braintrust 3.12.0 → 3.13.0

package/dist/workerd.mjs

@@ -89,6 +89,7 @@ var iso = {
   getRepoInfo: async (_settings) => void 0,
   getPastNAncestors: async () => [],
   getEnv: (_name) => void 0,
+  getBraintrustApiKey: async () => void 0,
   getCallerLocation: () => void 0,
   newAsyncLocalStorage: () => new DefaultAsyncLocalStorage(),
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
@@ -5430,6 +5431,19 @@ var JSONAttachment = class extends Attachment {
    */
   constructor(data, options) {
     const { filename = "data.json", pretty = false, state } = options ?? {};
+    const deferredJsonAttachment = globalThis.__BT_DATASET_PIPELINE_DEFER_JSON_ATTACHMENT__;
+    if (deferredJsonAttachment) {
+      super({
+        data: new Blob([]),
+        filename,
+        contentType: "application/json",
+        state
+      });
+      return deferredJsonAttachment(data, {
+        filename,
+        pretty
+      });
+    }
     const jsonString = pretty ? JSON.stringify(data, null, 2) : JSON.stringify(data);
     const blob = new Blob([jsonString], { type: "application/json" });
     super({
@@ -7422,10 +7436,11 @@ async function login(options = {}) {
 async function loginToState(options = {}) {
   const {
     appUrl = isomorph_default.getEnv("BRAINTRUST_APP_URL") || "https://www.braintrust.dev",
-    apiKey = isomorph_default.getEnv("BRAINTRUST_API_KEY"),
+    apiKey: apiKeyArg,
     orgName = isomorph_default.getEnv("BRAINTRUST_ORG_NAME"),
     fetch: fetch2 = globalThis.fetch
   } = options || {};
+  const apiKey = apiKeyArg !== void 0 ? apiKeyArg : await isomorph_default.getBraintrustApiKey();
   const appPublicUrl = isomorph_default.getEnv("BRAINTRUST_APP_PUBLIC_URL") || appUrl;
   const state = new BraintrustState(options);
   state.resetLoginInfo();
@@ -8666,9 +8681,15 @@ var SpanImpl = class _SpanImpl {
       const cachedSpan = {
         input: partialRecord.input,
         output: partialRecord.output,
+        expected: partialRecord.expected,
+        error: partialRecord.error,
+        scores: partialRecord.scores,
+        metrics: partialRecord.metrics,
         metadata: partialRecord.metadata,
+        tags: partialRecord.tags,
         span_id: this._spanId,
         span_parents: this._spanParents,
+        is_root: this._spanId === this._rootSpanId,
         span_attributes: partialRecord.span_attributes
       };
       this._state.spanCache.queueWrite(
@@ -9004,6 +9025,7 @@ var Dataset2 = class extends ObjectFetcher {
     metadata,
     tags,
     output,
+    origin,
     isMerge
   }) {
     return new LazyValue(async () => {
@@ -9018,6 +9040,7 @@ var Dataset2 = class extends ObjectFetcher {
         created: !isMerge ? (/* @__PURE__ */ new Date()).toISOString() : void 0,
         //if we're merging/updating an event we will not add this ts
         metadata,
+        origin,
         ...!!isMerge ? {
           [IS_MERGE_FIELD]: true
         } : {}
@@ -9037,6 +9060,7 @@ var Dataset2 = class extends ObjectFetcher {
    * about anything else that's relevant, that you can use to help find and analyze examples later. For example, you could log the
    * `prompt`, example's `id`, or anything else that would be useful to slice/dice later. The values in `metadata` can be any
    * JSON-serializable type, but its keys must be strings.
+   * @param event.origin (Optional) a reference to the source object this dataset record was derived from.
    * @param event.id (Optional) a unique identifier for the event. If you don't provide one, Braintrust will generate one for you.
    * @param event.output: (Deprecated) The output of your application. Use `expected` instead.
    * @returns The `id` of the logged record.
@@ -9047,7 +9071,8 @@ var Dataset2 = class extends ObjectFetcher {
     metadata,
     tags,
     id,
-    output
+    output,
+    origin
   }) {
     this.validateEvent({ metadata, expected, output, tags });
     const rowId = id || uuidv42();
@@ -9059,6 +9084,7 @@ var Dataset2 = class extends ObjectFetcher {
         metadata,
         tags,
         output,
+        origin,
         isMerge: false
       })
     );
@@ -26054,6 +26080,10 @@ function configureWorkerd() {
     }
     return process.env[name];
   };
+  isomorph_default.getBraintrustApiKey = async () => {
+    const value = isomorph_default.getEnv("BRAINTRUST_API_KEY");
+    return value?.trim() ? value : void 0;
+  };
   isomorph_default.hash = (data) => {
     let hash = 0;
     for (let i = 0; i < data.length; i++) {
@@ -26089,6 +26119,7 @@ __export(exports_exports, {
   DEFAULT_FETCH_BATCH_SIZE: () => DEFAULT_FETCH_BATCH_SIZE,
   DEFAULT_MAX_REQUEST_SIZE: () => DEFAULT_MAX_REQUEST_SIZE,
   Dataset: () => Dataset2,
+  DatasetPipeline: () => DatasetPipeline,
   ERR_PERMALINK: () => ERR_PERMALINK,
   Eval: () => Eval,
   EvalResultWithSummary: () => EvalResultWithSummary,
@@ -29347,10 +29378,12 @@ function formatExperimentSummary(summary) {
 // src/wrappers/shared/flush.ts
 async function summarizeAndFlush(experiment, options) {
   const shouldDisplay = options.displaySummary ?? true;
-  const summary = await experiment.summarize();
-  if (shouldDisplay) {
-    console.log(formatExperimentSummary(summary));
+  if (!shouldDisplay) {
+    await experiment.flush();
+    return;
   }
+  const summary = await experiment.summarize();
+  console.log(formatExperimentSummary(summary));
 }
 
 // src/wrappers/vitest/flush-manager.ts
@@ -31214,8 +31247,12 @@ var waterfall$1 = awaitify(waterfall);
 
 // src/trace.ts
 var SpanFetcher = class _SpanFetcher extends ObjectFetcher {
-  constructor(objectType, _objectId, rootSpanId, _state, spanTypeFilter) {
-    const filterExpr = _SpanFetcher.buildFilter(rootSpanId, spanTypeFilter);
+  constructor(objectType, _objectId, rootSpanId, _state, spanTypeFilter, includeScorers = false) {
+    const filterExpr = _SpanFetcher.buildFilter(
+      rootSpanId,
+      spanTypeFilter,
+      includeScorers
+    );
     super(objectType, void 0, void 0, {
       filter: filterExpr
     });
@@ -31224,16 +31261,17 @@ var SpanFetcher = class _SpanFetcher extends ObjectFetcher {
     this._state = _state;
     this.spanTypeFilter = spanTypeFilter;
   }
-  static buildFilter(rootSpanId, spanTypeFilter) {
+  static buildFilter(rootSpanId, spanTypeFilter, includeScorers = false) {
     const children = [
       // Base filter: root_span_id = 'value'
       {
         op: "eq",
         left: { op: "ident", name: ["root_span_id"] },
         right: { op: "literal", value: rootSpanId }
-      },
-      // Exclude span_attributes.purpose = 'score'
-      {
+      }
+    ];
+    if (!includeScorers) {
+      children.push({
         op: "or",
         children: [
           {
@@ -31246,8 +31284,8 @@ var SpanFetcher = class _SpanFetcher extends ObjectFetcher {
             right: { op: "literal", value: "scorer" }
           }
         ]
-      }
-    ];
+      });
+    }
     if (spanTypeFilter && spanTypeFilter.length > 0) {
       children.push({
         op: "in",
@@ -31273,35 +31311,49 @@ var CachedSpanFetcher = class {
   fetchFn;
   constructor(objectTypeOrFetchFn, objectId, rootSpanId, getState) {
     if (typeof objectTypeOrFetchFn === "function") {
-      this.fetchFn = objectTypeOrFetchFn;
+      this.fetchFn = (spanType) => objectTypeOrFetchFn(spanType);
     } else {
       const objectType = objectTypeOrFetchFn;
-      this.fetchFn = async (spanType) => {
+      this.fetchFn = async (spanType, includeScorers) => {
         const state = await getState();
         const fetcher = new SpanFetcher(
           objectType,
           objectId,
           rootSpanId,
           state,
-          spanType
+          spanType,
+          includeScorers
         );
         const rows = await fetcher.fetchedData();
-        return rows.filter((row) => row.span_attributes?.purpose !== "scorer").map((row) => ({
+        return rows.map((row) => ({
           input: row.input,
           output: row.output,
+          expected: row.expected,
+          error: row.error,
+          scores: row.scores,
+          metrics: row.metrics,
           metadata: row.metadata,
           span_id: row.span_id,
           span_parents: row.span_parents,
+          is_root: row.is_root,
           span_attributes: row.span_attributes,
           id: row.id,
           _xact_id: row._xact_id,
           _pagination_key: row._pagination_key,
-          root_span_id: row.root_span_id
+          root_span_id: row.root_span_id,
+          created: row.created,
+          tags: row.tags
         }));
       };
     }
   }
-  async getSpans({ spanType } = {}) {
+  async getSpans({
+    spanType,
+    includeScorers = false
+  } = {}) {
+    if (includeScorers) {
+      return this.fetchFn(spanType, true);
+    }
     if (this.allFetched) {
       return this.getFromCache(spanType);
     }
@@ -31318,7 +31370,7 @@ var CachedSpanFetcher = class {
     return this.getFromCache(spanType);
   }
   async fetchSpans(spanType) {
-    const spans = await this.fetchFn(spanType);
+    const spans = await this.fetchFn(spanType, false);
     for (const span of spans) {
       const type = span.span_attributes?.type ?? "";
       const existing = this.spanCache.get(type) ?? [];
@@ -31396,10 +31448,13 @@ var LocalTrace = class {
    * First checks the local span cache for recently logged spans, then falls
    * back to CachedSpanFetcher which handles BTQL fetching and caching.
    */
-  async getSpans({ spanType } = {}) {
+  async getSpans({
+    spanType,
+    includeScorers = false
+  } = {}) {
     const cachedSpans = this.state.spanCache.getByRootSpanId(this.rootSpanId);
     if (cachedSpans && cachedSpans.length > 0) {
-      let spans = cachedSpans.filter(
+      let spans = includeScorers ? cachedSpans : cachedSpans.filter(
         (span) => span.span_attributes?.purpose !== "scorer"
       );
       if (spanType && spanType.length > 0) {
@@ -31410,13 +31465,19 @@ var LocalTrace = class {
       return spans.map((span) => ({
         input: span.input,
         output: span.output,
+        expected: span.expected,
+        error: span.error,
+        scores: span.scores,
+        metrics: span.metrics,
         metadata: span.metadata,
         span_id: span.span_id,
         span_parents: span.span_parents,
-        span_attributes: span.span_attributes
+        is_root: span.is_root,
+        span_attributes: span.span_attributes,
+        tags: span.tags
       }));
     }
-    return this.cachedFetcher.getSpans({ spanType });
+    return this.cachedFetcher.getSpans({ spanType, includeScorers });
   }
   /**
    * Get the thread (preprocessed messages) for this trace.
@@ -32605,6 +32666,34 @@ var defaultReporter = {
   }
 };
 
+// src/dataset-pipeline.ts
+function DatasetPipeline(definition) {
+  if (!globalThis.__braintrust_dataset_pipelines) {
+    globalThis.__braintrust_dataset_pipelines = [];
+  }
+  const storedDefinition = {
+    name: definition.name,
+    source: {
+      projectId: definition.source.projectId,
+      projectName: definition.source.projectName,
+      orgName: definition.source.orgName,
+      filter: definition.source.filter,
+      scope: definition.source.scope ?? "span"
+    },
+    // eslint-disable-next-line @typescript-eslint/consistent-type-assertions, @typescript-eslint/no-explicit-any
+    transform: definition.transform,
+    target: {
+      projectId: definition.target.projectId,
+      projectName: definition.target.projectName,
+      orgName: definition.target.orgName,
+      datasetName: definition.target.datasetName,
+      description: definition.target.description,
+      metadata: definition.target.metadata
+    }
+  };
+  globalThis.__braintrust_dataset_pipelines.push(storedDefinition);
+}
+
 // src/framework2.ts
 import { z as z12 } from "zod/v3";
 var currentFilename = typeof __filename !== "undefined" ? __filename : "unknown";
@@ -33144,6 +33233,7 @@ export {
   DEFAULT_FETCH_BATCH_SIZE,
   DEFAULT_MAX_REQUEST_SIZE,
   Dataset2 as Dataset,
+  DatasetPipeline,
   ERR_PERMALINK,
   Eval,
   EvalResultWithSummary,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "braintrust",
-  "version": "3.12.0",
+  "version": "3.13.0",
   "description": "SDK for integrating Braintrust",
   "repository": {
     "type": "git",
@@ -138,29 +138,16 @@
     "test": "vitest run --exclude \"src/wrappers/**/*.test.ts\" --exclude \"src/otel/**/*.test.ts\" --exclude \"smoke/**/*.test.ts\" --exclude \"src/zod/**/*.test.ts\" --exclude \"tests/api-compatibility/**\"",
     "test:core": "pnpm prune && pnpm test",
     "test:checks": "pnpm run test:core && pnpm run test:vitest",
-    "test:external": "pnpm run test:external:sequential && node scripts/run-parallel.mjs test:external:ai-sdk-v5 test:external:ai-sdk-v6 test:external:claude-agent-sdk",
-    "test:external:sequential": "pnpm run test:external:openai && pnpm run test:external:anthropic && pnpm run test:external:google-genai",
-    "test:external:openai": "bash scripts/test-provider.sh test:openai openai",
-    "test:external:anthropic": "bash scripts/test-provider.sh test:anthropic @anthropic-ai/sdk",
-    "test:external:google-genai": "bash scripts/test-provider.sh test:google-genai @google/genai",
-    "test:external:ai-sdk": "pnpm run test:external:ai-sdk-v5 && pnpm run test:external:ai-sdk-v6",
-    "test:external:ai-sdk-v5": "cd src/wrappers/ai-sdk/tests/v5 && pnpm install --ignore-workspace && pnpm test",
-    "test:external:ai-sdk-v6": "cd src/wrappers/ai-sdk/tests/v6 && pnpm install --ignore-workspace && pnpm test",
-    "test:external:claude-agent-sdk": "cd src/wrappers/claude-agent-sdk && pnpm install && pnpm test",
-    "test:all": "pnpm run test:checks && pnpm run test:external",
+    "test:all": "pnpm run test:checks",
     "test:api-compat": "vitest run tests/api-compatibility/api-compatibility.test.ts",
-    "test:anthropic": "vitest run src/wrappers/anthropic.test.ts",
-    "test:openai": "vitest run src/wrappers/oai.test.ts",
     "test:otel": "vitest run --dir src/otel",
     "test:otel-no-deps": "vitest run src/otel/otel-no-deps.test.ts --reporter=verbose",
-    "test:google-genai": "vitest run src/wrappers/google-genai.test.ts",
     "test:ai-sdk-v1": "vitest run src/wrappers/ai-sdk-v1.test.ts",
     "test:ai-sdk-v2": "vitest run src/wrappers/ai-sdk-v2.test.ts src/wrappers/ai-sdk-v1.test.ts",
     "test:ai-sdk-v3": "vitest run src/wrappers/ai-sdk-v3.test.ts",
     "test:zod-v3": "vitest run src/zod/zod-v3-serialization.test.ts",
     "test:zod-v4": "vitest run src/zod/zod-v4-serialization.test.ts",
     "test:mastra": "vitest run src/wrappers/mastra.test.ts",
-    "test:claude-agent-sdk": "pnpm --filter @braintrust/claude-agent-sdk-tests test",
     "test:vitest": "pnpm --filter @braintrust/vitest-wrapper-tests test",
     "test:output": "tsx scripts/test-output.ts --with-comparison --with-metrics --with-progress",
     "bench": "tsx src/queue.bench.ts",
@@ -191,7 +178,6 @@
     "@types/node": "^20.10.5",
     "@types/pluralize": "^0.0.30",
     "@types/tar": "^6.1.13",
-    "@types/uuid": "^9.0.7",
     "@typescript-eslint/eslint-plugin": "^8.49.0",
     "@typescript-eslint/parser": "^8.49.0",
     "ai": "^6.0.0",
@@ -238,7 +224,7 @@
     "source-map": "^0.7.4",
     "termi-link": "^1.0.1",
     "unplugin": "^2.3.5",
-    "uuid": "^9.0.1",
+    "uuid": "^11.1.1",
     "zod-to-json-schema": "^3.25.0"
   },
   "peerDependencies": {

package/util/dist/index.d.mts

@@ -4842,6 +4842,7 @@ declare const ObjectReference: z.ZodObject<{
     _xact_id?: string | null | undefined;
     created?: string | null | undefined;
 }>;
+type ObjectReferenceType = z.infer<typeof ObjectReference>;
 
 type IdField = {
     id: string;
@@ -4862,7 +4863,7 @@ type OtherExperimentLogFields = {
     metadata: Record<string, unknown>;
     metrics: Record<string, unknown>;
     datasetRecordId: string;
-    origin: z.infer<typeof ObjectReference>;
+    origin: ObjectReferenceType;
     span_attributes: Record<string, unknown>;
     [ASYNC_SCORING_CONTROL_FIELD]: AsyncScoringControlType;
     [MERGE_PATHS_FIELD]: string[][];
@@ -4916,6 +4917,7 @@ type DatasetEvent = {
     tags?: string[];
     metadata?: unknown;
     created?: string;
+    origin?: ObjectReferenceType;
     id: string;
     dataset_id: string;
 } & ({

package/util/dist/index.d.ts

@@ -4842,6 +4842,7 @@ declare const ObjectReference: z.ZodObject<{
     _xact_id?: string | null | undefined;
     created?: string | null | undefined;
 }>;
+type ObjectReferenceType = z.infer<typeof ObjectReference>;
 
 type IdField = {
     id: string;
@@ -4862,7 +4863,7 @@ type OtherExperimentLogFields = {
     metadata: Record<string, unknown>;
     metrics: Record<string, unknown>;
     datasetRecordId: string;
-    origin: z.infer<typeof ObjectReference>;
+    origin: ObjectReferenceType;
     span_attributes: Record<string, unknown>;
     [ASYNC_SCORING_CONTROL_FIELD]: AsyncScoringControlType;
     [MERGE_PATHS_FIELD]: string[][];
@@ -4916,6 +4917,7 @@ type DatasetEvent = {
     tags?: string[];
     metadata?: unknown;
     created?: string;
+    origin?: ObjectReferenceType;
     id: string;
     dataset_id: string;
 } & ({