@agentmark-ai/shared-utils 0.7.0 → 0.8.0

package/dist/index.d.mts

@@ -175,6 +175,23 @@ interface ToolCall {
     result?: string;
     providerMetadata?: Record<string, any>;
 }
+/**
+ * A single document returned by a retriever / vector-store search, normalized
+ * from instrumentation that carries documents per-index (OpenInference's
+ * flattened `retrieval.documents.{i}.document.*`) or per-event (OpenLLMetry's
+ * `db.query.result` span events). Stored under a span's `outputObject` as
+ * `{ documents: RetrievalDocument[] }` so the UI can render a ranked panel
+ * (relevance score / distance, content, metadata) instead of a single blob.
+ */
+interface RetrievalDocument {
+    id?: string;
+    content?: string;
+    /** Relevance/similarity score (higher = more relevant). */
+    score?: number;
+    /** Vector distance (lower = closer). Some stores report this instead of a score. */
+    distance?: number;
+    metadata?: Record<string, any>;
+}
 interface OtelScope {
     name?: string;
     version?: string;
@@ -537,6 +554,17 @@ declare function messagesToPlainText(messages: Message[] | undefined): string |
 /** Valid semantic kind values. */
 declare const SEMANTIC_KINDS: readonly ["function", "llm", "tool", "agent", "retrieval", "embedding", "guardrail"];
 type SemanticKind = typeof SEMANTIC_KINDS[number];
+/**
+ * True when a span looks like a vector-store query: a recognized vector-DB
+ * `db.system`, any `db.vector.query.*` attribute, or `db.query.result` /
+ * `db.search.result` events. Distinguishes a vector search from a plain SQL
+ * query (which carries none of these). Shared by the semantic-kind resolver
+ * (→ classifies as "retrieval") and the dispatching transformer (→ routes to
+ * the OpenLLMetry extractor so the result events become documents).
+ */
+declare function hasVectorStoreSignature(attributes: Record<string, any>, events?: ReadonlyArray<{
+    name: string;
+}>): boolean;
 /**
  * Resolve the semantic kind of a span using a 9-level priority chain.
  *
@@ -660,6 +688,8 @@ declare class OtelGenAiTransformer implements ScopeTransformer {
  * @see https://github.com/Arize-ai/openinference/blob/main/spec/semantic_conventions.md
  */
 
+/** Join document contents into a readable, searchable text output. */
+declare function retrievalDocumentsToText(documents: RetrievalDocument[]): string | undefined;
 declare class OpenInferenceTransformer implements ScopeTransformer {
     classify(_span: OtelSpan, attributes: Record<string, any>): SpanType;
     transform(_span: OtelSpan, attributes: Record<string, any>): Partial<NormalizedSpan>;
@@ -710,11 +740,15 @@ declare class DispatchingTransformer implements ScopeTransformer {
     private readonly openInference;
     private readonly openLLMetry;
     private readonly otelGenAi;
-    /** Choose the extractor for a span from its attribute signature. OpenInference
-     *  is checked before OpenLLMetry because its markers (`llm.*`,
-     *  `openinference.span.kind`) are more specific; the bare OTel GenAI
-     *  transformer is the catch-all. */
-    select(attributes: Record<string, any>): ScopeTransformer;
+    /** Choose the extractor for a span from its attribute/event signature.
+     *  OpenInference is checked before OpenLLMetry because its markers (`llm.*`,
+     *  `openinference.span.kind`) are more specific. Vector-store query spans
+     *  route to the OpenLLMetry extractor — they may carry no `traceloop.*`
+     *  marker (a bare Pinecone span is just `db.system` + `db.query.result`
+     *  events), so without this they'd fall through to the OTel-GenAI catch-all
+     *  and their result documents would be dropped. The bare OTel GenAI
+     *  transformer is the final catch-all. */
+    select(attributes: Record<string, any>, span?: OtelSpan): ScopeTransformer;
     classify(span: OtelSpan, attributes: Record<string, any>): SpanType;
     transform(span: OtelSpan, attributes: Record<string, any>): Partial<NormalizedSpan>;
 }
@@ -727,4 +761,4 @@ declare function normalizeSpan(resource: OtelResource, scope: OtelScope, span: O
  */
 declare function normalizeOtlpSpans(resourceSpans: OtlpResourceSpans[]): NormalizedSpan[];
 
-export { AGENTMARK_SCOPE_NAME, AgentMarkTransformer, type AgentmarkConfig, type AgentmarkModelConfig, type AgentmarkModelSchema, type AgentmarkModelSettingsConfig, type AgentmarkModelSettingsSchema, AiSdkTransformer, type AiSdkVersion, type AttributeExtractor, AgentMarkTransformer as ClaudeAgentTransformer, DispatchingTransformer, type GenerateTypesLanguage, type IndexedMessageConfig, MastraTransformer, type McpServerConfig, type McpServers, type McpStdioServerConfig, type McpUrlServerConfig, type Message, type ModelSettingsTypeAspectRatio, type ModelSettingsTypeImageSize, type ModelSettingsTypeSelect, type ModelSettingsTypeSlider, type NormalizedSpan, OpenInferenceTransformer, OpenLLMetryTransformer, type OtelEvent, OtelGenAiTransformer, type OtelLink, type OtelResource, type OtelScope, type OtelSpan, type OtlpAttribute, type OtlpAttributeValue, type OtlpEvent, type OtlpLink, type OtlpResource, type OtlpResourceSpans, type OtlpScope, type OtlpScopeSpans, type OtlpSpan, SEMANTIC_KINDS, type ScopeTransformer, type SemanticKind, SpanType, type StandardMessageContent, type StandardTextContent, type StandardToolCallContent, type StandardToolResultContent, type ToolCall, TransformerRegistry, TypeClassifier, collectIndices, convertOtlpAttributes, createSignature, detectVersion, extractCustomMetadata, extractIndexedToolCalls, extractReasoningFromProviderMetadata, extractResourceScopeSpan, fetchPromptsFrontmatter, findPromptFiles, generateTypeDefinitions, generateUnique8CharString, messagesToPlainText, normalizeOtlpSpans, normalizeOtlpStatusCode, normalizeSpan, parseAgentMarkAttributes, parseIndexedMessages, parseMetadata, parseTokens, registry, resolveSemanticKind, toFrontMatter, typeClassifier, verifySignature };
+export { AGENTMARK_SCOPE_NAME, AgentMarkTransformer, type AgentmarkConfig, type AgentmarkModelConfig, type AgentmarkModelSchema, type AgentmarkModelSettingsConfig, type AgentmarkModelSettingsSchema, AiSdkTransformer, type AiSdkVersion, type AttributeExtractor, AgentMarkTransformer as ClaudeAgentTransformer, DispatchingTransformer, type GenerateTypesLanguage, type IndexedMessageConfig, MastraTransformer, type McpServerConfig, type McpServers, type McpStdioServerConfig, type McpUrlServerConfig, type Message, type ModelSettingsTypeAspectRatio, type ModelSettingsTypeImageSize, type ModelSettingsTypeSelect, type ModelSettingsTypeSlider, type NormalizedSpan, OpenInferenceTransformer, OpenLLMetryTransformer, type OtelEvent, OtelGenAiTransformer, type OtelLink, type OtelResource, type OtelScope, type OtelSpan, type OtlpAttribute, type OtlpAttributeValue, type OtlpEvent, type OtlpLink, type OtlpResource, type OtlpResourceSpans, type OtlpScope, type OtlpScopeSpans, type OtlpSpan, type RetrievalDocument, SEMANTIC_KINDS, type ScopeTransformer, type SemanticKind, SpanType, type StandardMessageContent, type StandardTextContent, type StandardToolCallContent, type StandardToolResultContent, type ToolCall, TransformerRegistry, TypeClassifier, collectIndices, convertOtlpAttributes, createSignature, detectVersion, extractCustomMetadata, extractIndexedToolCalls, extractReasoningFromProviderMetadata, extractResourceScopeSpan, fetchPromptsFrontmatter, findPromptFiles, generateTypeDefinitions, generateUnique8CharString, hasVectorStoreSignature, messagesToPlainText, normalizeOtlpSpans, normalizeOtlpStatusCode, normalizeSpan, parseAgentMarkAttributes, parseIndexedMessages, parseMetadata, parseTokens, registry, resolveSemanticKind, retrievalDocumentsToText, toFrontMatter, typeClassifier, verifySignature };

package/dist/index.d.ts

@@ -175,6 +175,23 @@ interface ToolCall {
     result?: string;
     providerMetadata?: Record<string, any>;
 }
+/**
+ * A single document returned by a retriever / vector-store search, normalized
+ * from instrumentation that carries documents per-index (OpenInference's
+ * flattened `retrieval.documents.{i}.document.*`) or per-event (OpenLLMetry's
+ * `db.query.result` span events). Stored under a span's `outputObject` as
+ * `{ documents: RetrievalDocument[] }` so the UI can render a ranked panel
+ * (relevance score / distance, content, metadata) instead of a single blob.
+ */
+interface RetrievalDocument {
+    id?: string;
+    content?: string;
+    /** Relevance/similarity score (higher = more relevant). */
+    score?: number;
+    /** Vector distance (lower = closer). Some stores report this instead of a score. */
+    distance?: number;
+    metadata?: Record<string, any>;
+}
 interface OtelScope {
     name?: string;
     version?: string;
@@ -537,6 +554,17 @@ declare function messagesToPlainText(messages: Message[] | undefined): string |
 /** Valid semantic kind values. */
 declare const SEMANTIC_KINDS: readonly ["function", "llm", "tool", "agent", "retrieval", "embedding", "guardrail"];
 type SemanticKind = typeof SEMANTIC_KINDS[number];
+/**
+ * True when a span looks like a vector-store query: a recognized vector-DB
+ * `db.system`, any `db.vector.query.*` attribute, or `db.query.result` /
+ * `db.search.result` events. Distinguishes a vector search from a plain SQL
+ * query (which carries none of these). Shared by the semantic-kind resolver
+ * (→ classifies as "retrieval") and the dispatching transformer (→ routes to
+ * the OpenLLMetry extractor so the result events become documents).
+ */
+declare function hasVectorStoreSignature(attributes: Record<string, any>, events?: ReadonlyArray<{
+    name: string;
+}>): boolean;
 /**
  * Resolve the semantic kind of a span using a 9-level priority chain.
  *
@@ -660,6 +688,8 @@ declare class OtelGenAiTransformer implements ScopeTransformer {
  * @see https://github.com/Arize-ai/openinference/blob/main/spec/semantic_conventions.md
  */
 
+/** Join document contents into a readable, searchable text output. */
+declare function retrievalDocumentsToText(documents: RetrievalDocument[]): string | undefined;
 declare class OpenInferenceTransformer implements ScopeTransformer {
     classify(_span: OtelSpan, attributes: Record<string, any>): SpanType;
     transform(_span: OtelSpan, attributes: Record<string, any>): Partial<NormalizedSpan>;
@@ -710,11 +740,15 @@ declare class DispatchingTransformer implements ScopeTransformer {
     private readonly openInference;
     private readonly openLLMetry;
     private readonly otelGenAi;
-    /** Choose the extractor for a span from its attribute signature. OpenInference
-     *  is checked before OpenLLMetry because its markers (`llm.*`,
-     *  `openinference.span.kind`) are more specific; the bare OTel GenAI
-     *  transformer is the catch-all. */
-    select(attributes: Record<string, any>): ScopeTransformer;
+    /** Choose the extractor for a span from its attribute/event signature.
+     *  OpenInference is checked before OpenLLMetry because its markers (`llm.*`,
+     *  `openinference.span.kind`) are more specific. Vector-store query spans
+     *  route to the OpenLLMetry extractor — they may carry no `traceloop.*`
+     *  marker (a bare Pinecone span is just `db.system` + `db.query.result`
+     *  events), so without this they'd fall through to the OTel-GenAI catch-all
+     *  and their result documents would be dropped. The bare OTel GenAI
+     *  transformer is the final catch-all. */
+    select(attributes: Record<string, any>, span?: OtelSpan): ScopeTransformer;
     classify(span: OtelSpan, attributes: Record<string, any>): SpanType;
     transform(span: OtelSpan, attributes: Record<string, any>): Partial<NormalizedSpan>;
 }
@@ -727,4 +761,4 @@ declare function normalizeSpan(resource: OtelResource, scope: OtelScope, span: O
  */
 declare function normalizeOtlpSpans(resourceSpans: OtlpResourceSpans[]): NormalizedSpan[];
 
-export { AGENTMARK_SCOPE_NAME, AgentMarkTransformer, type AgentmarkConfig, type AgentmarkModelConfig, type AgentmarkModelSchema, type AgentmarkModelSettingsConfig, type AgentmarkModelSettingsSchema, AiSdkTransformer, type AiSdkVersion, type AttributeExtractor, AgentMarkTransformer as ClaudeAgentTransformer, DispatchingTransformer, type GenerateTypesLanguage, type IndexedMessageConfig, MastraTransformer, type McpServerConfig, type McpServers, type McpStdioServerConfig, type McpUrlServerConfig, type Message, type ModelSettingsTypeAspectRatio, type ModelSettingsTypeImageSize, type ModelSettingsTypeSelect, type ModelSettingsTypeSlider, type NormalizedSpan, OpenInferenceTransformer, OpenLLMetryTransformer, type OtelEvent, OtelGenAiTransformer, type OtelLink, type OtelResource, type OtelScope, type OtelSpan, type OtlpAttribute, type OtlpAttributeValue, type OtlpEvent, type OtlpLink, type OtlpResource, type OtlpResourceSpans, type OtlpScope, type OtlpScopeSpans, type OtlpSpan, SEMANTIC_KINDS, type ScopeTransformer, type SemanticKind, SpanType, type StandardMessageContent, type StandardTextContent, type StandardToolCallContent, type StandardToolResultContent, type ToolCall, TransformerRegistry, TypeClassifier, collectIndices, convertOtlpAttributes, createSignature, detectVersion, extractCustomMetadata, extractIndexedToolCalls, extractReasoningFromProviderMetadata, extractResourceScopeSpan, fetchPromptsFrontmatter, findPromptFiles, generateTypeDefinitions, generateUnique8CharString, messagesToPlainText, normalizeOtlpSpans, normalizeOtlpStatusCode, normalizeSpan, parseAgentMarkAttributes, parseIndexedMessages, parseMetadata, parseTokens, registry, resolveSemanticKind, toFrontMatter, typeClassifier, verifySignature };
+export { AGENTMARK_SCOPE_NAME, AgentMarkTransformer, type AgentmarkConfig, type AgentmarkModelConfig, type AgentmarkModelSchema, type AgentmarkModelSettingsConfig, type AgentmarkModelSettingsSchema, AiSdkTransformer, type AiSdkVersion, type AttributeExtractor, AgentMarkTransformer as ClaudeAgentTransformer, DispatchingTransformer, type GenerateTypesLanguage, type IndexedMessageConfig, MastraTransformer, type McpServerConfig, type McpServers, type McpStdioServerConfig, type McpUrlServerConfig, type Message, type ModelSettingsTypeAspectRatio, type ModelSettingsTypeImageSize, type ModelSettingsTypeSelect, type ModelSettingsTypeSlider, type NormalizedSpan, OpenInferenceTransformer, OpenLLMetryTransformer, type OtelEvent, OtelGenAiTransformer, type OtelLink, type OtelResource, type OtelScope, type OtelSpan, type OtlpAttribute, type OtlpAttributeValue, type OtlpEvent, type OtlpLink, type OtlpResource, type OtlpResourceSpans, type OtlpScope, type OtlpScopeSpans, type OtlpSpan, type RetrievalDocument, SEMANTIC_KINDS, type ScopeTransformer, type SemanticKind, SpanType, type StandardMessageContent, type StandardTextContent, type StandardToolCallContent, type StandardToolResultContent, type ToolCall, TransformerRegistry, TypeClassifier, collectIndices, convertOtlpAttributes, createSignature, detectVersion, extractCustomMetadata, extractIndexedToolCalls, extractReasoningFromProviderMetadata, extractResourceScopeSpan, fetchPromptsFrontmatter, findPromptFiles, generateTypeDefinitions, generateUnique8CharString, hasVectorStoreSignature, messagesToPlainText, normalizeOtlpSpans, normalizeOtlpStatusCode, normalizeSpan, parseAgentMarkAttributes, parseIndexedMessages, parseMetadata, parseTokens, registry, resolveSemanticKind, retrievalDocumentsToText, toFrontMatter, typeClassifier, verifySignature };

package/dist/index.js

@@ -56,6 +56,7 @@ __export(index_exports, {
   findPromptFiles: () => findPromptFiles,
   generateTypeDefinitions: () => generateTypeDefinitions,
   generateUnique8CharString: () => generateUnique8CharString,
+  hasVectorStoreSignature: () => hasVectorStoreSignature,
   messagesToPlainText: () => messagesToPlainText,
   normalizeOtlpSpans: () => normalizeOtlpSpans,
   normalizeOtlpStatusCode: () => normalizeOtlpStatusCode,
@@ -66,6 +67,7 @@ __export(index_exports, {
   parseTokens: () => parseTokens,
   registry: () => registry,
   resolveSemanticKind: () => resolveSemanticKind,
+  retrievalDocumentsToText: () => retrievalDocumentsToText,
   toFrontMatter: () => toFrontMatter,
   typeClassifier: () => typeClassifier,
   verifySignature: () => verifySignature
@@ -2441,13 +2443,38 @@ function extractGenericOutput(attributes) {
   }
   return { output: String(value) };
 }
+function parseDocumentMetadata(raw) {
+  let parsed = raw;
+  if (typeof raw === "string") {
+    try {
+      parsed = JSON.parse(raw);
+    } catch {
+      return void 0;
+    }
+  }
+  if (parsed && typeof parsed === "object" && !Array.isArray(parsed)) return parsed;
+  return void 0;
+}
 function extractRetrievalDocuments(attributes) {
   const docs = [];
   for (const i of collectIndices(attributes, RETRIEVAL_PREFIX)) {
-    const content = attributes[`${RETRIEVAL_PREFIX}.${i}.document.content`];
-    if (typeof content === "string" && content.length > 0) docs.push(content);
-  }
-  return docs.length > 0 ? docs.join("\n\n") : void 0;
+    const base = `${RETRIEVAL_PREFIX}.${i}.document`;
+    const doc = {};
+    const id = attributes[`${base}.id`];
+    if (id !== void 0 && id !== null && String(id).length > 0) doc.id = String(id);
+    const content = attributes[`${base}.content`];
+    if (typeof content === "string" && content.length > 0) doc.content = content;
+    const score = toNumber(attributes[`${base}.score`]);
+    if (score !== void 0) doc.score = score;
+    const metadata = parseDocumentMetadata(attributes[`${base}.metadata`]);
+    if (metadata !== void 0) doc.metadata = metadata;
+    if (Object.keys(doc).length > 0) docs.push(doc);
+  }
+  return docs;
+}
+function retrievalDocumentsToText(documents) {
+  const text = documents.map((d) => d.content).filter((c) => typeof c === "string" && c.length > 0).join("\n\n");
+  return text.length > 0 ? text : void 0;
 }
 var OpenInferenceTransformer = class {
   classify(_span, attributes) {
@@ -2491,9 +2518,13 @@ var OpenInferenceTransformer = class {
     } else {
       Object.assign(result, extractGenericOutput(attributes));
     }
-    if (result.output === void 0) {
-      const docs = extractRetrievalDocuments(attributes);
-      if (docs) result.output = docs;
+    const documents = extractRetrievalDocuments(attributes);
+    if (documents.length > 0) {
+      if (result.outputObject === void 0) result.outputObject = { documents };
+      if (result.output === void 0) {
+        const text = retrievalDocumentsToText(documents);
+        if (text) result.output = text;
+      }
     }
     const toolName = attributes[Attrs2.TOOL_NAME];
     if (typeof toolName === "string" && toolName.length > 0) result.name = toolName;
@@ -2595,6 +2626,51 @@ function extractFromEvents(events, attrKey) {
   }
   return void 0;
 }
+function parseResultMetadata(raw) {
+  let parsed = raw;
+  if (typeof raw === "string") {
+    try {
+      parsed = JSON.parse(raw);
+    } catch {
+      return void 0;
+    }
+  }
+  if (parsed && typeof parsed === "object" && !Array.isArray(parsed)) return parsed;
+  return void 0;
+}
+var VECTOR_RESULT_EVENT_NAMES = /* @__PURE__ */ new Set(["db.query.result", "db.search.result"]);
+function extractVectorResultDocuments(events) {
+  var _a, _b;
+  if (!events) return [];
+  const docs = [];
+  for (const event of events) {
+    if (!VECTOR_RESULT_EVENT_NAMES.has(event.name)) continue;
+    const a = (_a = event.attributes) != null ? _a : {};
+    const get = (field) => {
+      var _a2;
+      return (_a2 = a[`db.query.result.${field}`]) != null ? _a2 : a[`db.search.result.${field}`];
+    };
+    const doc = {};
+    const id = get("id");
+    if (id !== void 0 && id !== null && String(id).length > 0) doc.id = String(id);
+    const content = get("document");
+    if (typeof content === "string" && content.length > 0) doc.content = content;
+    const score = toNumber(get("score"));
+    if (score !== void 0) doc.score = score;
+    const distance = toNumber(get("distance"));
+    if (distance !== void 0) doc.distance = distance;
+    let metadata = parseResultMetadata(get("metadata"));
+    if (metadata === void 0) {
+      const entity = get("entity");
+      if (entity !== void 0 && entity !== null && String(entity).length > 0) {
+        metadata = (_b = parseResultMetadata(entity)) != null ? _b : { entity: String(entity) };
+      }
+    }
+    if (metadata !== void 0) doc.metadata = metadata;
+    if (Object.keys(doc).length > 0) docs.push(doc);
+  }
+  return docs;
+}
 function extractFinishReason(attributes) {
   const direct = attributes[Attrs3.RESPONSE_FINISH_REASON];
   if (direct !== void 0) {
@@ -2664,6 +2740,16 @@ var OpenLLMetryTransformer = class {
         if (eventCompletion) result.output = eventCompletion;
       }
     }
+    if (result.outputObject === void 0) {
+      const documents = extractVectorResultDocuments(span.events);
+      if (documents.length > 0) {
+        result.outputObject = { documents };
+        if (result.output === void 0) {
+          const text = documents.map((d) => d.content).filter((c) => typeof c === "string" && c.length > 0).join("\n\n");
+          if (text) result.output = text;
+        }
+      }
+    }
     const entityName = attributes[Attrs3.ENTITY_NAME];
     if (typeof entityName === "string" && entityName.length > 0) result.traceName = entityName;
     const metadata = {};
@@ -2683,6 +2769,115 @@ var OpenLLMetryTransformer = class {
   }
 };
 
+// src/normalizer/resolvers/semantic-kind-resolver.ts
+var SEMANTIC_KINDS = ["function", "llm", "tool", "agent", "retrieval", "embedding", "guardrail"];
+var VALID_KINDS = new Set(SEMANTIC_KINDS);
+var OPENINFERENCE_MAP = {
+  "CHAIN": "function",
+  "LLM": "llm",
+  "TOOL": "tool",
+  "AGENT": "agent",
+  "RETRIEVER": "retrieval",
+  "EMBEDDING": "embedding",
+  "GUARDRAIL": "guardrail",
+  "RERANKER": "retrieval"
+};
+var VECTOR_DB_SYSTEMS = /* @__PURE__ */ new Set([
+  "pinecone",
+  "qdrant",
+  "weaviate",
+  "milvus",
+  "chroma",
+  "chromadb",
+  "marqo",
+  "lancedb"
+]);
+var VECTOR_RESULT_EVENT_NAMES2 = /* @__PURE__ */ new Set(["db.query.result", "db.search.result"]);
+function hasVectorStoreSignature(attributes, events) {
+  var _a;
+  const dbSystem = (_a = attributes["db.system"]) != null ? _a : attributes["db.system.name"];
+  if (typeof dbSystem === "string" && VECTOR_DB_SYSTEMS.has(dbSystem.toLowerCase())) return true;
+  if (Object.keys(attributes).some((k) => k.startsWith("db.vector.query."))) return true;
+  if (events == null ? void 0 : events.some((e) => VECTOR_RESULT_EVENT_NAMES2.has(e.name))) return true;
+  return false;
+}
+var FRAMEWORK_MAPPINGS = [
+  {
+    key: "ai.operationId",
+    // Vercel AI SDK
+    // The AI SDK emits ai.operationId WITH the "ai." prefix (e.g.
+    // "ai.generateText"); accept both prefixed and unprefixed so generation
+    // wrappers resolve to "llm" instead of falling through to "function".
+    map: {
+      "embed": "embedding",
+      "ai.embed": "embedding",
+      "generateText": "llm",
+      "ai.generateText": "llm",
+      "streamText": "llm",
+      "ai.streamText": "llm",
+      "generateObject": "llm",
+      "ai.generateObject": "llm",
+      "streamObject": "llm",
+      "ai.streamObject": "llm"
+    }
+  },
+  {
+    key: "traceloop.span.kind",
+    // Traceloop / OpenLLMetry
+    map: { "LLM": "llm", "TOOL": "tool", "AGENT": "agent", "WORKFLOW": "function", "TASK": "function" }
+  },
+  {
+    key: "langchain.run_type",
+    // LangChain via OTLP
+    map: { "llm": "llm", "chat_model": "llm", "retriever": "retrieval", "tool": "tool", "chain": "function", "embedding": "embedding" }
+  },
+  {
+    key: "genkit:type",
+    // Firebase Genkit
+    map: { "model": "llm", "tool": "tool", "flow": "function", "retriever": "retrieval", "embedder": "embedding" }
+  }
+];
+function resolveSemanticKind(normalized, allAttributes) {
+  if (normalized.semanticKind && VALID_KINDS.has(normalized.semanticKind)) {
+    return normalized.semanticKind;
+  }
+  const oiKind = allAttributes["openinference.span.kind"];
+  if (oiKind) {
+    const mapped = OPENINFERENCE_MAP[String(oiKind).toUpperCase()];
+    if (mapped) return mapped;
+  }
+  for (const { key, map } of FRAMEWORK_MAPPINGS) {
+    const val = allAttributes[key];
+    if (val) {
+      const mapped = map[String(val)];
+      if (mapped) return mapped;
+    }
+  }
+  if (hasVectorStoreSignature(allAttributes, normalized.events)) {
+    return "retrieval";
+  }
+  const opName = allAttributes["gen_ai.operation.name"];
+  if (opName) {
+    const op = String(opName).toLowerCase();
+    if (op === "chat" || op === "text_completion" || op === "generate_content") return "llm";
+    if (op === "embeddings") return "embedding";
+  }
+  if (normalized.type === "GENERATION" /* GENERATION */) {
+    return "llm";
+  }
+  if (normalized.model) {
+    return "llm";
+  }
+  if (normalized.toolCalls && normalized.toolCalls.length > 0) {
+    return "tool";
+  }
+  const name = (normalized.name || "").toLowerCase();
+  if (/retriev|search|rag/i.test(name)) return "retrieval";
+  if (/embed/i.test(name)) return "embedding";
+  if (/guard|safety/i.test(name)) return "guardrail";
+  return "function";
+}
+
 // src/normalizer/transformers/dispatching/index.ts
 var OPENINFERENCE_INDEXED = /^llm\.(input_messages|output_messages|token_count)\./;
 var OPENLLMETRY_INDEXED = /^gen_ai\.(prompt|completion)\.\d+\./;
@@ -2707,20 +2902,25 @@ var DispatchingTransformer = class {
     this.openLLMetry = new OpenLLMetryTransformer();
     this.otelGenAi = new OtelGenAiTransformer();
   }
-  /** Choose the extractor for a span from its attribute signature. OpenInference
-   *  is checked before OpenLLMetry because its markers (`llm.*`,
-   *  `openinference.span.kind`) are more specific; the bare OTel GenAI
-   *  transformer is the catch-all. */
-  select(attributes) {
+  /** Choose the extractor for a span from its attribute/event signature.
+   *  OpenInference is checked before OpenLLMetry because its markers (`llm.*`,
+   *  `openinference.span.kind`) are more specific. Vector-store query spans
+   *  route to the OpenLLMetry extractor — they may carry no `traceloop.*`
+   *  marker (a bare Pinecone span is just `db.system` + `db.query.result`
+   *  events), so without this they'd fall through to the OTel-GenAI catch-all
+   *  and their result documents would be dropped. The bare OTel GenAI
+   *  transformer is the final catch-all. */
+  select(attributes, span) {
     if (isOpenInference(attributes)) return this.openInference;
     if (isOpenLLMetry(attributes)) return this.openLLMetry;
+    if (hasVectorStoreSignature(attributes, span == null ? void 0 : span.events)) return this.openLLMetry;
     return this.otelGenAi;
   }
   classify(span, attributes) {
-    return this.select(attributes).classify(span, attributes);
+    return this.select(attributes, span).classify(span, attributes);
   }
   transform(span, attributes) {
-    return this.select(attributes).transform(span, attributes);
+    return this.select(attributes, span).transform(span, attributes);
   }
 };
 
@@ -2804,93 +3004,6 @@ function extractResourceScopeSpan(resourceSpans) {
   return result;
 }
 
-// src/normalizer/resolvers/semantic-kind-resolver.ts
-var SEMANTIC_KINDS = ["function", "llm", "tool", "agent", "retrieval", "embedding", "guardrail"];
-var VALID_KINDS = new Set(SEMANTIC_KINDS);
-var OPENINFERENCE_MAP = {
-  "CHAIN": "function",
-  "LLM": "llm",
-  "TOOL": "tool",
-  "AGENT": "agent",
-  "RETRIEVER": "retrieval",
-  "EMBEDDING": "embedding",
-  "GUARDRAIL": "guardrail",
-  "RERANKER": "retrieval"
-};
-var FRAMEWORK_MAPPINGS = [
-  {
-    key: "ai.operationId",
-    // Vercel AI SDK
-    // The AI SDK emits ai.operationId WITH the "ai." prefix (e.g.
-    // "ai.generateText"); accept both prefixed and unprefixed so generation
-    // wrappers resolve to "llm" instead of falling through to "function".
-    map: {
-      "embed": "embedding",
-      "ai.embed": "embedding",
-      "generateText": "llm",
-      "ai.generateText": "llm",
-      "streamText": "llm",
-      "ai.streamText": "llm",
-      "generateObject": "llm",
-      "ai.generateObject": "llm",
-      "streamObject": "llm",
-      "ai.streamObject": "llm"
-    }
-  },
-  {
-    key: "traceloop.span.kind",
-    // Traceloop / OpenLLMetry
-    map: { "LLM": "llm", "TOOL": "tool", "AGENT": "agent", "WORKFLOW": "function", "TASK": "function" }
-  },
-  {
-    key: "langchain.run_type",
-    // LangChain via OTLP
-    map: { "llm": "llm", "chat_model": "llm", "retriever": "retrieval", "tool": "tool", "chain": "function", "embedding": "embedding" }
-  },
-  {
-    key: "genkit:type",
-    // Firebase Genkit
-    map: { "model": "llm", "tool": "tool", "flow": "function", "retriever": "retrieval", "embedder": "embedding" }
-  }
-];
-function resolveSemanticKind(normalized, allAttributes) {
-  if (normalized.semanticKind && VALID_KINDS.has(normalized.semanticKind)) {
-    return normalized.semanticKind;
-  }
-  const oiKind = allAttributes["openinference.span.kind"];
-  if (oiKind) {
-    const mapped = OPENINFERENCE_MAP[String(oiKind).toUpperCase()];
-    if (mapped) return mapped;
-  }
-  for (const { key, map } of FRAMEWORK_MAPPINGS) {
-    const val = allAttributes[key];
-    if (val) {
-      const mapped = map[String(val)];
-      if (mapped) return mapped;
-    }
-  }
-  const opName = allAttributes["gen_ai.operation.name"];
-  if (opName) {
-    const op = String(opName).toLowerCase();
-    if (op === "chat" || op === "text_completion" || op === "generate_content") return "llm";
-    if (op === "embeddings") return "embedding";
-  }
-  if (normalized.type === "GENERATION" /* GENERATION */) {
-    return "llm";
-  }
-  if (normalized.model) {
-    return "llm";
-  }
-  if (normalized.toolCalls && normalized.toolCalls.length > 0) {
-    return "tool";
-  }
-  const name = (normalized.name || "").toLowerCase();
-  if (/retriev|search|rag/i.test(name)) return "retrieval";
-  if (/embed/i.test(name)) return "embedding";
-  if (/guard|safety/i.test(name)) return "guardrail";
-  return "function";
-}
-
 // src/normalizer/type-classifier.ts
 var TypeClassifier = class {
   classify(span, attributes) {
@@ -3059,6 +3172,7 @@ function deriveTraceIO(spans) {
   findPromptFiles,
   generateTypeDefinitions,
   generateUnique8CharString,
+  hasVectorStoreSignature,
   messagesToPlainText,
   normalizeOtlpSpans,
   normalizeOtlpStatusCode,
@@ -3069,6 +3183,7 @@ function deriveTraceIO(spans) {
   parseTokens,
   registry,
   resolveSemanticKind,
+  retrievalDocumentsToText,
   toFrontMatter,
   typeClassifier,
   verifySignature