: string;\n  endpoint?: string;\n}\n\nexport async function createEmbeddingProvider(config: EmbeddingConfig): Promise<EmbeddingProvider> {\n  if (config.type === 'voyage') {\n    const { VoyageEmbeddingProvider } = await import('./voyage');\n    if (!config.apiKey) throw new Error('apiKey is required for Voyage embedding provider');\n    return new VoyageEmbeddingProvider({\n      apiKey: config.apiKey,\n      model: config.model,\n      endpoint: config.endpoint,\n    });\n  }\n\n  if (config.type === 'ollama') {\n    const { OllamaEmbeddingProvider } = await import('./ollama');\n    return new OllamaEmbeddingProvider({\n      model: config.model,\n      baseURL: config.baseURL,\n    });\n  }\n\n  throw new Error(`Unknown embedding provider type: ${config.type}`);\n}\n","/**\n * Text Chunking Utilities\n *\n * Splits long text into overlapping chunks for embedding.\n * Each chunk is a passage that fits within the embedding model's context window.\n */\n\nexport interface ChunkingConfig {\n  chunkSize: number;   // approximate tokens per chunk\n  overlap: number;     // tokens of overlap between adjacent chunks\n}\n\nexport const DEFAULT_CHUNKING_CONFIG: ChunkingConfig = {\n  chunkSize: 512,\n  overlap: 64,\n};\n\n/**\n * Rough token count estimate: ~4 characters per token for English text.\n */\nfunction estimateTokens(text: string): number {\n  return Math.ceil(text.length / 4);\n}\n\n/**\n * Split text into overlapping chunks.\n *\n * Splits on paragraph boundaries when possible, falling back to sentence\n * boundaries, then word boundaries. Each chunk overlaps with the previous\n * by `overlap` tokens worth of text.\n */\nexport function chunkText(text: string, config: ChunkingConfig = DEFAULT_CHUNKING_CONFIG): string[] {\n  const totalTokens = estimateTokens(text);\n  if (totalTokens <= config.chunkSize) {\n    return [text];\n  }\n\n  const chunkChars = config.chunkSize * 4;\n  const overlapChars = config.overlap * 4;\n  const chunks: string[] = [];\n  let start = 0;\n\n  while (start < text.length) {\n    let end = Math.min(start + chunkChars, text.length);\n\n    // Try to break at a paragraph boundary\n    if (end < text.length) {\n      const paraBreak = text.lastIndexOf('\\n\\n', end);\n      if (paraBreak > start + chunkChars / 2) {\n        end = paraBreak;\n      } else {\n        // Try sentence boundary\n        const sentenceBreak = text.lastIndexOf('. ', end);\n        if (sentenceBreak > start + chunkChars / 2) {\n          end = sentenceBreak + 1;\n        } else {\n          // Try word boundary\n          const wordBreak = text.lastIndexOf(' ', end);\n          if (wordBreak > start + chunkChars / 2) {\n            end = wordBreak;\n          }\n        }\n      }\n    }\n\n    chunks.push(text.slice(start, end).trim());\n    start = end - overlapChars;\n    if (start >= text.length) break;\n  }\n\n  return chunks.filter(c => c.length > 0);\n}\n"],"mappings":";;;;;;;;;;;AAuBA,SAAS,iBAAiB,GAAa,GAAqB;AAC1D,MAAI,aAAa;AACjB,MAAI,QAAQ;AACZ,MAAI,QAAQ;AACZ,WAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,kBAAc,EAAE,CAAC,IAAI,EAAE,CAAC;AACxB,aAAS,EAAE,CAAC,IAAI,EAAE,CAAC;AACnB,aAAS,EAAE,CAAC,IAAI,EAAE,CAAC;AAAA,EACrB;AACA,QAAM,QAAQ,KAAK,KAAK,KAAK,IAAI,KAAK,KAAK,KAAK;AAChD,SAAO,UAAU,IAAI,IAAI,aAAa;AACxC;AAEO,IAAM,oBAAN,MAA+C;AAAA,EAC5C,YAA2B,CAAC;AAAA,EAC5B,cAA6B,CAAC;AAAA,EAC9B,YAAY;AAAA,EAEpB,MAAM,UAAyB;AAC7B,SAAK,YAAY;AAAA,EACnB;AAAA,EAEA,MAAM,aAA4B;AAChC,SAAK,YAAY;AAAA,EACnB;AAAA,EAEA,cAAuB;AACrB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAM,sBAAsB,YAAwB,QAAyC;AAE3F,SAAK,YAAY,KAAK,UAAU,OAAO,OAAK,EAAE,QAAQ,eAAe,OAAO,UAAU,CAAC;AAEvF,eAAW,SAAS,QAAQ;AAC1B,WAAK,UAAU,KAAK;AAAA,QAClB,IAAI,GAAG,UAAU,IAAI,MAAM,UAAU;AAAA,QACrC,QAAQ,MAAM;AAAA,QACd,SAAS;AAAA,UACP,YAAY,OAAO,UAAU;AAAA,UAC7B,YAAY,MAAM;AAAA,UAClB,MAAM,MAAM;AAAA,QACd;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AAAA,EAEA,MAAM,uBACJ,cACA,WACA,SACe;AACf,SAAK,cAAc,KAAK,YAAY,OAAO,OAAK,EAAE,OAAO,OAAO,YAAY,CAAC;AAC7E,SAAK,YAAY,KAAK;AAAA,MACpB,IAAI,OAAO,YAAY;AAAA,MACvB,QAAQ;AAAA,MACR,SAAS;AAAA,QACP,cAAc,OAAO,QAAQ,YAAY;AAAA,QACzC,YAAY,OAAO,QAAQ,UAAU;AAAA,QACrC,YAAY,QAAQ;AAAA,QACpB,aAAa,QAAQ;AAAA,QACrB,MAAM,QAAQ;AAAA,MAChB;AAAA,IACF,CAAC;AAAA,EACH;AAAA,EAEA,MAAM,sBAAsB,YAAuC;AACjE,SAAK,YAAY,KAAK,UAAU,OAAO,OAAK,EAAE,QAAQ,eAAe,OAAO,UAAU,CAAC;AAAA,EACzF;AAAA,EAEA,MAAM,uBAAuB,cAA2C;AACtE,SAAK,cAAc,KAAK,YAAY,OAAO,OAAK,EAAE,OAAO,OAAO,YAAY,CAAC;AAAA,EAC/E;AAAA,EAEA,MAAM,gBAAgB,WAAqB,MAAoD;AAC7F,WAAO,KAAK,OAAO,KAAK,WAAW,WAAW,IAAI;AAAA,EACpD;AAAA,EAEA,MAAM,kBAAkB,WAAqB,MAAoD;AAC/F,WAAO,KAAK,OAAO,KAAK,aAAa,WAAW,IAAI;AAAA,EACtD;AAAA,EAEQ,OAAO,QAAuB,WAAqB,MAA2C;AACpG,QAAI,WAAW;AAEf,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,KAAK;AACf,iBAAW,OAAO,OAAO,OAAK;AAC5B,YAAI,EAAE,cAAc,EAAE,QAAQ,eAAe,OAAO,EAAE,UAAU,EAAG,QAAO;AAC1E,YAAI,EAAE,qBAAqB,EAAE,QAAQ,eAAe,OAAO,EAAE,iBAAiB,EAAG,QAAO;AACxF,YAAI,EAAE,cAAc,EAAE,QAAQ,eAAe,EAAE,WAAY,QAAO;AAClE,YAAI,EAAE,eAAe,EAAE,YAAY,SAAS,GAAG;AAC7C,gBAAM,SAAS,EAAE,QAAQ,eAAe,CAAC;AACzC,cAAI,CAAC,EAAE,YAAY,KAAK,OAAK,OAAO,SAAS,CAAC,CAAC,EAAG,QAAO;AAAA,QAC3D;AACA,eAAO;AAAA,MACT,CAAC;AAAA,IACH;AAEA,UAAM,SAAS,SAAS,IAAI,QAAM;AAAA,MAChC,GAAG;AAAA,MACH,OAAO,iBAAiB,WAAW,EAAE,MAAM;AAAA,IAC7C,EAAE;AAEF,WAAO,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;AAEvC,QAAI,KAAK,gBAAgB;AACvB,YAAM,YAAY,KAAK;AACvB,aAAO,OACJ,OAAO,OAAK,EAAE,SAAS,SAAS,EAChC,MAAM,GAAG,KAAK,KAAK,EACnB,IAAI,OAAK,KAAK,SAAS,CAAC,CAAC;AAAA,IAC9B;AAEA,WAAO,OAAO,MAAM,GAAG,KAAK,KAAK,EAAE,IAAI,OAAK,KAAK,SAAS,CAAC,CAAC;AAAA,EAC9D;AAAA,EAEQ,SAAS,GAAwD;AACvE,WAAO;AAAA,MACL,IAAI,EAAE;AAAA,MACN,OAAO,EAAE;AAAA,MACT,YAAY,EAAE,QAAQ;AAAA,MACtB,cAAc,EAAE,QAAQ;AAAA,MACxB,MAAM,EAAE,QAAQ;AAAA,MAChB,aAAa,EAAE,QAAQ;AAAA,IACzB;AAAA,EACF;AACF;;;ACtIA,IAAI,WAA+B;AAEnC,eAAsB,kBAAkB,QAAiD;AACvF,MAAI,SAAU,QAAO;AAErB,MAAI,OAAO,SAAS,UAAU;AAC5B,UAAM,EAAE,mBAAAA,mBAAkB,IAAI,MAAM,OAAO,sBAAU;AACrD,eAAW,IAAIA,mBAAkB;AAAA,MAC/B,MAAM,OAAO,QAAQ;AAAA,MACrB,MAAM,OAAO,QAAQ;AAAA,MACrB,YAAY,OAAO;AAAA,IACrB,CAAC;AAAA,EACH,OAAO;AACL,eAAW,IAAI,kBAAkB;AAAA,EACnC;AAEA,QAAM,SAAS,QAAQ;AACvB,SAAO;AACT;AAEO,SAAS,iBAAqC;AACnD,SAAO;AACT;;;ACxBA,eAAsB,wBAAwB,QAAqD;AACjG,MAAI,OAAO,SAAS,UAAU;AAC5B,UAAM,EAAE,yBAAAC,yBAAwB,IAAI,MAAM,OAAO,sBAAU;AAC3D,QAAI,CAAC,OAAO,OAAQ,OAAM,IAAI,MAAM,kDAAkD;AACtF,WAAO,IAAIA,yBAAwB;AAAA,MACjC,QAAQ,OAAO;AAAA,MACf,OAAO,OAAO;AAAA,MACd,UAAU,OAAO;AAAA,IACnB,CAAC;AAAA,EACH;AAEA,MAAI,OAAO,SAAS,UAAU;AAC5B,UAAM,EAAE,yBAAAC,yBAAwB,IAAI,MAAM,OAAO,sBAAU;AAC3D,WAAO,IAAIA,yBAAwB;AAAA,MACjC,OAAO,OAAO;AAAA,MACd,SAAS,OAAO;AAAA,IAClB,CAAC;AAAA,EACH;AAEA,QAAM,IAAI,MAAM,oCAAoC,OAAO,IAAI,EAAE;AACnE;;;ACtBO,IAAM,0BAA0C;AAAA,EACrD,WAAW;AAAA,EACX,SAAS;AACX;AAKA,SAAS,eAAe,MAAsB;AAC5C,SAAO,KAAK,KAAK,KAAK,SAAS,CAAC;AAClC;AASO,SAAS,UAAU,MAAc,SAAyB,yBAAmC;AAClG,QAAM,cAAc,eAAe,IAAI;AACvC,MAAI,eAAe,OAAO,WAAW;AACnC,WAAO,CAAC,IAAI;AAAA,EACd;AAEA,QAAM,aAAa,OAAO,YAAY;AACtC,QAAM,eAAe,OAAO,UAAU;AACtC,QAAM,SAAmB,CAAC;AAC1B,MAAI,QAAQ;AAEZ,SAAO,QAAQ,KAAK,QAAQ;AAC1B,QAAI,MAAM,KAAK,IAAI,QAAQ,YAAY,KAAK,MAAM;AAGlD,QAAI,MAAM,KAAK,QAAQ;AACrB,YAAM,YAAY,KAAK,YAAY,QAAQ,GAAG;AAC9C,UAAI,YAAY,QAAQ,aAAa,GAAG;AACtC,cAAM;AAAA,MACR,OAAO;AAEL,cAAM,gBAAgB,KAAK,YAAY,MAAM,GAAG;AAChD,YAAI,gBAAgB,QAAQ,aAAa,GAAG;AAC1C,gBAAM,gBAAgB;AAAA,QACxB,OAAO;AAEL,gBAAM,YAAY,KAAK,YAAY,KAAK,GAAG;AAC3C,cAAI,YAAY,QAAQ,aAAa,GAAG;AACtC,kBAAM;AAAA,UACR;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAEA,WAAO,KAAK,KAAK,MAAM,OAAO,GAAG,EAAE,KAAK,CAAC;AACzC,YAAQ,MAAM;AACd,QAAI,SAAS,KAAK,OAAQ;AAAA,EAC5B;AAEA,SAAO,OAAO,OAAO,OAAK,EAAE,SAAS,CAAC;AACxC;","names":["QdrantVectorStore","VoyageEmbeddingProvider","OllamaEmbeddingProvider"]}

package/dist/ollama-H62PXJ5D.js

@@ -0,0 +1,7 @@
+import {
+  OllamaEmbeddingProvider
+} from "./chunk-LBAPGZOW.js";
+export {
+  OllamaEmbeddingProvider
+};
+//# sourceMappingURL=ollama-H62PXJ5D.js.map

package/dist/ollama-H62PXJ5D.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}

package/dist/qdrant-JXH3K3RP.js

@@ -0,0 +1,7 @@
+import {
+  QdrantVectorStore
+} from "./chunk-GZRNY7LY.js";
+export {
+  QdrantVectorStore
+};
+//# sourceMappingURL=qdrant-JXH3K3RP.js.map

package/dist/qdrant-JXH3K3RP.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}

package/dist/voyage-UWOFVIYM.js

@@ -0,0 +1,7 @@
+import {
+  VoyageEmbeddingProvider
+} from "./chunk-INCF7JMV.js";
+export {
+  VoyageEmbeddingProvider
+};
+//# sourceMappingURL=voyage-UWOFVIYM.js.map

package/dist/voyage-UWOFVIYM.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}

package/package.json

@@ -0,0 +1,52 @@
+{
+  "name": "@semiont/vectors",
+  "version": "0.4.12",
+  "type": "module",
+  "description": "Vector storage, embedding, and semantic search for Semiont",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js",
+      "default": "./dist/index.js"
+    }
+  },
+  "scripts": {
+    "build": "npm run typecheck && tsup",
+    "typecheck": "tsc --noEmit",
+    "clean": "rm -rf dist",
+    "test": "vitest run",
+    "test:watch": "vitest"
+  },
+  "dependencies": {
+    "@semiont/core": "*"
+  },
+  "peerDependencies": {
+    "@qdrant/js-client-rest": ">=1.0.0"
+  },
+  "peerDependenciesMeta": {
+    "@qdrant/js-client-rest": {
+      "optional": true
+    }
+  },
+  "devDependencies": {
+    "@qdrant/js-client-rest": "^1.13.0",
+    "tsup": "^8.5.1",
+    "typescript": "^5.6.3"
+  },
+  "files": [
+    "dist",
+    "README.md",
+    "LICENSE"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/The-AI-Alliance/semiont.git",
+    "directory": "packages/vectors"
+  },
+  "license": "Apache-2.0"
+}