@dxos/assistant 0.8.4-main.f9ba587 → 0.8.4-main.fcc0d83b33

package/README.md

@@ -1,4 +1,5 @@
 # @dxos/assistant
+
 Natural language interface to DXOS.
 
 Communicate with your DXOS applications through conversations using advanced LLM capabilities. This package provides a flexible system for building AI assistants that can help users interact with and control DXOS apps.

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

@@ -0,0 +1,202 @@
+// src/extraction/extraction.ts
+import * as Schema from "effect/Schema";
+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 ExtractionInput = Schema.Struct({
+  message: Message.Message,
+  objects: Schema.optional(Schema.Array(Obj.Unknown)),
+  options: Schema.optional(Schema.Struct({
+    timeout: Schema.optional(Schema.Number),
+    fallbackToRaw: Schema.optional(Schema.Boolean)
+  }))
+});
+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) {
+    } else {
+    }
+  } catch (error) {
+    if (params.options?.fallbackToRaw) {
+      log.warn("failed to process message, falling back to raw text", {
+        error
+      }, { "~LogMeta": "~LogMeta", F: __dxlog_file, L: 57, S: void 0 });
+      return {
+        message: params.input.message,
+        timeElapsed: 0
+      };
+    } else {
+      throw error;
+    }
+  }
+  return {
+    message: params.input.message,
+    timeElapsed: 0
+  };
+};
+
+// 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 _ner;
+var getNer = () => {
+  if (!_ner) {
+    _ner = pipeline("ner", "Xenova/bert-base-NER").then((ner) => {
+      log2.info("NER model is ready", void 0, { "~LogMeta": "~LogMeta", F: __dxlog_file2, L: 14, S: void 0 });
+      return ner;
+    });
+  }
+  return _ner;
+};
+var extractFullEntities = async (text) => {
+  const ner = await getNer();
+  const entities = await ner(text);
+  const tokenGroups = createTokenGroups(entities);
+  const combinedEntities = tokenGroups.map(combineNerTokens);
+  return combinedEntities;
+};
+var createTokenGroups = (tokens) => {
+  const tokenGroups = [];
+  const canBeCombined = (prevToken, currentToken) => {
+    const [currentState, currentEntity] = currentToken.entity.split("-");
+    const [prevState, prevEntity] = prevToken.entity.split("-");
+    if (prevEntity !== currentEntity) {
+      return false;
+    }
+    if (prevState === "O") {
+      return false;
+    }
+    if (currentState === "I") {
+      return true;
+    }
+    if ((prevState === "B" || prevState === "I") && currentState === "B") {
+      return currentToken.word.startsWith("#");
+    }
+    return false;
+  };
+  for (const token of tokens) {
+    if (tokenGroups.length === 0) {
+      tokenGroups.push([
+        token
+      ]);
+      continue;
+    }
+    const lastTokenGroup = tokenGroups.at(-1);
+    if (canBeCombined(lastTokenGroup.at(-1), token)) {
+      lastTokenGroup.push(token);
+      continue;
+    }
+    tokenGroups.push([
+      token
+    ]);
+  }
+  return tokenGroups;
+};
+var combineNerTokens = (group) => {
+  const combinedToken = {
+    entity: group.at(0).entity.split("-")[1],
+    score: group.reduce((acc, token) => acc + token.score, 0) / group.length,
+    index: group.at(0).index,
+    word: group.at(0).word,
+    start: group.at(0).start,
+    end: group.at(-1).end
+  };
+  for (const token of group.slice(1)) {
+    if (token.word.startsWith("#")) {
+      combinedToken.word += token.word.replace(/^#+/, "");
+    } else {
+      combinedToken.word += " " + token.word;
+    }
+  }
+  return combinedToken;
+};
+
+// src/extraction/quotes.ts
+import * as Schema2 from "effect/Schema";
+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 ReferencedQuotes = Schema2.Struct({
+  references: Schema2.Array(Schema2.Struct({
+    id: Schema2.String,
+    quote: Schema2.String
+  })).pipe(Schema2.mutable).annotations({
+    // TODO(burdon): Does this description make sense?
+    description: `
+      The references to the context objects that are mentioned in the transcript. 
+      Quote should match the original transcript text exactly, while id is the id of the context object.
+    `
+  })
+});
+var findQuotes = async (quotes, db) => {
+  const references = await Promise.all(quotes.map(async (quote) => {
+    const objects = await db.query(Query.select(Filter.text(quote, {
+      type: "vector"
+    }))).run();
+    log3.info("found quotes in vector index", {
+      objects,
+      quote
+    }, { "~LogMeta": "~LogMeta", F: __dxlog_file3, L: 26, S: void 0 });
+    return objects.length > 0 ? [
+      {
+        id: objects[0].id.toString(),
+        quote
+      }
+    ] : [];
+  }));
+  return {
+    references: references.flat()
+  };
+};
+var insertReferences = (text, quotes) => {
+  for (const quote of quotes.references) {
+    if (!ObjectId.isValid(quote.id)) {
+      continue;
+    }
+    const regex = new RegExp(quote.quote.replace(/[.*+?^${}()|[\]\\]/g, "\\$&"), "gi");
+    text = text.replace(regex, `[${quote.quote}](${DXN.fromLocalObjectId(quote.id).toString()})`);
+  }
+  return text;
+};
+export {
+  ExtractionInput,
+  ExtractionOutput,
+  ReferencedQuotes,
+  combineNerTokens,
+  createTokenGroups,
+  extractFullEntities,
+  extractionAnthropicFunction,
+  extractionNerFunction,
+  findQuotes,
+  getNer,
+  insertReferences,
+  processTranscriptMessage
+};
+//# sourceMappingURL=index.mjs.map

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

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "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"]
+}