bitfab 0.39.0 → 0.40.0

package/dist/index.d.cts

@@ -540,7 +540,11 @@ declare class HttpClient {
     lookupFunction<T>(name: string): Promise<T>;
     getAutoTracePolicy<T>(traceFunctionKey: string, protocol: string): Promise<T>;
     getTraceSpan(traceId: string, lookup: SpanLookup): Promise<CapturedSpan | null>;
-    private get;
+    /**
+     * GET a JSON endpoint on the service with the client's API key. Throws a
+     * `BitfabError` carrying the status text for any non-2xx response.
+     */
+    get<T>(endpoint: string): Promise<T>;
     /**
      * Queue an internal trace (from local BAML execution via `call()`) onto this
      * client's batching transport. `functionId` moves into the payload because
@@ -1034,6 +1038,141 @@ declare class BitfabVercelAiHandler {
     get middleware(): BitfabLanguageModelMiddleware;
 }
 
+interface DatasetGraderRef {
+    id: string;
+    name: string | null;
+}
+interface Dataset {
+    id: string;
+    traceFunctionKey: string;
+    name: string;
+    description: string | null;
+    traceCount: number;
+    graders: DatasetGraderRef[];
+    createdAt: string;
+    updatedAt: string;
+}
+interface SaveDatasetParams {
+    traceFunctionKey: string;
+    name: string;
+    description?: string;
+}
+interface SaveDatasetResult {
+    dataset: Dataset;
+    created: boolean;
+}
+interface ListDatasetsParams {
+    traceFunctionKey?: string;
+}
+interface DatasetTraceIds {
+    datasetId: string;
+    traceIds: string[];
+}
+interface AddDatasetTracesResult {
+    dataset: Dataset;
+    addedTraceIds: string[];
+    alreadyPresentTraceIds: string[];
+    skippedTraceIds: string[];
+}
+interface RemoveDatasetTracesResult {
+    dataset: Dataset;
+    removedTraceIds: string[];
+    notPresentTraceIds: string[];
+}
+interface AddDatasetGradersResult {
+    dataset: Dataset;
+    addedGraderIds: string[];
+    alreadyAssignedGraderIds: string[];
+    skippedGraderIds: string[];
+}
+interface RemoveDatasetGradersResult {
+    dataset: Dataset;
+    removedGraderIds: string[];
+    notAssignedGraderIds: string[];
+}
+type GraderRerunStatus = "pending" | "running" | "completed" | "errored";
+interface GraderRerunProgress {
+    completedTraces: number;
+    totalTraces: number;
+    graderCount: number;
+}
+interface GraderRerunResult {
+    tracesGraded: number;
+    gradersRun: number;
+}
+interface GraderRerun {
+    id: string;
+    status: GraderRerunStatus;
+    graderIds: string[];
+    progress: GraderRerunProgress | null;
+    result: GraderRerunResult | null;
+    error: string | null;
+    createdAt: string;
+    updatedAt: string;
+}
+interface RerunGradersOptions {
+    graderIds?: string[];
+    wait?: boolean;
+    timeoutMs?: number;
+    pollIntervalMs?: number;
+}
+interface RerunGradersResult {
+    run: GraderRerun;
+    joinedExisting: boolean;
+}
+/**
+ * Dataset operations for the authenticated organization, reached as
+ * `client.datasets`. A dataset is a named bucket of traces scoped to one trace
+ * function. Experiments replay against it and its graders score its members.
+ */
+declare class DatasetsClient {
+    private readonly httpClient;
+    constructor(httpClient: HttpClient);
+    /**
+     * Create a dataset, or update the one already named this way under the same
+     * trace function. `created` reports which happened. An omitted description
+     * leaves an existing one untouched.
+     */
+    save(params: SaveDatasetParams): Promise<SaveDatasetResult>;
+    /**
+     * List datasets, scoped to one trace function when `traceFunctionKey` is
+     * given and organization-wide otherwise.
+     */
+    list(params?: ListDatasetsParams): Promise<Dataset[]>;
+    /** Fetch one dataset by id. Rejects with a 404 `BitfabError` when it is not in this organization. */
+    get(datasetId: string): Promise<Dataset>;
+    /** The ids of every trace in the dataset, the same membership a replay with `datasetId` selects. */
+    listTraces(datasetId: string): Promise<DatasetTraceIds>;
+    /**
+     * Add traces to the dataset (1 to 100 ids per call). Traces outside the
+     * organization or under another trace function are reported in
+     * `skippedTraceIds` rather than failing the call.
+     */
+    addTraces(datasetId: string, traceIds: string[]): Promise<AddDatasetTracesResult>;
+    /** Remove traces from the dataset. The traces themselves are never deleted. */
+    removeTraces(datasetId: string, traceIds: string[]): Promise<RemoveDatasetTracesResult>;
+    /**
+     * Assign graders to the dataset (1 to 100 ids per call). Graders outside the
+     * organization or under another trace function are reported in
+     * `skippedGraderIds` rather than failing the call.
+     */
+    addGraders(datasetId: string, graderIds: string[]): Promise<AddDatasetGradersResult>;
+    /** Unassign graders from the dataset. */
+    removeGraders(datasetId: string, graderIds: string[]): Promise<RemoveDatasetGradersResult>;
+    /**
+     * Re-run graders over every trace in the dataset. Defaults to every assigned
+     * grader; an unassigned id is rejected. Waits for the run to finish (up to
+     * `timeoutMs`, default 90s) unless `wait` is `false`, and returns the last
+     * run state seen either way. A request matching an in-flight run joins it.
+     */
+    rerunGraders(datasetId: string, options?: RerunGradersOptions): Promise<RerunGradersResult>;
+    /**
+     * The dataset's active grader re-run, or the run named by `runId`. Returns
+     * `null` when nothing is active or the run does not belong to this dataset.
+     */
+    getGraderRerun(datasetId: string, runId?: string): Promise<GraderRerun | null>;
+}
+
 /**
  * LangGraph/LangChain callback handler for Bitfab tracing.
  *
@@ -2272,6 +2411,8 @@ declare class Bitfab {
     private readonly explicitlyEnabled;
     private readonly strict;
     private readonly httpClient;
+    /** Dataset operations for the authenticated organization. */
+    readonly datasets: DatasetsClient;
     private readonly bamlClient;
     private readonly dbSnapshot;
     private readonly autoTracePolicyRefreshes;
@@ -2955,7 +3096,7 @@ declare class BitfabFunction {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.39.0";
+declare const __version__ = "0.40.0";
 
 /**
  * Constants for the Bitfab SDK.
@@ -3083,4 +3224,4 @@ interface SeedResult {
  */
 declare function seedFromRegistry(registry: ReplayRegistry, pipeline: string, cases: readonly SeedCase[]): Promise<SeedResult>;
 
-export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AllowedEnvVars, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, type CaptureWhen, type CapturedSpan, type CodeChangeFile, type CurrentSpan, type CurrentTrace, DEFAULT_SERVICE_URL, type DbBranchOptions, DbBranchReplayError, type DbBranchTimings, type DbSnapshotConfig, type DbSnapshotProvider, type DbSnapshotRef, type DetachedTrace, HttpClient, type LangGraphIntegrationOptions, type MockOverride, type MockOverrideCtx, type MockOverrideInput, type MockOverrideResolver, type MockStrategy, type MockValue, NO_MOCK_OVERRIDE, type NodeMatcher, type NodeMethodDecorator, type NodeOptions, type ProviderDefinition, ReplayBranch, ReplayError, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, SUPPORTED_PROVIDERS, type SeedCase, type SeedResult, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceIngestionType, type TraceResponse, type TracingProcessor, type VercelCallParams, type VercelGenerateResult, type VercelStreamResult, type WrapBAMLOptions, type WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, seedFromRegistry, serializeReplayResult };
+export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AddDatasetGradersResult, type AddDatasetTracesResult, type AllowedEnvVars, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, type CaptureWhen, type CapturedSpan, type CodeChangeFile, type CurrentSpan, type CurrentTrace, DEFAULT_SERVICE_URL, type Dataset, type DatasetGraderRef, type DatasetTraceIds, DatasetsClient, type DbBranchOptions, DbBranchReplayError, type DbBranchTimings, type DbSnapshotConfig, type DbSnapshotProvider, type DbSnapshotRef, type DetachedTrace, type GraderRerun, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, HttpClient, type LangGraphIntegrationOptions, type ListDatasetsParams, type MockOverride, type MockOverrideCtx, type MockOverrideInput, type MockOverrideResolver, type MockStrategy, type MockValue, NO_MOCK_OVERRIDE, type NodeMatcher, type NodeMethodDecorator, type NodeOptions, type ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, SUPPORTED_PROVIDERS, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedResult, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceIngestionType, type TraceResponse, type TracingProcessor, type VercelCallParams, type VercelGenerateResult, type VercelStreamResult, type WrapBAMLOptions, type WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, seedFromRegistry, serializeReplayResult };

package/dist/index.d.ts

@@ -540,7 +540,11 @@ declare class HttpClient {
     lookupFunction<T>(name: string): Promise<T>;
     getAutoTracePolicy<T>(traceFunctionKey: string, protocol: string): Promise<T>;
     getTraceSpan(traceId: string, lookup: SpanLookup): Promise<CapturedSpan | null>;
-    private get;
+    /**
+     * GET a JSON endpoint on the service with the client's API key. Throws a
+     * `BitfabError` carrying the status text for any non-2xx response.
+     */
+    get<T>(endpoint: string): Promise<T>;
     /**
      * Queue an internal trace (from local BAML execution via `call()`) onto this
      * client's batching transport. `functionId` moves into the payload because
@@ -1034,6 +1038,141 @@ declare class BitfabVercelAiHandler {
     get middleware(): BitfabLanguageModelMiddleware;
 }
 
+interface DatasetGraderRef {
+    id: string;
+    name: string | null;
+}
+interface Dataset {
+    id: string;
+    traceFunctionKey: string;
+    name: string;
+    description: string | null;
+    traceCount: number;
+    graders: DatasetGraderRef[];
+    createdAt: string;
+    updatedAt: string;
+}
+interface SaveDatasetParams {
+    traceFunctionKey: string;
+    name: string;
+    description?: string;
+}
+interface SaveDatasetResult {
+    dataset: Dataset;
+    created: boolean;
+}
+interface ListDatasetsParams {
+    traceFunctionKey?: string;
+}
+interface DatasetTraceIds {
+    datasetId: string;
+    traceIds: string[];
+}
+interface AddDatasetTracesResult {
+    dataset: Dataset;
+    addedTraceIds: string[];
+    alreadyPresentTraceIds: string[];
+    skippedTraceIds: string[];
+}
+interface RemoveDatasetTracesResult {
+    dataset: Dataset;
+    removedTraceIds: string[];
+    notPresentTraceIds: string[];
+}
+interface AddDatasetGradersResult {
+    dataset: Dataset;
+    addedGraderIds: string[];
+    alreadyAssignedGraderIds: string[];
+    skippedGraderIds: string[];
+}
+interface RemoveDatasetGradersResult {
+    dataset: Dataset;
+    removedGraderIds: string[];
+    notAssignedGraderIds: string[];
+}
+type GraderRerunStatus = "pending" | "running" | "completed" | "errored";
+interface GraderRerunProgress {
+    completedTraces: number;
+    totalTraces: number;
+    graderCount: number;
+}
+interface GraderRerunResult {
+    tracesGraded: number;
+    gradersRun: number;
+}
+interface GraderRerun {
+    id: string;
+    status: GraderRerunStatus;
+    graderIds: string[];
+    progress: GraderRerunProgress | null;
+    result: GraderRerunResult | null;
+    error: string | null;
+    createdAt: string;
+    updatedAt: string;
+}
+interface RerunGradersOptions {
+    graderIds?: string[];
+    wait?: boolean;
+    timeoutMs?: number;
+    pollIntervalMs?: number;
+}
+interface RerunGradersResult {
+    run: GraderRerun;
+    joinedExisting: boolean;
+}
+/**
+ * Dataset operations for the authenticated organization, reached as
+ * `client.datasets`. A dataset is a named bucket of traces scoped to one trace
+ * function. Experiments replay against it and its graders score its members.
+ */
+declare class DatasetsClient {
+    private readonly httpClient;
+    constructor(httpClient: HttpClient);
+    /**
+     * Create a dataset, or update the one already named this way under the same
+     * trace function. `created` reports which happened. An omitted description
+     * leaves an existing one untouched.
+     */
+    save(params: SaveDatasetParams): Promise<SaveDatasetResult>;
+    /**
+     * List datasets, scoped to one trace function when `traceFunctionKey` is
+     * given and organization-wide otherwise.
+     */
+    list(params?: ListDatasetsParams): Promise<Dataset[]>;
+    /** Fetch one dataset by id. Rejects with a 404 `BitfabError` when it is not in this organization. */
+    get(datasetId: string): Promise<Dataset>;
+    /** The ids of every trace in the dataset, the same membership a replay with `datasetId` selects. */
+    listTraces(datasetId: string): Promise<DatasetTraceIds>;
+    /**
+     * Add traces to the dataset (1 to 100 ids per call). Traces outside the
+     * organization or under another trace function are reported in
+     * `skippedTraceIds` rather than failing the call.
+     */
+    addTraces(datasetId: string, traceIds: string[]): Promise<AddDatasetTracesResult>;
+    /** Remove traces from the dataset. The traces themselves are never deleted. */
+    removeTraces(datasetId: string, traceIds: string[]): Promise<RemoveDatasetTracesResult>;
+    /**
+     * Assign graders to the dataset (1 to 100 ids per call). Graders outside the
+     * organization or under another trace function are reported in
+     * `skippedGraderIds` rather than failing the call.
+     */
+    addGraders(datasetId: string, graderIds: string[]): Promise<AddDatasetGradersResult>;
+    /** Unassign graders from the dataset. */
+    removeGraders(datasetId: string, graderIds: string[]): Promise<RemoveDatasetGradersResult>;
+    /**
+     * Re-run graders over every trace in the dataset. Defaults to every assigned
+     * grader; an unassigned id is rejected. Waits for the run to finish (up to
+     * `timeoutMs`, default 90s) unless `wait` is `false`, and returns the last
+     * run state seen either way. A request matching an in-flight run joins it.
+     */
+    rerunGraders(datasetId: string, options?: RerunGradersOptions): Promise<RerunGradersResult>;
+    /**
+     * The dataset's active grader re-run, or the run named by `runId`. Returns
+     * `null` when nothing is active or the run does not belong to this dataset.
+     */
+    getGraderRerun(datasetId: string, runId?: string): Promise<GraderRerun | null>;
+}
+
 /**
  * LangGraph/LangChain callback handler for Bitfab tracing.
  *
@@ -2272,6 +2411,8 @@ declare class Bitfab {
     private readonly explicitlyEnabled;
     private readonly strict;
     private readonly httpClient;
+    /** Dataset operations for the authenticated organization. */
+    readonly datasets: DatasetsClient;
     private readonly bamlClient;
     private readonly dbSnapshot;
     private readonly autoTracePolicyRefreshes;
@@ -2955,7 +3096,7 @@ declare class BitfabFunction {
 /**
  * SDK version from package.json (injected at build time)
  */
-declare const __version__ = "0.39.0";
+declare const __version__ = "0.40.0";
 
 /**
  * Constants for the Bitfab SDK.
@@ -3083,4 +3224,4 @@ interface SeedResult {
  */
 declare function seedFromRegistry(registry: ReplayRegistry, pipeline: string, cases: readonly SeedCase[]): Promise<SeedResult>;
 
-export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AllowedEnvVars, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, type CaptureWhen, type CapturedSpan, type CodeChangeFile, type CurrentSpan, type CurrentTrace, DEFAULT_SERVICE_URL, type DbBranchOptions, DbBranchReplayError, type DbBranchTimings, type DbSnapshotConfig, type DbSnapshotProvider, type DbSnapshotRef, type DetachedTrace, HttpClient, type LangGraphIntegrationOptions, type MockOverride, type MockOverrideCtx, type MockOverrideInput, type MockOverrideResolver, type MockStrategy, type MockValue, NO_MOCK_OVERRIDE, type NodeMatcher, type NodeMethodDecorator, type NodeOptions, type ProviderDefinition, ReplayBranch, ReplayError, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, SUPPORTED_PROVIDERS, type SeedCase, type SeedResult, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceIngestionType, type TraceResponse, type TracingProcessor, type VercelCallParams, type VercelGenerateResult, type VercelStreamResult, type WrapBAMLOptions, type WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, seedFromRegistry, serializeReplayResult };
+export { type ActiveSpanContext, type AdaptContext, type AdaptInputsFn, type AddDatasetGradersResult, type AddDatasetTracesResult, type AllowedEnvVars, BITFAB_PROGRESS_PREFIX, type BamlExecutionResult, Bitfab, BitfabClaudeAgentHandler, type BitfabConfig, BitfabError, BitfabFunction, BitfabLangGraphCallbackHandler as BitfabLangChainCallbackHandler, BitfabLangGraphCallbackHandler, BitfabLangGraphIntegration, type BitfabLanguageModelMiddleware, BitfabOpenAIAgentHandler, BitfabOpenAITracingProcessor, BitfabVercelAiHandler, type CaptureWhen, type CapturedSpan, type CodeChangeFile, type CurrentSpan, type CurrentTrace, DEFAULT_SERVICE_URL, type Dataset, type DatasetGraderRef, type DatasetTraceIds, DatasetsClient, type DbBranchOptions, DbBranchReplayError, type DbBranchTimings, type DbSnapshotConfig, type DbSnapshotProvider, type DbSnapshotRef, type DetachedTrace, type GraderRerun, type GraderRerunProgress, type GraderRerunResult, type GraderRerunStatus, HttpClient, type LangGraphIntegrationOptions, type ListDatasetsParams, type MockOverride, type MockOverrideCtx, type MockOverrideInput, type MockOverrideResolver, type MockStrategy, type MockValue, NO_MOCK_OVERRIDE, type NodeMatcher, type NodeMethodDecorator, type NodeOptions, type ProviderDefinition, type RemoveDatasetGradersResult, type RemoveDatasetTracesResult, ReplayBranch, ReplayError, type ReplayItem, type ReplayItemFinishProgress, type ReplayItemStartProgress, type ReplayOptions, type ReplayOptionsFactory, type ReplayProgress, type ReplayProgressItem, type ReplayRegistration, type ReplayRegistry, type ReplayRegistryContext, type ReplayRegistryOptions, type ReplayResult, type RerunGradersOptions, type RerunGradersResult, SUPPORTED_PROVIDERS, type SaveDatasetParams, type SaveDatasetResult, type SeedCase, type SeedResult, type SpanLookup, type SpanMethodDecorator, type SpanMethodDecoratorContext, type SpanNodeMeta, type SpanOccurrence, type SpanOptions, type SpanType, type TokenUsage, type TraceIngestionType, type TraceResponse, type TracingProcessor, type VercelCallParams, type VercelGenerateResult, type VercelStreamResult, type WrapBAMLOptions, type WrappedBamlFn, __version__, defineReplayRegistry, finalizers, flushTraces, getCurrentReplayBranch, getCurrentSpan, getCurrentTrace, reportReplayProgress, seedFromRegistry, serializeReplayResult };

package/dist/index.js

@@ -16,6 +16,7 @@ import {
   BitfabOpenAIAgentHandler,
   BitfabOpenAITracingProcessor,
   BitfabVercelAiHandler,
+  DatasetsClient,
   SUPPORTED_PROVIDERS,
   defineReplayRegistry,
   finalizers,
@@ -23,7 +24,7 @@ import {
   getCurrentSpan,
   getCurrentTrace,
   seedFromRegistry
-} from "./chunk-B4EJ2EQT.js";
+} from "./chunk-HBPJSNC2.js";
 import "./chunk-ZUD7OFYB.js";
 import {
   BITFAB_PROGRESS_PREFIX,
@@ -32,14 +33,14 @@ import {
   ReplayError,
   reportReplayProgress,
   serializeReplayResult
-} from "./chunk-7OKE7FGZ.js";
+} from "./chunk-SMOV6KN5.js";
 import {
   BitfabError,
   DEFAULT_SERVICE_URL,
   HttpClient,
   __version__,
   flushTraces
-} from "./chunk-OXE6M7CZ.js";
+} from "./chunk-DYURJJSV.js";
 import "./chunk-H6LZRFMN.js";
 export {
   BITFAB_PROGRESS_PREFIX,
@@ -54,6 +55,7 @@ export {
   BitfabOpenAITracingProcessor,
   BitfabVercelAiHandler,
   DEFAULT_SERVICE_URL,
+  DatasetsClient,
   DbBranchReplayError,
   HttpClient,
   NO_MOCK_OVERRIDE,

package/dist/node.cjs

@@ -97,7 +97,7 @@ var __version__, __packageName__;
 var init_version_generated = __esm({
   "src/version.generated.ts"() {
     "use strict";
-    __version__ = "0.39.0";
+    __version__ = "0.40.0";
     __packageName__ = "bitfab";
   }
 });
@@ -1748,6 +1748,10 @@ var init_http = __esm({
         const response = await this.get(endpoint);
         return response.span;
       }
+      /**
+       * GET a JSON endpoint on the service with the client's API key. Throws a
+       * `BitfabError` carrying the status text for any non-2xx response.
+       */
       async get(endpoint) {
         const url = `${this.serviceUrl}${endpoint}`;
         const controller = new AbortController();
@@ -1761,7 +1765,10 @@ var init_http = __esm({
           if (!response.ok) {
             const errorText = await response.text();
             throw new BitfabError(
-              `HTTP ${response.status}: ${errorText.slice(0, 500)}`
+              `HTTP ${response.status}: ${errorText.slice(0, 500)}`,
+              void 0,
+              response.status,
+              parseRetryAfterMs(readHeader(response, "retry-after"))
             );
           }
           return await response.json();
@@ -3213,6 +3220,7 @@ __export(node_exports, {
   BitfabOpenAITracingProcessor: () => BitfabOpenAITracingProcessor,
   BitfabVercelAiHandler: () => BitfabVercelAiHandler,
   DEFAULT_SERVICE_URL: () => DEFAULT_SERVICE_URL,
+  DatasetsClient: () => DatasetsClient,
   DbBranchReplayError: () => DbBranchReplayError,
   HttpClient: () => HttpClient,
   NO_MOCK_OVERRIDE: () => NO_MOCK_OVERRIDE,
@@ -4299,6 +4307,131 @@ async function runFunctionWithBaml(bamlSource, inputs, providers, envVars) {
 // src/client.ts
 init_constants();
 
+// src/datasets.ts
+var DEFAULT_RERUN_TIMEOUT_MS = 9e4;
+var DEFAULT_RERUN_POLL_INTERVAL_MS = 1e3;
+var TERMINAL_RERUN_STATUSES = /* @__PURE__ */ new Set([
+  "completed",
+  "errored"
+]);
+function sleep(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+function datasetPath(datasetId, suffix = "") {
+  return `/api/sdk/datasets/${encodeURIComponent(datasetId)}${suffix}`;
+}
+var DatasetsClient = class {
+  constructor(httpClient) {
+    this.httpClient = httpClient;
+  }
+  /**
+   * Create a dataset, or update the one already named this way under the same
+   * trace function. `created` reports which happened. An omitted description
+   * leaves an existing one untouched.
+   */
+  async save(params) {
+    return this.httpClient.request("/api/sdk/datasets", {
+      traceFunctionKey: params.traceFunctionKey,
+      name: params.name,
+      ...params.description === void 0 ? {} : { description: params.description }
+    });
+  }
+  /**
+   * List datasets, scoped to one trace function when `traceFunctionKey` is
+   * given and organization-wide otherwise.
+   */
+  async list(params = {}) {
+    const query = params.traceFunctionKey === void 0 ? "" : `?traceFunctionKey=${encodeURIComponent(params.traceFunctionKey)}`;
+    const response = await this.httpClient.get(
+      `/api/sdk/datasets${query}`
+    );
+    return response.datasets;
+  }
+  /** Fetch one dataset by id. Rejects with a 404 `BitfabError` when it is not in this organization. */
+  async get(datasetId) {
+    const response = await this.httpClient.get(
+      datasetPath(datasetId)
+    );
+    return response.dataset;
+  }
+  /** The ids of every trace in the dataset, the same membership a replay with `datasetId` selects. */
+  async listTraces(datasetId) {
+    return this.httpClient.get(
+      datasetPath(datasetId, "/traces")
+    );
+  }
+  /**
+   * Add traces to the dataset (1 to 100 ids per call). Traces outside the
+   * organization or under another trace function are reported in
+   * `skippedTraceIds` rather than failing the call.
+   */
+  async addTraces(datasetId, traceIds) {
+    return this.httpClient.request(
+      datasetPath(datasetId, "/traces"),
+      { traceIds }
+    );
+  }
+  /** Remove traces from the dataset. The traces themselves are never deleted. */
+  async removeTraces(datasetId, traceIds) {
+    return this.httpClient.request(
+      datasetPath(datasetId, "/removeTraces"),
+      { traceIds }
+    );
+  }
+  /**
+   * Assign graders to the dataset (1 to 100 ids per call). Graders outside the
+   * organization or under another trace function are reported in
+   * `skippedGraderIds` rather than failing the call.
+   */
+  async addGraders(datasetId, graderIds) {
+    return this.httpClient.request(
+      datasetPath(datasetId, "/graders"),
+      { graderIds }
+    );
+  }
+  /** Unassign graders from the dataset. */
+  async removeGraders(datasetId, graderIds) {
+    return this.httpClient.request(
+      datasetPath(datasetId, "/removeGraders"),
+      { graderIds }
+    );
+  }
+  /**
+   * Re-run graders over every trace in the dataset. Defaults to every assigned
+   * grader; an unassigned id is rejected. Waits for the run to finish (up to
+   * `timeoutMs`, default 90s) unless `wait` is `false`, and returns the last
+   * run state seen either way. A request matching an in-flight run joins it.
+   */
+  async rerunGraders(datasetId, options = {}) {
+    const started = await this.httpClient.request(
+      datasetPath(datasetId, "/rerunGraders"),
+      options.graderIds === void 0 ? {} : { graderIds: options.graderIds }
+    );
+    if (options.wait === false) {
+      return started;
+    }
+    const deadline = Date.now() + (options.timeoutMs ?? DEFAULT_RERUN_TIMEOUT_MS);
+    const interval = options.pollIntervalMs ?? DEFAULT_RERUN_POLL_INTERVAL_MS;
+    let run = started.run;
+    while (!TERMINAL_RERUN_STATUSES.has(run.status) && Date.now() < deadline) {
+      await sleep(interval);
+      run = await this.getGraderRerun(datasetId, run.id) ?? run;
+    }
+    return { run, joinedExisting: started.joinedExisting };
+  }
+  /**
+   * The dataset's active grader re-run, or the run named by `runId`. Returns
+   * `null` when nothing is active or the run does not belong to this dataset.
+   */
+  async getGraderRerun(datasetId, runId) {
+    const query = runId === void 0 ? "" : `?runId=${encodeURIComponent(runId)}`;
+    const response = await this.httpClient.get(
+      datasetPath(datasetId, `/rerunGraders${query}`)
+    );
+    return response.run;
+  }
+};
+
 // src/dbSnapshot.ts
 init_errors();
 var SUPPORTED_PROVIDERS = ["neon"];
@@ -6034,6 +6167,7 @@ var Bitfab = class {
       serviceUrl: this.serviceUrl,
       timeout: this.timeout
     });
+    this.datasets = new DatasetsClient(this.httpClient);
   }
   /**
    * Decorate a class method as an automatically expanded trace root.
@@ -7851,6 +7985,7 @@ assertAsyncStorageRegistered();
   BitfabOpenAITracingProcessor,
   BitfabVercelAiHandler,
   DEFAULT_SERVICE_URL,
+  DatasetsClient,
   DbBranchReplayError,
   HttpClient,
   NO_MOCK_OVERRIDE,