@dxos/assistant 0.8.4-main.c85a9c8dae → 0.8.4-main.cb12b3f963

package/dist/lib/neutral/extraction/index.mjs

@@ -1,10 +1,10 @@
 // src/extraction/extraction.ts
 import * as Schema from "effect/Schema";
-import { asyncTimeout } from "@dxos/async";
+import { Operation } from "@dxos/compute";
 import { Obj } from "@dxos/echo";
 import { log } from "@dxos/log";
 import { Message } from "@dxos/types";
-var __dxlog_file = "/__w/dxos/dxos/packages/core/assistant/src/extraction/extraction.ts";
+var __dxlog_file = "/__w/dxos/dxos/packages/core/compute/assistant/src/extraction/extraction.ts";
 var ExtractionInput = Schema.Struct({
   message: Message.Message,
   objects: Schema.optional(Schema.Array(Obj.Unknown)),
@@ -17,23 +17,34 @@ var ExtractionOutput = Schema.Struct({
   message: Message.Message,
   timeElapsed: Schema.Number
 });
+var extractionAnthropicFunction = Operation.make({
+  meta: {
+    key: "org.dxos.function.extraction.anthropic",
+    name: "Entity Extraction (Anthropic)",
+    description: "Extract entities from transcript using Anthropic LLM."
+  },
+  input: ExtractionInput,
+  output: ExtractionOutput
+});
+var extractionNerFunction = Operation.make({
+  meta: {
+    key: "org.dxos.function.extraction.ner",
+    name: "Entity Extraction (NER)",
+    description: "Extract entities from transcript using Named Entity Recognition."
+  },
+  input: ExtractionInput,
+  output: ExtractionOutput
+});
 var processTranscriptMessage = async (params) => {
   try {
     if (params.options?.timeout && params.options.timeout > 0) {
-      return await asyncTimeout(params.executor.invoke(params.function, params.input), params.options.timeout);
     } else {
-      return await params.executor.invoke(params.function, params.input);
     }
   } catch (error) {
     if (params.options?.fallbackToRaw) {
       log.warn("failed to process message, falling back to raw text", {
         error
-      }, {
-        F: __dxlog_file,
-        L: 67,
-        S: void 0,
-        C: (f, a) => f(...a)
-      });
+      }, { "~LogMeta": "~LogMeta", F: __dxlog_file, L: 57, S: void 0 });
       return {
         message: params.input.message,
         timeElapsed: 0
@@ -42,48 +53,21 @@ var processTranscriptMessage = async (params) => {
       throw error;
     }
   }
+  return {
+    message: params.input.message,
+    timeElapsed: 0
+  };
 };
 
-// src/extraction/extraction-llm-function/extraction-llm-function.ts
-import { defineFunction } from "@dxos/functions";
-var extractionAnthropicFunction = defineFunction({
-  key: "dxos.org/function/extraction/extract-entities",
-  name: "Extract Entities",
-  description: "Extract entities from the transcript message and add them to the message.",
-  inputSchema: ExtractionInput,
-  outputSchema: ExtractionOutput,
-  handler: async ({ data: { message, objects }, context }) => {
-    throw new Error("not implemented");
-  }
-});
-
-// src/extraction/extraction-ner-function/extraction-ner-function.ts
-import { defineFunction as defineFunction2 } from "@dxos/functions";
-var extractionNerFunction = defineFunction2({
-  key: "dxos.org/function/extraction/extract-entities",
-  name: "Extract Entities",
-  description: "Extracts named entities from a text",
-  inputSchema: ExtractionInput,
-  outputSchema: ExtractionOutput,
-  handler: async ({ data: { message, options }, context }) => {
-    throw new Error("not implemented");
-  }
-});
-
 // src/extraction/extraction-ner-function/named-entity-recognition.ts
 import { pipeline } from "@xenova/transformers";
 import { log as log2 } from "@dxos/log";
-var __dxlog_file2 = "/__w/dxos/dxos/packages/core/assistant/src/extraction/extraction-ner-function/named-entity-recognition.ts";
+var __dxlog_file2 = "/__w/dxos/dxos/packages/core/compute/assistant/src/extraction/extraction-ner-function/named-entity-recognition.ts";
 var _ner;
 var getNer = () => {
   if (!_ner) {
     _ner = pipeline("ner", "Xenova/bert-base-NER").then((ner) => {
-      log2.info("NER model is ready", void 0, {
-        F: __dxlog_file2,
-        L: 24,
-        S: void 0,
-        C: (f, a) => f(...a)
-      });
+      log2.info("NER model is ready", void 0, { "~LogMeta": "~LogMeta", F: __dxlog_file2, L: 14, S: void 0 });
       return ner;
     });
   }
@@ -158,7 +142,7 @@ import { Filter, Query } from "@dxos/echo";
 import { ObjectId } from "@dxos/keys";
 import { DXN } from "@dxos/keys";
 import { log as log3 } from "@dxos/log";
-var __dxlog_file3 = "/__w/dxos/dxos/packages/core/assistant/src/extraction/quotes.ts";
+var __dxlog_file3 = "/__w/dxos/dxos/packages/core/compute/assistant/src/extraction/quotes.ts";
 var ReferencedQuotes = Schema2.Struct({
   references: Schema2.Array(Schema2.Struct({
     id: Schema2.String,
@@ -179,12 +163,7 @@ var findQuotes = async (quotes, db) => {
     log3.info("found quotes in vector index", {
       objects,
       quote
-    }, {
-      F: __dxlog_file3,
-      L: 34,
-      S: void 0,
-      C: (f, a) => f(...a)
-    });
+    }, { "~LogMeta": "~LogMeta", F: __dxlog_file3, L: 26, S: void 0 });
     return objects.length > 0 ? [
       {
         id: objects[0].id.toString(),

package/dist/lib/neutral/extraction/index.mjs.map

@@ -1,7 +1,7 @@
 {
   "version": 3,
-  "sources": ["../../../../src/extraction/extraction.ts", "../../../../src/extraction/extraction-llm-function/extraction-llm-function.ts", "../../../../src/extraction/extraction-ner-function/extraction-ner-function.ts", "../../../../src/extraction/extraction-ner-function/named-entity-recognition.ts", "../../../../src/extraction/quotes.ts"],
-  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport * as Schema from 'effect/Schema';\n\nimport { asyncTimeout } from '@dxos/async';\nimport { Obj } from '@dxos/echo';\nimport { type FunctionDefinition } from '@dxos/functions';\nimport { type FunctionExecutor } from '@dxos/functions-runtime';\nimport { log } from '@dxos/log';\nimport { Message } from '@dxos/types';\n\nexport const ExtractionInput = Schema.Struct({\n  message: Message.Message,\n  objects: Schema.optional(Schema.Array(Obj.Unknown)),\n  options: Schema.optional(\n    Schema.Struct({\n      timeout: Schema.optional(Schema.Number),\n      fallbackToRaw: Schema.optional(Schema.Boolean),\n    }),\n  ),\n});\nexport interface ExtractionInput extends Schema.Schema.Type<typeof ExtractionInput> {}\n\nexport const ExtractionOutput = Schema.Struct({\n  message: Message.Message,\n  timeElapsed: Schema.Number,\n});\nexport interface ExtractionOutput extends Schema.Schema.Type<typeof ExtractionOutput> {}\n\nexport type ProcessTranscriptMessageProps = {\n  input: ExtractionInput;\n  function: FunctionDefinition<ExtractionInput, ExtractionOutput>;\n  executor: FunctionExecutor;\n\n  options?: {\n    /**\n     * Timeout for the entity extraction.\n     */\n    timeout?: number;\n\n    /**\n     * Fallback to raw text if the entity extraction fails.\n     * Otherwise the function will throw an error.\n     * @default false\n     */\n    fallbackToRaw?: boolean;\n  };\n};\n\nexport type ExtractionFunction = FunctionDefinition<ExtractionInput, ExtractionOutput>;\n\n/**\n * Extract entities from the transcript message and add them to the message.\n */\n// TODO(dmaretskyi): Move context to a vector search index.\nexport const processTranscriptMessage = async (params: ProcessTranscriptMessageProps): Promise<ExtractionOutput> => {\n  try {\n    if (params.options?.timeout && params.options.timeout > 0) {\n      return await asyncTimeout(params.executor.invoke(params.function, params.input), params.options.timeout);\n    } else {\n      return await params.executor.invoke(params.function, params.input);\n    }\n  } catch (error) {\n    if (params.options?.fallbackToRaw) {\n      log.warn('failed to process message, falling back to raw text', { error });\n      return {\n        message: params.input.message,\n        timeElapsed: 0,\n      };\n    } else {\n      throw error;\n    }\n  }\n};\n", "//\n// Copyright 2025 DXOS.org\n//\n\nimport { type FunctionDefinition, defineFunction } from '@dxos/functions';\n\nimport { ExtractionInput, ExtractionOutput } from '../extraction';\n\nexport const extractionAnthropicFunction: FunctionDefinition<ExtractionInput, ExtractionOutput> = defineFunction({\n  key: 'dxos.org/function/extraction/extract-entities',\n  name: 'Extract Entities',\n  description: 'Extract entities from the transcript message and add them to the message.',\n  inputSchema: ExtractionInput,\n  outputSchema: ExtractionOutput,\n  handler: async ({ data: { message, objects }, context }) => {\n    throw new Error('not implemented');\n    // const startTime = performance.now();\n    // const ai = context.getService(AiService.AiService);\n    // const session = new AiSession({ operationModel: 'configured' });\n    // const result = await session.runStructured(ReferencedQuotes, {\n    //   generationOptions: {\n    //     model: '@anthropic/claude-3-5-haiku-20241022',\n    //   },\n    //   client: ai.client,\n    //   systemPrompt: PROMPT,\n    //   history: [\n    //     Obj.make(DataType.Message.Message, {\n    //       created: new Date().toISOString(),\n    //       sender: { role: 'user' },\n    //       blocks: [\n    //         {\n    //           _tag: 'text',\n    //           text: `<context>${JSON.stringify(objects)}</context>`,\n    //         } satisfies DataType.ContentBlock.Text,\n    //         {\n    //           _tag: 'text',\n    //           text: `<transcript>${JSON.stringify(message.blocks)}</transcript>`,\n    //         } satisfies DataType.ContentBlock.Text,\n    //       ],\n    //     }),\n    //   ],\n    //   artifacts: [],\n    //   prompt: '',\n    //   tools: [],\n    // } as any); // TODO(burdon): Rewrite test.\n\n    // return {\n    //   timeElapsed: performance.now() - startTime,\n    //   message: create(DataType.Message.Message, {\n    //     ...message,\n    //     blocks: message.blocks.map((block) =>\n    //       block._tag !== 'transcript'\n    //         ? block\n    //         : {\n    //             ...block,\n    //             text: insertReferences(block.text, result),\n    //           },\n    //     ),\n    //   }),\n    // };\n  },\n});\n", "//\n// Copyright 2025 DXOS.org\n//\n\nimport { defineFunction } from '@dxos/functions';\n\nimport { ExtractionInput, ExtractionOutput } from '../extraction';\n\n/**\n * Named Entity Recognition model\n */\n// TODO(mykola): Make it use vector index to query objects and not pass objects as input.\nexport const extractionNerFunction = defineFunction({\n  key: 'dxos.org/function/extraction/extract-entities',\n  name: 'Extract Entities',\n  description: 'Extracts named entities from a text',\n  inputSchema: ExtractionInput,\n  outputSchema: ExtractionOutput,\n  handler: async ({ data: { message, options }, context }) => {\n    throw new Error('not implemented');\n    // log.info('input', { message, options });\n    // const startTime = performance.now();\n    // const { db } = context.getService(Database.Service);\n    // const { extractFullEntities } = await import('./named-entity-recognition');\n\n    // const entitiesPromise = Promise.all(\n    //   message.blocks.map(async (block) => {\n    //     invariant(block._tag === 'transcript' || block._tag === 'text');\n    //     return extractFullEntities(block.text);\n    //   }),\n    // ).then((entities) => entities.flat());\n    // const entities = options?.timeout ? await asyncTimeout(entitiesPromise, options.timeout) : await entitiesPromise;\n\n    // log.info('entities', { entities });\n    // const quoteReferences = await findQuotes(\n    //   entities.map((entity) => entity.word),\n    //   db,\n    // );\n\n    // const blocksWithReferences = message.blocks.map((block) => {\n    //   invariant(block._tag === 'transcript' || block._tag === 'text');\n    //   return { ...block, text: insertReferences(block._tag, quoteReferences) };\n    // });\n\n    // const messageWithReferences = create(Message.Message, {\n    //   ...message,\n    //   blocks: blocksWithReferences,\n    // });\n\n    // log.info('output', { messageWithReferences });\n    // return {\n    //   message: messageWithReferences,\n    //   timeElapsed: performance.now() - startTime,\n    // };\n  },\n});\n", "//\n// Copyright 2025 DXOS.org\n//\n\nimport {\n  type TokenClassificationOutput,\n  type TokenClassificationPipelineType,\n  type TokenClassificationSingle,\n  pipeline,\n} from '@xenova/transformers';\n\nimport { log } from '@dxos/log';\n\nlet _ner: Promise<TokenClassificationPipelineType>;\n\n/**\n * Named Entity Recognition pipeline.\n * Initializes the pipeline on first call.\n * @returns The singleton promise that resolves to a token classification pipeline.\n */\nexport const getNer = () => {\n  if (!_ner) {\n    _ner = pipeline('ner', 'Xenova/bert-base-NER').then((ner) => {\n      log.info('NER model is ready');\n      return ner;\n    });\n  }\n\n  return _ner;\n};\n\n/**\n * Extracts entities from text using the named entity recognition pipeline and does simple grouping of tokens.\n */\nexport const extractFullEntities = async (text: string): Promise<TokenClassificationSingle[]> => {\n  const ner = await getNer();\n  const entities = (await ner(text)) as TokenClassificationOutput;\n  const tokenGroups = createTokenGroups(entities);\n  const combinedEntities = tokenGroups.map(combineNerTokens);\n  return combinedEntities;\n};\n\nexport const createTokenGroups = (tokens: TokenClassificationSingle[]): TokenClassificationSingle[][] => {\n  const tokenGroups: TokenClassificationSingle[][] = [];\n\n  const canBeCombined = (prevToken: TokenClassificationSingle, currentToken: TokenClassificationSingle) => {\n    const [currentState, currentEntity] = currentToken.entity.split('-'); // B, I, O\n    const [prevState, prevEntity] = prevToken.entity.split('-'); // B, I, O\n    if (prevEntity !== currentEntity) {\n      return false;\n    }\n    if (prevState === 'O') {\n      return false;\n    }\n    if (currentState === 'I') {\n      return true;\n    }\n    if ((prevState === 'B' || prevState === 'I') && currentState === 'B') {\n      return currentToken.word.startsWith('#');\n    }\n    return false;\n  };\n\n  for (const token of tokens) {\n    if (tokenGroups.length === 0) {\n      tokenGroups.push([token]);\n      continue;\n    }\n    const lastTokenGroup = tokenGroups.at(-1)!;\n    if (canBeCombined(lastTokenGroup.at(-1)!, token)) {\n      lastTokenGroup.push(token);\n      continue;\n    }\n    tokenGroups.push([token]);\n  }\n\n  return tokenGroups;\n};\n\nexport const combineNerTokens = (group: TokenClassificationSingle[]): TokenClassificationSingle => {\n  const combinedToken: TokenClassificationSingle = {\n    entity: group.at(0)!.entity.split('-')[1],\n    score: group.reduce((acc, token) => acc + token.score, 0) / group.length,\n    index: group.at(0)!.index,\n    word: group.at(0)!.word,\n    start: group.at(0)!.start,\n    end: group.at(-1)!.end,\n  };\n\n  for (const token of group.slice(1)) {\n    if (token.word.startsWith('#')) {\n      combinedToken.word += token.word.replace(/^#+/, '');\n    } else {\n      combinedToken.word += ' ' + token.word;\n    }\n  }\n\n  return combinedToken;\n};\n", "//\n// Copyright 2025 DXOS.org\n//\n\nimport * as Schema from 'effect/Schema';\n\nimport { type Database, Filter, Query } from '@dxos/echo';\nimport { ObjectId } from '@dxos/keys';\nimport { DXN } from '@dxos/keys';\nimport { log } from '@dxos/log';\n\nexport const ReferencedQuotes = Schema.Struct({\n  references: Schema.Array(\n    Schema.Struct({\n      id: Schema.String,\n      quote: Schema.String, // TODO(burdon): Quote?\n    }),\n  )\n    .pipe(Schema.mutable)\n    .annotations({\n      // TODO(burdon): Does this description make sense?\n      description: `\n      The references to the context objects that are mentioned in the transcript. \n      Quote should match the original transcript text exactly, while id is the id of the context object.\n    `,\n    }),\n});\nexport type ReferencedQuotes = Schema.Schema.Type<typeof ReferencedQuotes>;\n\nexport const findQuotes = async (quotes: string[], db: Database.Database): Promise<ReferencedQuotes> => {\n  const references = await Promise.all(\n    quotes.map(async (quote) => {\n      const objects = await db.query(Query.select(Filter.text(quote, { type: 'vector' }))).run();\n      log.info('found quotes in vector index', { objects, quote });\n      return objects.length > 0 ? [{ id: objects[0].id.toString(), quote }] : [];\n    }),\n  );\n  return { references: references.flat() };\n};\n\n/**\n * Finds and replaces all quotes with DXNs references.\n */\n// TODO(dmaretskyi): Lookup and verifiy ids from provided context.\nexport const insertReferences = (text: string, quotes: ReferencedQuotes) => {\n  for (const quote of quotes.references) {\n    if (!ObjectId.isValid(quote.id)) {\n      continue;\n    }\n\n    // Use a case-insensitive regular expression to replace the quote.\n    const regex = new RegExp(quote.quote.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&'), 'gi');\n    text = text.replace(regex, `[${quote.quote}](${DXN.fromLocalObjectId(quote.id).toString()})`);\n  }\n\n  return text;\n};\n"],
-  "mappings": ";AAIA,YAAYA,YAAY;AAExB,SAASC,oBAAoB;AAC7B,SAASC,WAAW;AAGpB,SAASC,WAAW;AACpB,SAASC,eAAe;;AAEjB,IAAMC,kBAAyBC,cAAO;EAC3CC,SAASH,QAAQA;EACjBI,SAAgBC,gBAAgBC,aAAMR,IAAIS,OAAO,CAAA;EACjDC,SAAgBH,gBACPH,cAAO;IACZO,SAAgBJ,gBAAgBK,aAAM;IACtCC,eAAsBN,gBAAgBO,cAAO;EAC/C,CAAA,CAAA;AAEJ,CAAA;AAGO,IAAMC,mBAA0BX,cAAO;EAC5CC,SAASH,QAAQA;EACjBc,aAAoBJ;AACtB,CAAA;AA6BO,IAAMK,2BAA2B,OAAOC,WAAAA;AAC7C,MAAI;AACF,QAAIA,OAAOR,SAASC,WAAWO,OAAOR,QAAQC,UAAU,GAAG;AACzD,aAAO,MAAMZ,aAAamB,OAAOC,SAASC,OAAOF,OAAOG,UAAUH,OAAOI,KAAK,GAAGJ,OAAOR,QAAQC,OAAO;IACzG,OAAO;AACL,aAAO,MAAMO,OAAOC,SAASC,OAAOF,OAAOG,UAAUH,OAAOI,KAAK;IACnE;EACF,SAASC,OAAO;AACd,QAAIL,OAAOR,SAASG,eAAe;AACjCZ,UAAIuB,KAAK,uDAAuD;QAAED;MAAM,GAAA;;;;;;AACxE,aAAO;QACLlB,SAASa,OAAOI,MAAMjB;QACtBW,aAAa;MACf;IACF,OAAO;AACL,YAAMO;IACR;EACF;AACF;;;ACvEA,SAAkCE,sBAAsB;AAIjD,IAAMC,8BAAqFC,eAAe;EAC/GC,KAAK;EACLC,MAAM;EACNC,aAAa;EACbC,aAAaC;EACbC,cAAcC;EACdC,SAAS,OAAO,EAAEC,MAAM,EAAEC,SAASC,QAAO,GAAIC,QAAO,MAAE;AACrD,UAAM,IAAIC,MAAM,iBAAA;EA6ClB;AACF,CAAA;;;ACzDA,SAASC,kBAAAA,uBAAsB;AAQxB,IAAMC,wBAAwBC,gBAAe;EAClDC,KAAK;EACLC,MAAM;EACNC,aAAa;EACbC,aAAaC;EACbC,cAAcC;EACdC,SAAS,OAAO,EAAEC,MAAM,EAAEC,SAASC,QAAO,GAAIC,QAAO,MAAE;AACrD,UAAM,IAAIC,MAAM,iBAAA;EAmClB;AACF,CAAA;;;ACnDA,SAIEC,gBACK;AAEP,SAASC,OAAAA,YAAW;;AAEpB,IAAIC;AAOG,IAAMC,SAAS,MAAA;AACpB,MAAI,CAACD,MAAM;AACTA,WAAOF,SAAS,OAAO,sBAAA,EAAwBI,KAAK,CAACC,QAAAA;AACnDJ,MAAAA,KAAIK,KAAK,sBAAA,QAAA;;;;;;AACT,aAAOD;IACT,CAAA;EACF;AAEA,SAAOH;AACT;AAKO,IAAMK,sBAAsB,OAAOC,SAAAA;AACxC,QAAMH,MAAM,MAAMF,OAAAA;AAClB,QAAMM,WAAY,MAAMJ,IAAIG,IAAAA;AAC5B,QAAME,cAAcC,kBAAkBF,QAAAA;AACtC,QAAMG,mBAAmBF,YAAYG,IAAIC,gBAAAA;AACzC,SAAOF;AACT;AAEO,IAAMD,oBAAoB,CAACI,WAAAA;AAChC,QAAML,cAA6C,CAAA;AAEnD,QAAMM,gBAAgB,CAACC,WAAsCC,iBAAAA;AAC3D,UAAM,CAACC,cAAcC,aAAAA,IAAiBF,aAAaG,OAAOC,MAAM,GAAA;AAChE,UAAM,CAACC,WAAWC,UAAAA,IAAcP,UAAUI,OAAOC,MAAM,GAAA;AACvD,QAAIE,eAAeJ,eAAe;AAChC,aAAO;IACT;AACA,QAAIG,cAAc,KAAK;AACrB,aAAO;IACT;AACA,QAAIJ,iBAAiB,KAAK;AACxB,aAAO;IACT;AACA,SAAKI,cAAc,OAAOA,cAAc,QAAQJ,iBAAiB,KAAK;AACpE,aAAOD,aAAaO,KAAKC,WAAW,GAAA;IACtC;AACA,WAAO;EACT;AAEA,aAAWC,SAASZ,QAAQ;AAC1B,QAAIL,YAAYkB,WAAW,GAAG;AAC5BlB,kBAAYmB,KAAK;QAACF;OAAM;AACxB;IACF;AACA,UAAMG,iBAAiBpB,YAAYqB,GAAG,EAAC;AACvC,QAAIf,cAAcc,eAAeC,GAAG,EAAC,GAAKJ,KAAAA,GAAQ;AAChDG,qBAAeD,KAAKF,KAAAA;AACpB;IACF;AACAjB,gBAAYmB,KAAK;MAACF;KAAM;EAC1B;AAEA,SAAOjB;AACT;AAEO,IAAMI,mBAAmB,CAACkB,UAAAA;AAC/B,QAAMC,gBAA2C;IAC/CZ,QAAQW,MAAMD,GAAG,CAAA,EAAIV,OAAOC,MAAM,GAAA,EAAK,CAAA;IACvCY,OAAOF,MAAMG,OAAO,CAACC,KAAKT,UAAUS,MAAMT,MAAMO,OAAO,CAAA,IAAKF,MAAMJ;IAClES,OAAOL,MAAMD,GAAG,CAAA,EAAIM;IACpBZ,MAAMO,MAAMD,GAAG,CAAA,EAAIN;IACnBa,OAAON,MAAMD,GAAG,CAAA,EAAIO;IACpBC,KAAKP,MAAMD,GAAG,EAAC,EAAIQ;EACrB;AAEA,aAAWZ,SAASK,MAAMQ,MAAM,CAAA,GAAI;AAClC,QAAIb,MAAMF,KAAKC,WAAW,GAAA,GAAM;AAC9BO,oBAAcR,QAAQE,MAAMF,KAAKgB,QAAQ,OAAO,EAAA;IAClD,OAAO;AACLR,oBAAcR,QAAQ,MAAME,MAAMF;IACpC;EACF;AAEA,SAAOQ;AACT;;;AC9FA,YAAYS,aAAY;AAExB,SAAwBC,QAAQC,aAAa;AAC7C,SAASC,gBAAgB;AACzB,SAASC,WAAW;AACpB,SAASC,OAAAA,YAAW;;AAEb,IAAMC,mBAA0BC,eAAO;EAC5CC,YAAmBC,cACVF,eAAO;IACZG,IAAWC;IACXC,OAAcD;EAChB,CAAA,CAAA,EAECE,KAAYC,eAAO,EACnBC,YAAY;;IAEXC,aAAa;;;;EAIf,CAAA;AACJ,CAAA;AAGO,IAAMC,aAAa,OAAOC,QAAkBC,OAAAA;AACjD,QAAMX,aAAa,MAAMY,QAAQC,IAC/BH,OAAOI,IAAI,OAAOV,UAAAA;AAChB,UAAMW,UAAU,MAAMJ,GAAGK,MAAMtB,MAAMuB,OAAOxB,OAAOyB,KAAKd,OAAO;MAAEe,MAAM;IAAS,CAAA,CAAA,CAAA,EAAKC,IAAG;AACxFvB,IAAAA,KAAIwB,KAAK,gCAAgC;MAAEN;MAASX;IAAM,GAAA;;;;;;AAC1D,WAAOW,QAAQO,SAAS,IAAI;MAAC;QAAEpB,IAAIa,QAAQ,CAAA,EAAGb,GAAGqB,SAAQ;QAAInB;MAAM;QAAK,CAAA;EAC1E,CAAA,CAAA;AAEF,SAAO;IAAEJ,YAAYA,WAAWwB,KAAI;EAAG;AACzC;AAMO,IAAMC,mBAAmB,CAACP,MAAcR,WAAAA;AAC7C,aAAWN,SAASM,OAAOV,YAAY;AACrC,QAAI,CAACL,SAAS+B,QAAQtB,MAAMF,EAAE,GAAG;AAC/B;IACF;AAGA,UAAMyB,QAAQ,IAAIC,OAAOxB,MAAMA,MAAMyB,QAAQ,uBAAuB,MAAA,GAAS,IAAA;AAC7EX,WAAOA,KAAKW,QAAQF,OAAO,IAAIvB,MAAMA,KAAK,KAAKR,IAAIkC,kBAAkB1B,MAAMF,EAAE,EAAEqB,SAAQ,CAAA,GAAK;EAC9F;AAEA,SAAOL;AACT;",
-  "names": ["Schema", "asyncTimeout", "Obj", "log", "Message", "ExtractionInput", "Struct", "message", "objects", "optional", "Array", "Unknown", "options", "timeout", "Number", "fallbackToRaw", "Boolean", "ExtractionOutput", "timeElapsed", "processTranscriptMessage", "params", "executor", "invoke", "function", "input", "error", "warn", "defineFunction", "extractionAnthropicFunction", "defineFunction", "key", "name", "description", "inputSchema", "ExtractionInput", "outputSchema", "ExtractionOutput", "handler", "data", "message", "objects", "context", "Error", "defineFunction", "extractionNerFunction", "defineFunction", "key", "name", "description", "inputSchema", "ExtractionInput", "outputSchema", "ExtractionOutput", "handler", "data", "message", "options", "context", "Error", "pipeline", "log", "_ner", "getNer", "then", "ner", "info", "extractFullEntities", "text", "entities", "tokenGroups", "createTokenGroups", "combinedEntities", "map", "combineNerTokens", "tokens", "canBeCombined", "prevToken", "currentToken", "currentState", "currentEntity", "entity", "split", "prevState", "prevEntity", "word", "startsWith", "token", "length", "push", "lastTokenGroup", "at", "group", "combinedToken", "score", "reduce", "acc", "index", "start", "end", "slice", "replace", "Schema", "Filter", "Query", "ObjectId", "DXN", "log", "ReferencedQuotes", "Struct", "references", "Array", "id", "String", "quote", "pipe", "mutable", "annotations", "description", "findQuotes", "quotes", "db", "Promise", "all", "map", "objects", "query", "select", "text", "type", "run", "info", "length", "toString", "flat", "insertReferences", "isValid", "regex", "RegExp", "replace", "fromLocalObjectId"]
+  "sources": ["../../../../src/extraction/extraction.ts", "../../../../src/extraction/extraction-ner-function/named-entity-recognition.ts", "../../../../src/extraction/quotes.ts"],
+  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport * as Schema from 'effect/Schema';\n\nimport { Operation } from '@dxos/compute';\nimport { Obj } from '@dxos/echo';\nimport { log } from '@dxos/log';\nimport { Message } from '@dxos/types';\n\nexport const ExtractionInput = Schema.Struct({\n  message: Message.Message,\n  objects: Schema.optional(Schema.Array(Obj.Unknown)),\n  options: Schema.optional(\n    Schema.Struct({\n      timeout: Schema.optional(Schema.Number),\n      fallbackToRaw: Schema.optional(Schema.Boolean),\n    }),\n  ),\n});\nexport interface ExtractionInput extends Schema.Schema.Type<typeof ExtractionInput> {}\n\nexport const ExtractionOutput = Schema.Struct({\n  message: Message.Message,\n  timeElapsed: Schema.Number,\n});\nexport interface ExtractionOutput extends Schema.Schema.Type<typeof ExtractionOutput> {}\n\nexport type ProcessTranscriptMessageProps = {\n  input: ExtractionInput;\n  function: Operation.Definition<ExtractionInput, ExtractionOutput>;\n\n  options?: {\n    /**\n     * Timeout for the entity extraction.\n     */\n    timeout?: number;\n\n    /**\n     * Fallback to raw text if the entity extraction fails.\n     * Otherwise the function will throw an error.\n     * @default false\n     */\n    fallbackToRaw?: boolean;\n  };\n};\n\nexport type ExtractionFunction = Operation.Definition<ExtractionInput, ExtractionOutput>;\n\n/**\n * Placeholder for Anthropic-based entity extraction.\n * Implementation is currently disabled pending restoration of extraction logic.\n */\nexport const extractionAnthropicFunction = Operation.make({\n  meta: {\n    key: 'org.dxos.function.extraction.anthropic',\n    name: 'Entity Extraction (Anthropic)',\n    description: 'Extract entities from transcript using Anthropic LLM.',\n  },\n  input: ExtractionInput,\n  output: ExtractionOutput,\n});\n\n/**\n * Placeholder for NER-based entity extraction.\n * Implementation is currently disabled pending restoration of extraction logic.\n */\nexport const extractionNerFunction = Operation.make({\n  meta: {\n    key: 'org.dxos.function.extraction.ner',\n    name: 'Entity Extraction (NER)',\n    description: 'Extract entities from transcript using Named Entity Recognition.',\n  },\n  input: ExtractionInput,\n  output: ExtractionOutput,\n});\n\n/**\n * Extract entities from the transcript message and add them to the message.\n */\n// TODO(dmaretskyi): Move context to a vector search index.\nexport const processTranscriptMessage = async (params: ProcessTranscriptMessageProps): Promise<ExtractionOutput> => {\n  try {\n    if (params.options?.timeout && params.options.timeout > 0) {\n      // return await asyncTimeout(params.executor.invoke(params.function, params.input), params.options.timeout);\n    } else {\n      // return await params.executor.invoke(params.function, params.input);\n    }\n  } catch (error) {\n    if (params.options?.fallbackToRaw) {\n      log.warn('failed to process message, falling back to raw text', { error });\n      return {\n        message: params.input.message,\n        timeElapsed: 0,\n      };\n    } else {\n      throw error;\n    }\n  }\n\n  // TODO(dmaretskyi): Restore extraction logic.\n  return { message: params.input.message, timeElapsed: 0 };\n};\n", "//\n// Copyright 2025 DXOS.org\n//\n\nimport {\n  type TokenClassificationOutput,\n  type TokenClassificationPipelineType,\n  type TokenClassificationSingle,\n  pipeline,\n} from '@xenova/transformers';\n\nimport { log } from '@dxos/log';\n\nlet _ner: Promise<TokenClassificationPipelineType>;\n\n/**\n * Named Entity Recognition pipeline.\n * Initializes the pipeline on first call.\n * @returns The singleton promise that resolves to a token classification pipeline.\n */\nexport const getNer = () => {\n  if (!_ner) {\n    _ner = pipeline('ner', 'Xenova/bert-base-NER').then((ner) => {\n      log.info('NER model is ready');\n      return ner;\n    });\n  }\n\n  return _ner;\n};\n\n/**\n * Extracts entities from text using the named entity recognition pipeline and does simple grouping of tokens.\n */\nexport const extractFullEntities = async (text: string): Promise<TokenClassificationSingle[]> => {\n  const ner = await getNer();\n  const entities = (await ner(text)) as TokenClassificationOutput;\n  const tokenGroups = createTokenGroups(entities);\n  const combinedEntities = tokenGroups.map(combineNerTokens);\n  return combinedEntities;\n};\n\nexport const createTokenGroups = (tokens: TokenClassificationSingle[]): TokenClassificationSingle[][] => {\n  const tokenGroups: TokenClassificationSingle[][] = [];\n\n  const canBeCombined = (prevToken: TokenClassificationSingle, currentToken: TokenClassificationSingle) => {\n    const [currentState, currentEntity] = currentToken.entity.split('-'); // B, I, O\n    const [prevState, prevEntity] = prevToken.entity.split('-'); // B, I, O\n    if (prevEntity !== currentEntity) {\n      return false;\n    }\n    if (prevState === 'O') {\n      return false;\n    }\n    if (currentState === 'I') {\n      return true;\n    }\n    if ((prevState === 'B' || prevState === 'I') && currentState === 'B') {\n      return currentToken.word.startsWith('#');\n    }\n    return false;\n  };\n\n  for (const token of tokens) {\n    if (tokenGroups.length === 0) {\n      tokenGroups.push([token]);\n      continue;\n    }\n    const lastTokenGroup = tokenGroups.at(-1)!;\n    if (canBeCombined(lastTokenGroup.at(-1)!, token)) {\n      lastTokenGroup.push(token);\n      continue;\n    }\n    tokenGroups.push([token]);\n  }\n\n  return tokenGroups;\n};\n\nexport const combineNerTokens = (group: TokenClassificationSingle[]): TokenClassificationSingle => {\n  const combinedToken: TokenClassificationSingle = {\n    entity: group.at(0)!.entity.split('-')[1],\n    score: group.reduce((acc, token) => acc + token.score, 0) / group.length,\n    index: group.at(0)!.index,\n    word: group.at(0)!.word,\n    start: group.at(0)!.start,\n    end: group.at(-1)!.end,\n  };\n\n  for (const token of group.slice(1)) {\n    if (token.word.startsWith('#')) {\n      combinedToken.word += token.word.replace(/^#+/, '');\n    } else {\n      combinedToken.word += ' ' + token.word;\n    }\n  }\n\n  return combinedToken;\n};\n", "//\n// Copyright 2025 DXOS.org\n//\n\nimport * as Schema from 'effect/Schema';\n\nimport { type Database, Filter, Query } from '@dxos/echo';\nimport { ObjectId } from '@dxos/keys';\nimport { DXN } from '@dxos/keys';\nimport { log } from '@dxos/log';\n\nexport const ReferencedQuotes = Schema.Struct({\n  references: Schema.Array(\n    Schema.Struct({\n      id: Schema.String,\n      quote: Schema.String, // TODO(burdon): Quote?\n    }),\n  )\n    .pipe(Schema.mutable)\n    .annotations({\n      // TODO(burdon): Does this description make sense?\n      description: `\n      The references to the context objects that are mentioned in the transcript. \n      Quote should match the original transcript text exactly, while id is the id of the context object.\n    `,\n    }),\n});\nexport type ReferencedQuotes = Schema.Schema.Type<typeof ReferencedQuotes>;\n\nexport const findQuotes = async (quotes: string[], db: Database.Database): Promise<ReferencedQuotes> => {\n  const references = await Promise.all(\n    quotes.map(async (quote) => {\n      const objects = await db.query(Query.select(Filter.text(quote, { type: 'vector' }))).run();\n      log.info('found quotes in vector index', { objects, quote });\n      return objects.length > 0 ? [{ id: objects[0].id.toString(), quote }] : [];\n    }),\n  );\n  return { references: references.flat() };\n};\n\n/**\n * Finds and replaces all quotes with DXNs references.\n */\n// TODO(dmaretskyi): Lookup and verifiy ids from provided context.\nexport const insertReferences = (text: string, quotes: ReferencedQuotes) => {\n  for (const quote of quotes.references) {\n    if (!ObjectId.isValid(quote.id)) {\n      continue;\n    }\n\n    // Use a case-insensitive regular expression to replace the quote.\n    const regex = new RegExp(quote.quote.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\$&'), 'gi');\n    text = text.replace(regex, `[${quote.quote}](${DXN.fromLocalObjectId(quote.id).toString()})`);\n  }\n\n  return text;\n};\n"],
+  "mappings": ";AAIA,YAAYA,YAAY;AAExB,SAASC,iBAAiB;AAC1B,SAASC,WAAW;AACpB,SAASC,WAAW;AACpB,SAASC,eAAe;AAExB,IAAA,eAAaC;AAEXC,IAASN,kBAAuBO,cAAML;EACtCM,SAASR,QAAOS;WAEHT,gBAAgBA,aAAOU,IAAM,OAAA,CAAA;WACtCC,gBAAsBF,cAAST;IACjC,SAAA,gBAAA,aAAA;IAED,eAAA,gBAAA,cAAA;EAGH,CAAA,CAAA;;AAEEY,IAAAA,mBAA0B,cAAA;EACzB,SAAA,QAAA;EAwBH,aAAA;;AAMIC,IAAK,8BAAA,UAAA,KAAA;QACLC;IACAC,KAAAA;IACF,MAAA;IACAC,aAAOX;EACPY;EACC,OAAA;EAEH,QAAA;;AAMIJ,IAAK,wBAAA,UAAA,KAAA;QACLC;IACAC,KAAAA;IACF,MAAA;IACAC,aAAOX;EACPY;EACC,OAAA;EAEH,QAAA;;IAMQC,2BAA2BA,OAAOV,WAAQW;MAC5C;AACF,QAAA,OAAO,SAAA,WAAA,OAAA,QAAA,UAAA,GAAA;IAEP,OAAA;IAEA;WACEhB,OAAQ;eAA0DiB,SAAAA,eAAAA;AAAM,UAAA,KAAA,uDAAA;QACxE;uBACWF,YAAY,GAACG,cAAO,GAAA,IAAA,GAAA,OAAA,CAAA;aAC7BT;QACF,SAAA,OAAA,MAAA;QACK,aAAA;MACL;IACF,OAAA;AACF,YAAA;IAEA;EACA;SAAwCA;IAAe,SAAA,OAAA,MAAA;IACvD,aAAA;;;;;ACnGF,SAIEU,gBACK;AAEP,SAASC,OAAAA,YAAW;AAEpB,IAAIC,gBAAAA;;AASAA,IAAOF,SAAS,MAAA;aACdC;WACA,SAAOE,OAAAA,sBAAAA,EAAAA,KAAAA,CAAAA,QAAAA;AACT,MAAAF,KAAA,KAAA,sBAAA,QAAA,EAAA,YAAA,YAAA,GAAAG,eAAA,GAAA,IAAA,GAAA,OAAA,CAAA;AACF,aAAA;IAEA,CAAA;EACA;AAEF,SAAA;;AAKQC,IAAAA,sBAAsBC,OAAAA,SAAAA;AAC5B,QAAMC,MAAAA,MAAAA,OAAcC;AACpB,QAAMC,WAAAA,MAAAA,IAAmBF,IAAAA;AACzB,QAAA,cAAOE,kBAAAA,QAAAA;AACP,QAAA,mBAAA,YAAA,IAAA,gBAAA;AAEF,SAAO;;AAGL,IAAMC,oBAAiBC,CAAAA,WAAsCC;QAC3D,cAAOC,CAAAA;QACP,gBAAkBC,CAAAA,WAAW,iBAAaC;AAC1C,UAAID,CAAAA,cAAeE,aAAe,IAAA,aAAA,OAAA,MAAA,GAAA;UAChC,CAAA,WAAO,UAAA,IAAA,UAAA,OAAA,MAAA,GAAA;AACT,QAAA,eAAA,eAAA;AACIC,aAAAA;;AAEJ,QAAA,cAAA,KAAA;AACIJ,aAAAA;;AAEJ,QAAA,iBAAA,KAAA;AACKI,aAAAA;;AAEL,SAAA,cAAA,OAAA,cAAA,QAAA,iBAAA,KAAA;AACA,aAAO,aAAA,KAAA,WAAA,GAAA;IACT;AAEA,WAAWC;;aAEPX,SAAYY,QAAK;oBAACD,WAAAA,GAAAA;kBAAM,KAAA;QACxB;MACF,CAAA;AACA;IACA;UACEE,iBAAmB,YAACF,GAAAA,EAAAA;QACpB,cAAA,eAAA,GAAA,EAAA,GAAA,KAAA,GAAA;AACF,qBAAA,KAAA,KAAA;AACAX;;gBAAwB,KAAA;MAC1B;IAEA,CAAA;EACA;AAEF,SAAO;;IAEHQ,mBAAqBA,CAAAA,UAAa;QAClCM,gBAAaC;IACbC,QAAOC,MAAMC,GAAG,CAAA,EAAIF,OAAK,MAAA,GAAA,EAAA,CAAA;IACzBG,OAAMF,MAAMC,OAAOC,CAAAA,KAAI,UAAA,MAAA,MAAA,OAAA,CAAA,IAAA,MAAA;IACvBC,OAAOH,MAAMC,GAAG,CAAA,EAAIE;IACpBC,MAAKJ,MAAMC,GAAI,CAAA,EAAIG;IACrB,OAAA,MAAA,GAAA,CAAA,EAAA;IAEA,KAAK,MAAMV,GAAAA,EAASM,EAAAA;;aAEhBK,SAAAA,MAAkB,MAAIX,CAAAA,GAAAA;AACxB,QAAA,MAAO,KAAA,WAAA,GAAA,GAAA;AACLW,oBAAcH,QAAQ,MAAMR,KAAAA,QAAU,OAAA,EAAA;IACxC,OAAA;AACF,oBAAA,QAAA,MAAA,MAAA;IAEA;EACA;;;;;AC9FF,YAAYY,aAAY;AAExB,SAAwBC,QAAQC,aAAa;AAC7C,SAASC,gBAAgB;AACzB,SAASC,WAAW;AACpB,SAASC,OAAAA,YAAW;AAEpB,IAAAC,gBAAaC;IAGHP,mBAAa,eAAA;cACHQ,cAAM,eAAA;IAGrBC,IAAKT;IAEJ,OAAA;MACAU,KAAAA,eAAc,EAAA,YAAA;;;;;;EAQpB,CAAA;;IAGM,aAAgB,OAASC,QAAMT,OAAMU;qBAAkC,MAAA,QAAA,IAAA,OAAA,IAAA,OAAA,UAAA;AAAS,UAAKC,UAAG,MAAA,GAAA,MAAA,MAAA,OAAA,OAAA,KAAA,OAAA;MACpFC,MAAK;SAAkCC,IAAAA;SAASC,KAAAA,gCAAAA;MAAM;MAC1D;qBAA6B,YAAA,GAAAV,eAAA,GAAA,IAAA,GAAA,OAAA,CAAA;mBAAMS,SAAWE,IAAGC;;QAAkB,IAAA,QAAA,CAAA,EAAA,GAAA,SAAA;QAAO;MAC5E;IAEF,IAAO,CAAA;;AAAgC,SAAA;IACvC,YAAA,WAAA,KAAA;EAEF;;IAMSf,mBAAiBa,CAAAA,MAAQ,WAAG;aAC/B,SAAA,OAAA,YAAA;AACF,QAAA,CAAA,SAAA,QAAA,MAAA,EAAA,GAAA;AAEA;IACA;AAEF,UAAA,QAAA,IAAA,OAAA,MAAA,MAAA,QAAA,uBAAA,MAAA,GAAA,IAAA;AAEA,WAAOG,KAAAA,QAAAA,OAAAA,IAAAA,MAAAA,KAAAA,KAAAA,IAAAA,kBAAAA,MAAAA,EAAAA,EAAAA,SAAAA,CAAAA,GAAAA;EACP;;;",
+  "names": ["Schema", "Operation", "Obj", "log", "Message", "ExtractionInput", "objects", "Array", "options", "optional", "Number", "fallbackToRaw", "timeElapsed", "key", "name", "description", "input", "output", "params", "timeout", "error", "message", "pipeline", "log", "_ner", "ner", "__dxlog_file", "entities", "text", "tokenGroups", "createTokenGroups", "combinedEntities", "canBeCombined", "prevToken", "currentToken", "currentState", "prevEntity", "entity", "currentEntity", "prevState", "token", "push", "lastTokenGroup", "score", "reduce", "index", "group", "at", "word", "start", "end", "combinedToken", "Schema", "Filter", "Query", "ObjectId", "DXN", "log", "__dxlog_file", "ReferencedQuotes", "String", "pipe", "description", "query", "select", "run", "info", "objects", "quote", "id", "toString", "text"]
 }