@anvia/lancedb 0.2.4 → 0.2.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/dist/index.js +10 -6
  2. package/dist/index.js.map +1 -1
  3. package/package.json +2 -2
package/dist/index.js CHANGED
@@ -81,13 +81,14 @@ function lanceRows(document) {
81
81
  assertNoReservedMetadata(document.metadata);
82
82
  return document.embeddings.map((embedding, index) => {
83
83
  const logicalId = document.embeddings.length === 1 ? document.id : `${document.id}#embedding:${index}`;
84
- return {
84
+ const row = {
85
85
  _rowid: rowId(logicalId),
86
86
  [documentIdColumn]: document.id,
87
87
  [documentColumn]: serializeDocument(document.document),
88
- [vectorColumn]: embedding.vector,
89
- ...document.metadata ?? {}
88
+ [vectorColumn]: embedding.vector
90
89
  };
90
+ Object.assign(row, document.metadata);
91
+ return row;
91
92
  });
92
93
  }
93
94
  function parseQueryResults(response, threshold) {
@@ -103,9 +104,12 @@ function parseQueryResults(response, threshold) {
103
104
  const result = {
104
105
  id,
105
106
  score,
106
- document: parseDocument(record[documentColumn]),
107
- ...metadataFromColumns(record)
107
+ document: parseDocument(record[documentColumn])
108
108
  };
109
+ const metadata = metadataFromColumns(record);
110
+ if (metadata !== void 0) {
111
+ result.metadata = metadata;
112
+ }
109
113
  const current = byId.get(id);
110
114
  if (current === void 0 || result.score > current.score) {
111
115
  byId.set(id, result);
@@ -125,7 +129,7 @@ function metadataFromColumns(record) {
125
129
  ([key]) => !key.startsWith(reservedColumnPrefix) && key !== vectorColumn && key !== "_rowid" && key !== "_distance"
126
130
  )
127
131
  );
128
- return Object.keys(metadata).length === 0 ? {} : { metadata };
132
+ return Object.keys(metadata).length === 0 ? void 0 : metadata;
129
133
  }
130
134
 
131
135
  // src/search-index.ts
package/dist/index.js.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"sources":["../src/filters.ts","../src/search-index.ts","../src/helpers.ts","../src/types.ts","../src/store.ts"],"sourcesContent":["import type { VectorFilter } from \"@anvia/core/vector-store\";\n\nexport function filterToLanceExpr(filter: VectorFilter | undefined): string | undefined {\n if (filter === undefined) {\n return undefined;\n }\n\n return translateFilter(filter);\n}\n\nfunction translateFilter(filter: VectorFilter): string {\n switch (filter.type) {\n case \"eq\":\n if (typeof filter.value === \"string\") {\n return `${filter.key} = '${escapeSql(filter.value)}'`;\n }\n if (typeof filter.value === \"boolean\") {\n return `${filter.key} = ${filter.value ? \"TRUE\" : \"FALSE\"}`;\n }\n if (filter.value === null) {\n return `${filter.key} IS NULL`;\n }\n return `${filter.key} = ${filter.value}`;\n case \"gt\":\n return `${filter.key} > ${filter.value}`;\n case \"lt\":\n return `${filter.key} < ${filter.value}`;\n case \"and\":\n return filter.filters.map((f) => `(${translateFilter(f)})`).join(\" AND \");\n case \"or\":\n return filter.filters.map((f) => `(${translateFilter(f)})`).join(\" OR \");\n }\n}\n\nfunction escapeSql(value: string): string {\n return value.replace(/'/g, \"''\");\n}\n","import { type EmbeddingModel, embedText, type VectorMetadata } from \"@anvia/core/embeddings\";\nimport type { Tool } from \"@anvia/core/tool\";\nimport {\n createVectorSearchTool,\n type VectorSearchIndex,\n type VectorSearchRequest,\n type VectorSearchResult,\n type VectorSearchToolOptions,\n} from \"@anvia/core/vector-store\";\nimport { filterToLanceExpr } from \"./filters.js\";\nimport { parseQueryResults } from \"./helpers.js\";\nimport type { LanceDBConnectionLike } from \"./types.js\";\n\nexport class LanceDBVectorIndex<T, Metadata extends VectorMetadata = VectorMetadata>\n implements VectorSearchIndex<T, Metadata>\n{\n constructor(\n private readonly model: EmbeddingModel,\n private readonly connection: LanceDBConnectionLike,\n private readonly tableName: string,\n ) {}\n\n async search(request: VectorSearchRequest): Promise<Array<VectorSearchResult<T, Metadata>>> {\n const queryEmbedding = await embedText(this.model, request.query);\n const table = await this.connection.openTable(this.tableName);\n const filterExpr = filterToLanceExpr(request.filter);\n\n const query = table.search(queryEmbedding.vector).limit(request.topK);\n\n let response: unknown[];\n if (filterExpr !== undefined) {\n response = (await query.filter(filterExpr)) as unknown[];\n } else {\n response = (await query.toArray()) as unknown[];\n }\n\n return parseQueryResults<T, Metadata>(response, request.threshold);\n }\n\n async searchIds(request: VectorSearchRequest): Promise<Array<{ score: number; id: string }>> {\n return (await this.search(request)).map(({ score, id }) => ({ score, id }));\n }\n\n asTool(options: VectorSearchToolOptions): Tool<{ query: string; topK?: number }, unknown> {\n return createVectorSearchTool(this, options);\n }\n}\n","import { createHash } from \"node:crypto\";\nimport type { EmbeddedDocument, VectorMetadata } from \"@anvia/core/embeddings\";\nimport type { VectorSearchResult } from \"@anvia/core/vector-store\";\nimport { documentColumn, documentIdColumn, reservedColumnPrefix, vectorColumn } from \"./types.js\";\n\nexport function assertNoReservedMetadata(metadata: VectorMetadata | undefined): void {\n for (const key of Object.keys(metadata ?? {})) {\n if (key.startsWith(reservedColumnPrefix)) {\n throw new Error(`Metadata key ${key} is reserved for Anvia LanceDB columns`);\n }\n }\n}\n\nexport function rowId(id: string): string {\n return createHash(\"sha256\").update(id).digest(\"hex\").slice(0, 32);\n}\n\nexport function serializeDocument(document: unknown): string {\n return typeof document === \"string\" ? document : JSON.stringify(document);\n}\n\nexport function parseDocument<T>(document: unknown): T {\n if (document === null || document === undefined) {\n return \"\" as T;\n }\n if (typeof document !== \"string\") {\n return document as T;\n }\n try {\n return JSON.parse(document) as T;\n } catch {\n return document as T;\n }\n}\n\nexport function lanceRows<T, Metadata extends VectorMetadata>(\n document: EmbeddedDocument<T, Metadata>,\n): Array<Record<string, unknown>> {\n if (document.embeddings.length === 0) {\n throw new Error(`Document ${document.id} has no embeddings`);\n }\n assertNoReservedMetadata(document.metadata);\n\n return document.embeddings.map((embedding, index) => {\n const logicalId =\n document.embeddings.length === 1 ? document.id : `${document.id}#embedding:${index}`;\n return {\n _rowid: rowId(logicalId),\n [documentIdColumn]: document.id,\n [documentColumn]: serializeDocument(document.document),\n [vectorColumn]: embedding.vector,\n ...(document.metadata ?? {}),\n };\n });\n}\n\nexport function parseQueryResults<T, Metadata extends VectorMetadata>(\n response: unknown[],\n threshold: number | undefined,\n): Array<VectorSearchResult<T, Metadata>> {\n const byId = new Map<string, VectorSearchResult<T, Metadata>>();\n\n for (const row of response) {\n const record = row as Record<string, unknown>;\n const distance = typeof record._distance === \"number\" ? record._distance : 0;\n const score = 1 - distance;\n\n if (threshold !== undefined && score < threshold) {\n continue;\n }\n\n const id = String(record[documentIdColumn] ?? \"\");\n const result = {\n id,\n score,\n document: parseDocument(record[documentColumn]),\n ...metadataFromColumns<Metadata>(record),\n } as VectorSearchResult<T, Metadata>;\n const current = byId.get(id);\n if (current === undefined || result.score > current.score) {\n byId.set(id, result);\n }\n }\n\n return [...byId.values()];\n}\n\nexport async function defaultLanceDBConnection(\n uri?: string,\n): Promise<import(\"./types.js\").LanceDBConnectionLike> {\n const lancedb = await import(\"@lancedb/lancedb\");\n return lancedb.connect(\n uri ?? \"~/.anvia/lancedb\",\n ) as unknown as import(\"./types.js\").LanceDBConnectionLike;\n}\n\nfunction metadataFromColumns<Metadata extends VectorMetadata>(\n record: Record<string, unknown>,\n): { metadata?: Metadata | undefined } {\n const metadata = Object.fromEntries(\n Object.entries(record).filter(\n ([key]) =>\n !key.startsWith(reservedColumnPrefix) &&\n key !== vectorColumn &&\n key !== \"_rowid\" &&\n key !== \"_distance\",\n ),\n ) as Metadata;\n return Object.keys(metadata).length === 0 ? {} : { metadata };\n}\n","export const documentIdColumn = \"__anvia_document_id\";\nexport const documentColumn = \"__anvia_document\";\nexport const vectorColumn = \"__anvia_vector\";\nexport const reservedColumnPrefix = \"__anvia_\";\n\nexport type LanceDBDistance = \"cosine\" | \"l2\" | \"dot\";\n\nexport type LanceDBTableLike = {\n add(rows: Record<string, unknown>[]): Promise<unknown>;\n search(vector: number[]): {\n limit(n: number): {\n filter(expr: string | undefined): Promise<unknown[]>;\n toArray(): Promise<unknown[]>;\n };\n };\n countRows(): Promise<number>;\n delete(filter: string): Promise<unknown>;\n};\n\nexport type LanceDBConnectionLike = {\n openTable(name: string): Promise<LanceDBTableLike>;\n tableNames(): Promise<string[]>;\n createTable(name: string, data: Record<string, unknown>[]): Promise<LanceDBTableLike>;\n};\n\nexport type LanceDBVectorStoreConnectOptions = {\n client?: LanceDBConnectionLike | undefined;\n uri?: string | undefined;\n tableName: string;\n vectorSize: number;\n createIfMissing?: boolean | undefined;\n distance?: LanceDBDistance | undefined;\n};\n","import type { EmbeddedDocument, EmbeddingModel, VectorMetadata } from \"@anvia/core/embeddings\";\nimport { defaultLanceDBConnection, lanceRows } from \"./helpers.js\";\nimport { LanceDBVectorIndex } from \"./search-index.js\";\nimport type {\n LanceDBConnectionLike,\n LanceDBDistance,\n LanceDBVectorStoreConnectOptions,\n} from \"./types.js\";\n\nexport class LanceDBVectorStore<T, Metadata extends VectorMetadata = VectorMetadata> {\n private constructor(\n private readonly connection: LanceDBConnectionLike,\n private readonly tableName: string,\n private readonly distance: LanceDBDistance,\n private readonly vectorSize: number,\n ) {}\n\n static async connect<T, Metadata extends VectorMetadata = VectorMetadata>(\n options: LanceDBVectorStoreConnectOptions,\n ): Promise<LanceDBVectorStore<T, Metadata>> {\n const connection = options.client ?? (await defaultLanceDBConnection(options.uri));\n const distance = options.distance ?? \"cosine\";\n\n if (options.createIfMissing === false) {\n await connection.openTable(options.tableName);\n return new LanceDBVectorStore<T, Metadata>(\n connection,\n options.tableName,\n distance,\n options.vectorSize,\n );\n }\n\n const tableNames = await connection.tableNames();\n if (!tableNames.includes(options.tableName)) {\n await connection.createTable(options.tableName, []);\n }\n\n return new LanceDBVectorStore<T, Metadata>(\n connection,\n options.tableName,\n distance,\n options.vectorSize,\n );\n }\n\n async upsertDocuments(documents: Array<EmbeddedDocument<T, Metadata>>): Promise<void> {\n const table = await this.connection.openTable(this.tableName);\n const rows = documents.flatMap((document) => lanceRows(document));\n await table.add(rows);\n }\n\n index(model: EmbeddingModel): LanceDBVectorIndex<T, Metadata> {\n return new LanceDBVectorIndex(model, this.connection, this.tableName);\n }\n}\n"],"mappings":";AAEO,SAAS,kBAAkB,QAAsD;AACtF,MAAI,WAAW,QAAW;AACxB,WAAO;AAAA,EACT;AAEA,SAAO,gBAAgB,MAAM;AAC/B;AAEA,SAAS,gBAAgB,QAA8B;AACrD,UAAQ,OAAO,MAAM;AAAA,IACnB,KAAK;AACH,UAAI,OAAO,OAAO,UAAU,UAAU;AACpC,eAAO,GAAG,OAAO,GAAG,OAAO,UAAU,OAAO,KAAK,CAAC;AAAA,MACpD;AACA,UAAI,OAAO,OAAO,UAAU,WAAW;AACrC,eAAO,GAAG,OAAO,GAAG,MAAM,OAAO,QAAQ,SAAS,OAAO;AAAA,MAC3D;AACA,UAAI,OAAO,UAAU,MAAM;AACzB,eAAO,GAAG,OAAO,GAAG;AAAA,MACtB;AACA,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,OAAO,QAAQ,IAAI,CAAC,MAAM,IAAI,gBAAgB,CAAC,CAAC,GAAG,EAAE,KAAK,OAAO;AAAA,IAC1E,KAAK;AACH,aAAO,OAAO,QAAQ,IAAI,CAAC,MAAM,IAAI,gBAAgB,CAAC,CAAC,GAAG,EAAE,KAAK,MAAM;AAAA,EAC3E;AACF;AAEA,SAAS,UAAU,OAAuB;AACxC,SAAO,MAAM,QAAQ,MAAM,IAAI;AACjC;;;ACpCA,SAA8B,iBAAsC;AAEpE;AAAA,EACE;AAAA,OAKK;;;ACRP,SAAS,kBAAkB;;;ACApB,IAAM,mBAAmB;AACzB,IAAM,iBAAiB;AACvB,IAAM,eAAe;AACrB,IAAM,uBAAuB;;;ADE7B,SAAS,yBAAyB,UAA4C;AACnF,aAAW,OAAO,OAAO,KAAK,YAAY,CAAC,CAAC,GAAG;AAC7C,QAAI,IAAI,WAAW,oBAAoB,GAAG;AACxC,YAAM,IAAI,MAAM,gBAAgB,GAAG,wCAAwC;AAAA,IAC7E;AAAA,EACF;AACF;AAEO,SAAS,MAAM,IAAoB;AACxC,SAAO,WAAW,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,KAAK,EAAE,MAAM,GAAG,EAAE;AAClE;AAEO,SAAS,kBAAkB,UAA2B;AAC3D,SAAO,OAAO,aAAa,WAAW,WAAW,KAAK,UAAU,QAAQ;AAC1E;AAEO,SAAS,cAAiB,UAAsB;AACrD,MAAI,aAAa,QAAQ,aAAa,QAAW;AAC/C,WAAO;AAAA,EACT;AACA,MAAI,OAAO,aAAa,UAAU;AAChC,WAAO;AAAA,EACT;AACA,MAAI;AACF,WAAO,KAAK,MAAM,QAAQ;AAAA,EAC5B,QAAQ;AACN,WAAO;AAAA,EACT;AACF;AAEO,SAAS,UACd,UACgC;AAChC,MAAI,SAAS,WAAW,WAAW,GAAG;AACpC,UAAM,IAAI,MAAM,YAAY,SAAS,EAAE,oBAAoB;AAAA,EAC7D;AACA,2BAAyB,SAAS,QAAQ;AAE1C,SAAO,SAAS,WAAW,IAAI,CAAC,WAAW,UAAU;AACnD,UAAM,YACJ,SAAS,WAAW,WAAW,IAAI,SAAS,KAAK,GAAG,SAAS,EAAE,cAAc,KAAK;AACpF,WAAO;AAAA,MACL,QAAQ,MAAM,SAAS;AAAA,MACvB,CAAC,gBAAgB,GAAG,SAAS;AAAA,MAC7B,CAAC,cAAc,GAAG,kBAAkB,SAAS,QAAQ;AAAA,MACrD,CAAC,YAAY,GAAG,UAAU;AAAA,MAC1B,GAAI,SAAS,YAAY,CAAC;AAAA,IAC5B;AAAA,EACF,CAAC;AACH;AAEO,SAAS,kBACd,UACA,WACwC;AACxC,QAAM,OAAO,oBAAI,IAA6C;AAE9D,aAAW,OAAO,UAAU;AAC1B,UAAM,SAAS;AACf,UAAM,WAAW,OAAO,OAAO,cAAc,WAAW,OAAO,YAAY;AAC3E,UAAM,QAAQ,IAAI;AAElB,QAAI,cAAc,UAAa,QAAQ,WAAW;AAChD;AAAA,IACF;AAEA,UAAM,KAAK,OAAO,OAAO,gBAAgB,KAAK,EAAE;AAChD,UAAM,SAAS;AAAA,MACb;AAAA,MACA;AAAA,MACA,UAAU,cAAc,OAAO,cAAc,CAAC;AAAA,MAC9C,GAAG,oBAA8B,MAAM;AAAA,IACzC;AACA,UAAM,UAAU,KAAK,IAAI,EAAE;AAC3B,QAAI,YAAY,UAAa,OAAO,QAAQ,QAAQ,OAAO;AACzD,WAAK,IAAI,IAAI,MAAM;AAAA,IACrB;AAAA,EACF;AAEA,SAAO,CAAC,GAAG,KAAK,OAAO,CAAC;AAC1B;AAEA,eAAsB,yBACpB,KACqD;AACrD,QAAM,UAAU,MAAM,OAAO,kBAAkB;AAC/C,SAAO,QAAQ;AAAA,IACb,OAAO;AAAA,EACT;AACF;AAEA,SAAS,oBACP,QACqC;AACrC,QAAM,WAAW,OAAO;AAAA,IACtB,OAAO,QAAQ,MAAM,EAAE;AAAA,MACrB,CAAC,CAAC,GAAG,MACH,CAAC,IAAI,WAAW,oBAAoB,KACpC,QAAQ,gBACR,QAAQ,YACR,QAAQ;AAAA,IACZ;AAAA,EACF;AACA,SAAO,OAAO,KAAK,QAAQ,EAAE,WAAW,IAAI,CAAC,IAAI,EAAE,SAAS;AAC9D;;;ADhGO,IAAM,qBAAN,MAEP;AAAA,EACE,YACmB,OACA,YACA,WACjB;AAHiB;AACA;AACA;AAAA,EAChB;AAAA,EAHgB;AAAA,EACA;AAAA,EACA;AAAA,EAGnB,MAAM,OAAO,SAA+E;AAC1F,UAAM,iBAAiB,MAAM,UAAU,KAAK,OAAO,QAAQ,KAAK;AAChE,UAAM,QAAQ,MAAM,KAAK,WAAW,UAAU,KAAK,SAAS;AAC5D,UAAM,aAAa,kBAAkB,QAAQ,MAAM;AAEnD,UAAM,QAAQ,MAAM,OAAO,eAAe,MAAM,EAAE,MAAM,QAAQ,IAAI;AAEpE,QAAI;AACJ,QAAI,eAAe,QAAW;AAC5B,iBAAY,MAAM,MAAM,OAAO,UAAU;AAAA,IAC3C,OAAO;AACL,iBAAY,MAAM,MAAM,QAAQ;AAAA,IAClC;AAEA,WAAO,kBAA+B,UAAU,QAAQ,SAAS;AAAA,EACnE;AAAA,EAEA,MAAM,UAAU,SAA6E;AAC3F,YAAQ,MAAM,KAAK,OAAO,OAAO,GAAG,IAAI,CAAC,EAAE,OAAO,GAAG,OAAO,EAAE,OAAO,GAAG,EAAE;AAAA,EAC5E;AAAA,EAEA,OAAO,SAAmF;AACxF,WAAO,uBAAuB,MAAM,OAAO;AAAA,EAC7C;AACF;;;AGrCO,IAAM,qBAAN,MAAM,oBAAwE;AAAA,EAC3E,YACW,YACA,WACA,UACA,YACjB;AAJiB;AACA;AACA;AACA;AAAA,EAChB;AAAA,EAJgB;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EAGnB,aAAa,QACX,SAC0C;AAC1C,UAAM,aAAa,QAAQ,UAAW,MAAM,yBAAyB,QAAQ,GAAG;AAChF,UAAM,WAAW,QAAQ,YAAY;AAErC,QAAI,QAAQ,oBAAoB,OAAO;AACrC,YAAM,WAAW,UAAU,QAAQ,SAAS;AAC5C,aAAO,IAAI;AAAA,QACT;AAAA,QACA,QAAQ;AAAA,QACR;AAAA,QACA,QAAQ;AAAA,MACV;AAAA,IACF;AAEA,UAAM,aAAa,MAAM,WAAW,WAAW;AAC/C,QAAI,CAAC,WAAW,SAAS,QAAQ,SAAS,GAAG;AAC3C,YAAM,WAAW,YAAY,QAAQ,WAAW,CAAC,CAAC;AAAA,IACpD;AAEA,WAAO,IAAI;AAAA,MACT;AAAA,MACA,QAAQ;AAAA,MACR;AAAA,MACA,QAAQ;AAAA,IACV;AAAA,EACF;AAAA,EAEA,MAAM,gBAAgB,WAAgE;AACpF,UAAM,QAAQ,MAAM,KAAK,WAAW,UAAU,KAAK,SAAS;AAC5D,UAAM,OAAO,UAAU,QAAQ,CAAC,aAAa,UAAU,QAAQ,CAAC;AAChE,UAAM,MAAM,IAAI,IAAI;AAAA,EACtB;AAAA,EAEA,MAAM,OAAwD;AAC5D,WAAO,IAAI,mBAAmB,OAAO,KAAK,YAAY,KAAK,SAAS;AAAA,EACtE;AACF;","names":[]}
1
+ {"version":3,"sources":["../src/filters.ts","../src/search-index.ts","../src/helpers.ts","../src/types.ts","../src/store.ts"],"sourcesContent":["import type { VectorFilter } from \"@anvia/core/vector-store\";\n\nexport function filterToLanceExpr(filter: VectorFilter | undefined): string | undefined {\n if (filter === undefined) {\n return undefined;\n }\n\n return translateFilter(filter);\n}\n\nfunction translateFilter(filter: VectorFilter): string {\n switch (filter.type) {\n case \"eq\":\n if (typeof filter.value === \"string\") {\n return `${filter.key} = '${escapeSql(filter.value)}'`;\n }\n if (typeof filter.value === \"boolean\") {\n return `${filter.key} = ${filter.value ? \"TRUE\" : \"FALSE\"}`;\n }\n if (filter.value === null) {\n return `${filter.key} IS NULL`;\n }\n return `${filter.key} = ${filter.value}`;\n case \"gt\":\n return `${filter.key} > ${filter.value}`;\n case \"lt\":\n return `${filter.key} < ${filter.value}`;\n case \"and\":\n return filter.filters.map((f) => `(${translateFilter(f)})`).join(\" AND \");\n case \"or\":\n return filter.filters.map((f) => `(${translateFilter(f)})`).join(\" OR \");\n }\n}\n\nfunction escapeSql(value: string): string {\n return value.replace(/'/g, \"''\");\n}\n","import { type EmbeddingModel, embedText, type VectorMetadata } from \"@anvia/core/embeddings\";\nimport type { Tool } from \"@anvia/core/tool\";\nimport {\n createVectorSearchTool,\n type VectorSearchIndex,\n type VectorSearchRequest,\n type VectorSearchResult,\n type VectorSearchToolOptions,\n} from \"@anvia/core/vector-store\";\nimport { filterToLanceExpr } from \"./filters.js\";\nimport { parseQueryResults } from \"./helpers.js\";\nimport type { LanceDBConnectionLike } from \"./types.js\";\n\nexport class LanceDBVectorIndex<T, Metadata extends VectorMetadata = VectorMetadata>\n implements VectorSearchIndex<T, Metadata>\n{\n constructor(\n private readonly model: EmbeddingModel,\n private readonly connection: LanceDBConnectionLike,\n private readonly tableName: string,\n ) {}\n\n async search(request: VectorSearchRequest): Promise<Array<VectorSearchResult<T, Metadata>>> {\n const queryEmbedding = await embedText(this.model, request.query);\n const table = await this.connection.openTable(this.tableName);\n const filterExpr = filterToLanceExpr(request.filter);\n\n const query = table.search(queryEmbedding.vector).limit(request.topK);\n\n let response: unknown[];\n if (filterExpr !== undefined) {\n response = (await query.filter(filterExpr)) as unknown[];\n } else {\n response = (await query.toArray()) as unknown[];\n }\n\n return parseQueryResults<T, Metadata>(response, request.threshold);\n }\n\n async searchIds(request: VectorSearchRequest): Promise<Array<{ score: number; id: string }>> {\n return (await this.search(request)).map(({ score, id }) => ({ score, id }));\n }\n\n asTool(options: VectorSearchToolOptions): Tool<{ query: string; topK?: number }, unknown> {\n return createVectorSearchTool(this, options);\n }\n}\n","import { createHash } from \"node:crypto\";\nimport type { EmbeddedDocument, VectorMetadata } from \"@anvia/core/embeddings\";\nimport type { VectorSearchResult } from \"@anvia/core/vector-store\";\nimport { documentColumn, documentIdColumn, reservedColumnPrefix, vectorColumn } from \"./types.js\";\n\nexport function assertNoReservedMetadata(metadata: VectorMetadata | undefined): void {\n for (const key of Object.keys(metadata ?? {})) {\n if (key.startsWith(reservedColumnPrefix)) {\n throw new Error(`Metadata key ${key} is reserved for Anvia LanceDB columns`);\n }\n }\n}\n\nexport function rowId(id: string): string {\n return createHash(\"sha256\").update(id).digest(\"hex\").slice(0, 32);\n}\n\nexport function serializeDocument(document: unknown): string {\n return typeof document === \"string\" ? document : JSON.stringify(document);\n}\n\nexport function parseDocument<T>(document: unknown): T {\n if (document === null || document === undefined) {\n return \"\" as T;\n }\n if (typeof document !== \"string\") {\n return document as T;\n }\n try {\n return JSON.parse(document) as T;\n } catch {\n return document as T;\n }\n}\n\nexport function lanceRows<T, Metadata extends VectorMetadata>(\n document: EmbeddedDocument<T, Metadata>,\n): Array<Record<string, unknown>> {\n if (document.embeddings.length === 0) {\n throw new Error(`Document ${document.id} has no embeddings`);\n }\n assertNoReservedMetadata(document.metadata);\n\n return document.embeddings.map((embedding, index) => {\n const logicalId =\n document.embeddings.length === 1 ? document.id : `${document.id}#embedding:${index}`;\n const row: Record<string, unknown> = {\n _rowid: rowId(logicalId),\n [documentIdColumn]: document.id,\n [documentColumn]: serializeDocument(document.document),\n [vectorColumn]: embedding.vector,\n };\n Object.assign(row, document.metadata);\n return row;\n });\n}\n\nexport function parseQueryResults<T, Metadata extends VectorMetadata>(\n response: unknown[],\n threshold: number | undefined,\n): Array<VectorSearchResult<T, Metadata>> {\n const byId = new Map<string, VectorSearchResult<T, Metadata>>();\n\n for (const row of response) {\n const record = row as Record<string, unknown>;\n const distance = typeof record._distance === \"number\" ? record._distance : 0;\n const score = 1 - distance;\n\n if (threshold !== undefined && score < threshold) {\n continue;\n }\n\n const id = String(record[documentIdColumn] ?? \"\");\n const result: VectorSearchResult<T, Metadata> = {\n id,\n score,\n document: parseDocument(record[documentColumn]),\n };\n const metadata = metadataFromColumns<Metadata>(record);\n if (metadata !== undefined) {\n result.metadata = metadata;\n }\n const current = byId.get(id);\n if (current === undefined || result.score > current.score) {\n byId.set(id, result);\n }\n }\n\n return [...byId.values()];\n}\n\nexport async function defaultLanceDBConnection(\n uri?: string,\n): Promise<import(\"./types.js\").LanceDBConnectionLike> {\n const lancedb = await import(\"@lancedb/lancedb\");\n return lancedb.connect(\n uri ?? \"~/.anvia/lancedb\",\n ) as unknown as import(\"./types.js\").LanceDBConnectionLike;\n}\n\nfunction metadataFromColumns<Metadata extends VectorMetadata>(\n record: Record<string, unknown>,\n): Metadata | undefined {\n const metadata = Object.fromEntries(\n Object.entries(record).filter(\n ([key]) =>\n !key.startsWith(reservedColumnPrefix) &&\n key !== vectorColumn &&\n key !== \"_rowid\" &&\n key !== \"_distance\",\n ),\n ) as Metadata;\n return Object.keys(metadata).length === 0 ? undefined : metadata;\n}\n","export const documentIdColumn = \"__anvia_document_id\";\nexport const documentColumn = \"__anvia_document\";\nexport const vectorColumn = \"__anvia_vector\";\nexport const reservedColumnPrefix = \"__anvia_\";\n\nexport type LanceDBDistance = \"cosine\" | \"l2\" | \"dot\";\n\nexport type LanceDBTableLike = {\n add(rows: Record<string, unknown>[]): Promise<unknown>;\n search(vector: number[]): {\n limit(n: number): {\n filter(expr: string | undefined): Promise<unknown[]>;\n toArray(): Promise<unknown[]>;\n };\n };\n countRows(): Promise<number>;\n delete(filter: string): Promise<unknown>;\n};\n\nexport type LanceDBConnectionLike = {\n openTable(name: string): Promise<LanceDBTableLike>;\n tableNames(): Promise<string[]>;\n createTable(name: string, data: Record<string, unknown>[]): Promise<LanceDBTableLike>;\n};\n\nexport type LanceDBVectorStoreConnectOptions = {\n client?: LanceDBConnectionLike | undefined;\n uri?: string | undefined;\n tableName: string;\n vectorSize: number;\n createIfMissing?: boolean | undefined;\n distance?: LanceDBDistance | undefined;\n};\n","import type { EmbeddedDocument, EmbeddingModel, VectorMetadata } from \"@anvia/core/embeddings\";\nimport { defaultLanceDBConnection, lanceRows } from \"./helpers.js\";\nimport { LanceDBVectorIndex } from \"./search-index.js\";\nimport type {\n LanceDBConnectionLike,\n LanceDBDistance,\n LanceDBVectorStoreConnectOptions,\n} from \"./types.js\";\n\nexport class LanceDBVectorStore<T, Metadata extends VectorMetadata = VectorMetadata> {\n private constructor(\n private readonly connection: LanceDBConnectionLike,\n private readonly tableName: string,\n private readonly distance: LanceDBDistance,\n private readonly vectorSize: number,\n ) {}\n\n static async connect<T, Metadata extends VectorMetadata = VectorMetadata>(\n options: LanceDBVectorStoreConnectOptions,\n ): Promise<LanceDBVectorStore<T, Metadata>> {\n const connection = options.client ?? (await defaultLanceDBConnection(options.uri));\n const distance = options.distance ?? \"cosine\";\n\n if (options.createIfMissing === false) {\n await connection.openTable(options.tableName);\n return new LanceDBVectorStore<T, Metadata>(\n connection,\n options.tableName,\n distance,\n options.vectorSize,\n );\n }\n\n const tableNames = await connection.tableNames();\n if (!tableNames.includes(options.tableName)) {\n await connection.createTable(options.tableName, []);\n }\n\n return new LanceDBVectorStore<T, Metadata>(\n connection,\n options.tableName,\n distance,\n options.vectorSize,\n );\n }\n\n async upsertDocuments(documents: Array<EmbeddedDocument<T, Metadata>>): Promise<void> {\n const table = await this.connection.openTable(this.tableName);\n const rows = documents.flatMap((document) => lanceRows(document));\n await table.add(rows);\n }\n\n index(model: EmbeddingModel): LanceDBVectorIndex<T, Metadata> {\n return new LanceDBVectorIndex(model, this.connection, this.tableName);\n }\n}\n"],"mappings":";AAEO,SAAS,kBAAkB,QAAsD;AACtF,MAAI,WAAW,QAAW;AACxB,WAAO;AAAA,EACT;AAEA,SAAO,gBAAgB,MAAM;AAC/B;AAEA,SAAS,gBAAgB,QAA8B;AACrD,UAAQ,OAAO,MAAM;AAAA,IACnB,KAAK;AACH,UAAI,OAAO,OAAO,UAAU,UAAU;AACpC,eAAO,GAAG,OAAO,GAAG,OAAO,UAAU,OAAO,KAAK,CAAC;AAAA,MACpD;AACA,UAAI,OAAO,OAAO,UAAU,WAAW;AACrC,eAAO,GAAG,OAAO,GAAG,MAAM,OAAO,QAAQ,SAAS,OAAO;AAAA,MAC3D;AACA,UAAI,OAAO,UAAU,MAAM;AACzB,eAAO,GAAG,OAAO,GAAG;AAAA,MACtB;AACA,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,GAAG,OAAO,GAAG,MAAM,OAAO,KAAK;AAAA,IACxC,KAAK;AACH,aAAO,OAAO,QAAQ,IAAI,CAAC,MAAM,IAAI,gBAAgB,CAAC,CAAC,GAAG,EAAE,KAAK,OAAO;AAAA,IAC1E,KAAK;AACH,aAAO,OAAO,QAAQ,IAAI,CAAC,MAAM,IAAI,gBAAgB,CAAC,CAAC,GAAG,EAAE,KAAK,MAAM;AAAA,EAC3E;AACF;AAEA,SAAS,UAAU,OAAuB;AACxC,SAAO,MAAM,QAAQ,MAAM,IAAI;AACjC;;;ACpCA,SAA8B,iBAAsC;AAEpE;AAAA,EACE;AAAA,OAKK;;;ACRP,SAAS,kBAAkB;;;ACApB,IAAM,mBAAmB;AACzB,IAAM,iBAAiB;AACvB,IAAM,eAAe;AACrB,IAAM,uBAAuB;;;ADE7B,SAAS,yBAAyB,UAA4C;AACnF,aAAW,OAAO,OAAO,KAAK,YAAY,CAAC,CAAC,GAAG;AAC7C,QAAI,IAAI,WAAW,oBAAoB,GAAG;AACxC,YAAM,IAAI,MAAM,gBAAgB,GAAG,wCAAwC;AAAA,IAC7E;AAAA,EACF;AACF;AAEO,SAAS,MAAM,IAAoB;AACxC,SAAO,WAAW,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,KAAK,EAAE,MAAM,GAAG,EAAE;AAClE;AAEO,SAAS,kBAAkB,UAA2B;AAC3D,SAAO,OAAO,aAAa,WAAW,WAAW,KAAK,UAAU,QAAQ;AAC1E;AAEO,SAAS,cAAiB,UAAsB;AACrD,MAAI,aAAa,QAAQ,aAAa,QAAW;AAC/C,WAAO;AAAA,EACT;AACA,MAAI,OAAO,aAAa,UAAU;AAChC,WAAO;AAAA,EACT;AACA,MAAI;AACF,WAAO,KAAK,MAAM,QAAQ;AAAA,EAC5B,QAAQ;AACN,WAAO;AAAA,EACT;AACF;AAEO,SAAS,UACd,UACgC;AAChC,MAAI,SAAS,WAAW,WAAW,GAAG;AACpC,UAAM,IAAI,MAAM,YAAY,SAAS,EAAE,oBAAoB;AAAA,EAC7D;AACA,2BAAyB,SAAS,QAAQ;AAE1C,SAAO,SAAS,WAAW,IAAI,CAAC,WAAW,UAAU;AACnD,UAAM,YACJ,SAAS,WAAW,WAAW,IAAI,SAAS,KAAK,GAAG,SAAS,EAAE,cAAc,KAAK;AACpF,UAAM,MAA+B;AAAA,MACnC,QAAQ,MAAM,SAAS;AAAA,MACvB,CAAC,gBAAgB,GAAG,SAAS;AAAA,MAC7B,CAAC,cAAc,GAAG,kBAAkB,SAAS,QAAQ;AAAA,MACrD,CAAC,YAAY,GAAG,UAAU;AAAA,IAC5B;AACA,WAAO,OAAO,KAAK,SAAS,QAAQ;AACpC,WAAO;AAAA,EACT,CAAC;AACH;AAEO,SAAS,kBACd,UACA,WACwC;AACxC,QAAM,OAAO,oBAAI,IAA6C;AAE9D,aAAW,OAAO,UAAU;AAC1B,UAAM,SAAS;AACf,UAAM,WAAW,OAAO,OAAO,cAAc,WAAW,OAAO,YAAY;AAC3E,UAAM,QAAQ,IAAI;AAElB,QAAI,cAAc,UAAa,QAAQ,WAAW;AAChD;AAAA,IACF;AAEA,UAAM,KAAK,OAAO,OAAO,gBAAgB,KAAK,EAAE;AAChD,UAAM,SAA0C;AAAA,MAC9C;AAAA,MACA;AAAA,MACA,UAAU,cAAc,OAAO,cAAc,CAAC;AAAA,IAChD;AACA,UAAM,WAAW,oBAA8B,MAAM;AACrD,QAAI,aAAa,QAAW;AAC1B,aAAO,WAAW;AAAA,IACpB;AACA,UAAM,UAAU,KAAK,IAAI,EAAE;AAC3B,QAAI,YAAY,UAAa,OAAO,QAAQ,QAAQ,OAAO;AACzD,WAAK,IAAI,IAAI,MAAM;AAAA,IACrB;AAAA,EACF;AAEA,SAAO,CAAC,GAAG,KAAK,OAAO,CAAC;AAC1B;AAEA,eAAsB,yBACpB,KACqD;AACrD,QAAM,UAAU,MAAM,OAAO,kBAAkB;AAC/C,SAAO,QAAQ;AAAA,IACb,OAAO;AAAA,EACT;AACF;AAEA,SAAS,oBACP,QACsB;AACtB,QAAM,WAAW,OAAO;AAAA,IACtB,OAAO,QAAQ,MAAM,EAAE;AAAA,MACrB,CAAC,CAAC,GAAG,MACH,CAAC,IAAI,WAAW,oBAAoB,KACpC,QAAQ,gBACR,QAAQ,YACR,QAAQ;AAAA,IACZ;AAAA,EACF;AACA,SAAO,OAAO,KAAK,QAAQ,EAAE,WAAW,IAAI,SAAY;AAC1D;;;ADpGO,IAAM,qBAAN,MAEP;AAAA,EACE,YACmB,OACA,YACA,WACjB;AAHiB;AACA;AACA;AAAA,EAChB;AAAA,EAHgB;AAAA,EACA;AAAA,EACA;AAAA,EAGnB,MAAM,OAAO,SAA+E;AAC1F,UAAM,iBAAiB,MAAM,UAAU,KAAK,OAAO,QAAQ,KAAK;AAChE,UAAM,QAAQ,MAAM,KAAK,WAAW,UAAU,KAAK,SAAS;AAC5D,UAAM,aAAa,kBAAkB,QAAQ,MAAM;AAEnD,UAAM,QAAQ,MAAM,OAAO,eAAe,MAAM,EAAE,MAAM,QAAQ,IAAI;AAEpE,QAAI;AACJ,QAAI,eAAe,QAAW;AAC5B,iBAAY,MAAM,MAAM,OAAO,UAAU;AAAA,IAC3C,OAAO;AACL,iBAAY,MAAM,MAAM,QAAQ;AAAA,IAClC;AAEA,WAAO,kBAA+B,UAAU,QAAQ,SAAS;AAAA,EACnE;AAAA,EAEA,MAAM,UAAU,SAA6E;AAC3F,YAAQ,MAAM,KAAK,OAAO,OAAO,GAAG,IAAI,CAAC,EAAE,OAAO,GAAG,OAAO,EAAE,OAAO,GAAG,EAAE;AAAA,EAC5E;AAAA,EAEA,OAAO,SAAmF;AACxF,WAAO,uBAAuB,MAAM,OAAO;AAAA,EAC7C;AACF;;;AGrCO,IAAM,qBAAN,MAAM,oBAAwE;AAAA,EAC3E,YACW,YACA,WACA,UACA,YACjB;AAJiB;AACA;AACA;AACA;AAAA,EAChB;AAAA,EAJgB;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EAGnB,aAAa,QACX,SAC0C;AAC1C,UAAM,aAAa,QAAQ,UAAW,MAAM,yBAAyB,QAAQ,GAAG;AAChF,UAAM,WAAW,QAAQ,YAAY;AAErC,QAAI,QAAQ,oBAAoB,OAAO;AACrC,YAAM,WAAW,UAAU,QAAQ,SAAS;AAC5C,aAAO,IAAI;AAAA,QACT;AAAA,QACA,QAAQ;AAAA,QACR;AAAA,QACA,QAAQ;AAAA,MACV;AAAA,IACF;AAEA,UAAM,aAAa,MAAM,WAAW,WAAW;AAC/C,QAAI,CAAC,WAAW,SAAS,QAAQ,SAAS,GAAG;AAC3C,YAAM,WAAW,YAAY,QAAQ,WAAW,CAAC,CAAC;AAAA,IACpD;AAEA,WAAO,IAAI;AAAA,MACT;AAAA,MACA,QAAQ;AAAA,MACR;AAAA,MACA,QAAQ;AAAA,IACV;AAAA,EACF;AAAA,EAEA,MAAM,gBAAgB,WAAgE;AACpF,UAAM,QAAQ,MAAM,KAAK,WAAW,UAAU,KAAK,SAAS;AAC5D,UAAM,OAAO,UAAU,QAAQ,CAAC,aAAa,UAAU,QAAQ,CAAC;AAChE,UAAM,MAAM,IAAI,IAAI;AAAA,EACtB;AAAA,EAEA,MAAM,OAAwD;AAC5D,WAAO,IAAI,mBAAmB,OAAO,KAAK,YAAY,KAAK,SAAS;AAAA,EACtE;AACF;","names":[]}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@anvia/lancedb",
3
- "version": "0.2.4",
3
+ "version": "0.2.5",
4
4
  "description": "LanceDB vector store adapter for Anvia.",
5
5
  "author": "anvia",
6
6
  "maintainer": "Indra Zulfi",
@@ -33,7 +33,7 @@
33
33
  "tsup": "^8.5.0",
34
34
  "typescript": "^5.9.3",
35
35
  "vitest": "^4.0.8",
36
- "@anvia/core": "0.12.3"
36
+ "@anvia/core": "0.13.1"
37
37
  },
38
38
  "peerDependencies": {
39
39
  "@anvia/core": ">=0.7.1 <1.0.0"