braintrust 3.2.0 → 3.3.0

package/dist/index.mjs

@@ -5921,6 +5921,7 @@ function utf8ByteLength(value) {
 function now() {
   return (/* @__PURE__ */ new Date()).getTime();
 }
+var DEFAULT_FLUSH_BACKPRESSURE_BYTES = 10 * 1024 * 1024;
 var TestBackgroundLogger = class {
   items = [];
   maskingFunction = null;
@@ -5933,6 +5934,12 @@ var TestBackgroundLogger = class {
   async flush() {
     return Promise.resolve();
   }
+  pendingFlushBytes() {
+    return 0;
+  }
+  flushBackpressureBytes() {
+    return DEFAULT_FLUSH_BACKPRESSURE_BYTES;
+  }
   async drain() {
     const items = this.items;
     this.items = [];
@@ -5990,7 +5997,8 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
   queueDropLoggingPeriod = 60;
   failedPublishPayloadsDir = void 0;
   allPublishPayloadsDir = void 0;
-  flushChunkSize = 25;
+  _flushBackpressureBytes = DEFAULT_FLUSH_BACKPRESSURE_BYTES;
+  _pendingBytes = 0;
   _disabled = false;
   queueDropLoggingState = {
     numDropped: 0,
@@ -6030,11 +6038,16 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (!isNaN(queueDropLoggingPeriodEnv)) {
       this.queueDropLoggingPeriod = queueDropLoggingPeriodEnv;
     }
-    const flushChunkSizeEnv = Number(
-      isomorph_default.getEnv("BRAINTRUST_LOG_FLUSH_CHUNK_SIZE")
+    if (isomorph_default.getEnv("BRAINTRUST_LOG_FLUSH_CHUNK_SIZE")) {
+      console.warn(
+        "BRAINTRUST_LOG_FLUSH_CHUNK_SIZE is deprecated and no longer has any effect. Log flushing now sends all items at once and batches them automatically. This environment variable will be removed in a future major release."
+      );
+    }
+    const flushBackpressureBytesEnv = Number(
+      isomorph_default.getEnv("BRAINTRUST_FLUSH_BACKPRESSURE_BYTES")
     );
-    if (!isNaN(flushChunkSizeEnv) && flushChunkSizeEnv > 0) {
-      this.flushChunkSize = flushChunkSizeEnv;
+    if (!isNaN(flushBackpressureBytesEnv) && flushBackpressureBytesEnv > 0) {
+      this._flushBackpressureBytes = flushBackpressureBytesEnv;
     }
     const failedPublishPayloadsDirEnv = isomorph_default.getEnv(
       "BRAINTRUST_FAILED_PUBLISH_PAYLOADS_DIR"
@@ -6058,6 +6071,12 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
   setMaskingFunction(maskingFunction) {
     this.maskingFunction = maskingFunction;
   }
+  pendingFlushBytes() {
+    return this._pendingBytes;
+  }
+  flushBackpressureBytes() {
+    return this._flushBackpressureBytes;
+  }
   log(items) {
     if (this._disabled) {
       return;
@@ -6120,14 +6139,7 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (wrappedItems.length === 0) {
       return;
     }
-    const chunkSize = Math.max(1, Math.min(batchSize, this.flushChunkSize));
-    let index = 0;
-    while (index < wrappedItems.length) {
-      const chunk = wrappedItems.slice(index, index + chunkSize);
-      await this.flushWrappedItemsChunk(chunk, batchSize);
-      index += chunk.length;
-    }
-    wrappedItems.length = 0;
+    await this.flushWrappedItemsChunk(wrappedItems, batchSize);
     if (this.queue.length() > 0) {
       await this.flushOnce(args);
     }
@@ -6140,9 +6152,13 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (allItems.length === 0) {
       return;
     }
-    const allItemsWithMeta = allItems.map(
-      (item) => stringifyWithOverflowMeta(item)
-    );
+    let chunkBytes = 0;
+    const allItemsWithMeta = allItems.map((item) => {
+      const withMeta = stringifyWithOverflowMeta(item);
+      chunkBytes += withMeta.str.length;
+      return withMeta;
+    });
+    this._pendingBytes += chunkBytes;
     const maxRequestSizeResult = await this.getMaxRequestSize();
     const batches = batchItems({
       items: allItemsWithMeta,
@@ -6161,6 +6177,7 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
       })()
     );
     const results = await Promise.all(postPromises);
+    this._pendingBytes = Math.max(0, this._pendingBytes - chunkBytes);
     const failingResultErrors = results.map((r) => r.type === "success" ? void 0 : r.value).filter((r) => r !== void 0);
     if (failingResultErrors.length) {
       throw new AggregateError(
@@ -12047,6 +12064,7 @@ __export(exports_exports, {
   promptDefinitionToPromptData: () => promptDefinitionToPromptData,
   promptDefinitionWithToolsSchema: () => promptDefinitionWithToolsSchema,
   registerOtelFlush: () => registerOtelFlush,
+  registerSandbox: () => registerSandbox,
   registerTemplatePlugin: () => registerTemplatePlugin,
   renderMessage: () => renderMessage,
   renderPromptParams: () => renderPromptParams,
@@ -12083,6 +12101,99 @@ __export(exports_exports, {
   wrapVitest: () => wrapVitest
 });
 
+// src/sandbox.ts
+import { z as z9 } from "zod/v3";
+var SANDBOX_GROUP_NAME_METADATA_KEY = "_bt_sandbox_group_name";
+async function registerSandbox(options) {
+  const state = options.state ?? _internalGetGlobalState();
+  await state.login({
+    apiKey: options.apiKey,
+    appUrl: options.appUrl,
+    orgName: options.orgName
+  });
+  const projectResponse = await state.appConn().post_json("api/project/register", {
+    project_name: options.project,
+    org_id: state.orgId
+  });
+  const projectId = projectResponse.project.id;
+  if (!state.orgName) {
+    throw new Error("Organization name is required to register sandbox evals");
+  }
+  const runtimeContext = {
+    runtime: "node",
+    version: process.version.slice(1)
+  };
+  const listResponse = await state.proxyConn().post(
+    "function/sandbox-list",
+    {
+      sandbox_spec: {
+        provider: options.sandbox.provider,
+        snapshot_ref: options.sandbox.snapshotRef
+      },
+      entrypoints: options.entrypoints,
+      project_id: projectId
+    },
+    {
+      headers: {
+        "x-bt-org-name": state.orgName
+      }
+    }
+  );
+  const evaluatorDefinitions = z9.record(z9.unknown()).parse(await listResponse.json());
+  const functions = [];
+  for (const [evalName, evaluatorDefinition] of Object.entries(
+    evaluatorDefinitions
+  )) {
+    const functionName = evalName;
+    const functionSlug = slugify(evalName, { lower: true, strict: true });
+    const functionDef = {
+      project_id: projectId,
+      org_name: state.orgName,
+      name: functionName,
+      slug: functionSlug,
+      function_type: "sandbox",
+      function_data: {
+        type: "code",
+        data: {
+          type: "bundle",
+          runtime_context: runtimeContext,
+          location: {
+            type: "sandbox",
+            sandbox_spec: {
+              provider: options.sandbox.provider,
+              snapshot_ref: options.sandbox.snapshotRef
+            },
+            entrypoints: options.entrypoints,
+            eval_name: evalName,
+            evaluator_definition: evaluatorDefinition
+          },
+          bundle_id: null,
+          preview: null
+        }
+      },
+      metadata: {
+        ...options.metadata ?? {},
+        [SANDBOX_GROUP_NAME_METADATA_KEY]: options.name
+      },
+      if_exists: options.ifExists ?? "replace"
+    };
+    if (options.description !== void 0) {
+      functionDef.description = options.description;
+    }
+    const response = await state.apiConn().post_json("v1/function", functionDef);
+    functions.push({
+      evalName,
+      id: response.id,
+      name: response.name,
+      slug: response.slug
+    });
+  }
+  return {
+    projectId,
+    functions
+  };
+}
+
 // src/functions/invoke.ts
 async function invoke(args) {
   const {
@@ -18555,29 +18666,29 @@ var SimpleProgressReporter = class {
 };
 
 // src/eval-parameters.ts
-import { z as z10 } from "zod/v3";
+import { z as z11 } from "zod/v3";
 import Ajv from "ajv";
 
 // src/prompt-schemas.ts
-import { z as z9 } from "zod/v3";
-var promptContentsSchema = z9.union([
-  z9.object({
-    prompt: z9.string()
+import { z as z10 } from "zod/v3";
+var promptContentsSchema = z10.union([
+  z10.object({
+    prompt: z10.string()
   }),
-  z9.object({
-    messages: z9.array(ChatCompletionMessageParam)
+  z10.object({
+    messages: z10.array(ChatCompletionMessageParam)
   })
 ]);
 var promptDefinitionSchema = promptContentsSchema.and(
-  z9.object({
-    model: z9.string(),
+  z10.object({
+    model: z10.string(),
     params: ModelParams.optional(),
-    templateFormat: z9.enum(["mustache", "nunjucks", "none"]).optional()
+    templateFormat: z10.enum(["mustache", "nunjucks", "none"]).optional()
   })
 );
 var promptDefinitionWithToolsSchema = promptDefinitionSchema.and(
-  z9.object({
-    tools: z9.array(ToolFunctionDefinition).optional()
+  z10.object({
+    tools: z10.array(ToolFunctionDefinition).optional()
   })
 );
 function promptDefinitionToPromptData(promptDefinition, rawTools) {
@@ -18600,15 +18711,15 @@ function promptDefinitionToPromptData(promptDefinition, rawTools) {
 }
 
 // src/eval-parameters.ts
-var evalParametersSchema = z10.record(
-  z10.string(),
-  z10.union([
-    z10.object({
-      type: z10.literal("prompt"),
+var evalParametersSchema = z11.record(
+  z11.string(),
+  z11.union([
+    z11.object({
+      type: z11.literal("prompt"),
       default: promptDefinitionWithToolsSchema.optional(),
-      description: z10.string().optional()
+      description: z11.string().optional()
     }),
-    z10.instanceof(z10.ZodType)
+    z11.instanceof(z11.ZodType)
     // For Zod schemas
   ])
 );
@@ -19227,7 +19338,8 @@ async function runEvaluatorInternal(experiment, evaluator, progressReporter, fil
           });
         } else {
           const result = await experiment.traced(callback, baseEvent);
-          if (evaluator.maxConcurrency !== void 0) {
+          const bgLogger = experiment.loggingState.bgLogger();
+          if (evaluator.maxConcurrency !== void 0 && bgLogger.pendingFlushBytes() >= bgLogger.flushBackpressureBytes()) {
             await experiment.flush();
           }
           return result;
@@ -19491,7 +19603,7 @@ var defaultReporter = {
 };
 
 // src/framework2.ts
-import { z as z11 } from "zod/v3";
+import { z as z12 } from "zod/v3";
 var currentFilename = typeof __filename !== "undefined" ? __filename : "unknown";
 var ProjectBuilder = class {
   create(opts) {
@@ -19883,7 +19995,7 @@ var ProjectNameIdMap = class {
       const response = await _internalGetGlobalState().appConn().post_json("api/project/register", {
         project_name: projectName
       });
-      const result = z11.object({
+      const result = z12.object({
         project: Project
       }).parse(response);
       const projectId = result.project.id;
@@ -19897,7 +20009,7 @@ var ProjectNameIdMap = class {
       const response = await _internalGetGlobalState().appConn().post_json("api/project/get", {
         id: projectId
       });
-      const result = z11.array(Project).nonempty().parse(response);
+      const result = z12.array(Project).nonempty().parse(response);
       const projectName = result[0].name;
       this.idToName[projectId] = projectName;
       this.nameToId[projectName] = projectId;
@@ -19913,73 +20025,73 @@ var ProjectNameIdMap = class {
 };
 
 // dev/types.ts
-import { z as z12 } from "zod/v3";
-var evalBodySchema = z12.object({
-  name: z12.string(),
-  parameters: z12.record(z12.string(), z12.unknown()).nullish(),
+import { z as z13 } from "zod/v3";
+var evalBodySchema = z13.object({
+  name: z13.string(),
+  parameters: z13.record(z13.string(), z13.unknown()).nullish(),
   data: RunEval.shape.data,
-  scores: z12.array(
-    z12.object({
+  scores: z13.array(
+    z13.object({
       function_id: FunctionId,
-      name: z12.string()
+      name: z13.string()
     })
   ).nullish(),
-  experiment_name: z12.string().nullish(),
-  project_id: z12.string().nullish(),
+  experiment_name: z13.string().nullish(),
+  project_id: z13.string().nullish(),
   parent: InvokeParent.optional(),
-  stream: z12.boolean().optional()
+  stream: z13.boolean().optional()
 });
-var staticParametersSchema = z12.record(
-  z12.string(),
-  z12.union([
-    z12.object({
-      type: z12.literal("prompt"),
+var staticParametersSchema = z13.record(
+  z13.string(),
+  z13.union([
+    z13.object({
+      type: z13.literal("prompt"),
       default: PromptData.optional(),
-      description: z12.string().optional()
+      description: z13.string().optional()
     }),
-    z12.object({
-      type: z12.literal("data"),
-      schema: z12.record(z12.unknown()),
-      default: z12.unknown().optional(),
-      description: z12.string().optional()
+    z13.object({
+      type: z13.literal("data"),
+      schema: z13.record(z13.unknown()),
+      default: z13.unknown().optional(),
+      description: z13.string().optional()
     })
   ])
 );
-var parametersSchema = z12.object({
-  type: z12.literal("object"),
-  properties: z12.record(z12.string(), z12.record(z12.unknown())),
-  required: z12.array(z12.string()).optional(),
-  additionalProperties: z12.boolean().optional()
+var parametersSchema = z13.object({
+  type: z13.literal("object"),
+  properties: z13.record(z13.string(), z13.record(z13.unknown())),
+  required: z13.array(z13.string()).optional(),
+  additionalProperties: z13.boolean().optional()
 });
-var parametersSourceSchema = z12.object({
-  parametersId: z12.string().optional(),
-  slug: z12.string(),
-  name: z12.string(),
-  projectId: z12.string().optional(),
-  version: z12.string().optional()
+var parametersSourceSchema = z13.object({
+  parametersId: z13.string().optional(),
+  slug: z13.string(),
+  name: z13.string(),
+  projectId: z13.string().optional(),
+  version: z13.string().optional()
 });
-var parametersContainerSchema = z12.object({
-  type: z12.literal("braintrust.parameters"),
+var parametersContainerSchema = z13.object({
+  type: z13.literal("braintrust.parameters"),
   schema: parametersSchema,
   source: parametersSourceSchema
 });
-var staticParametersContainerSchema = z12.object({
-  type: z12.literal("braintrust.staticParameters"),
+var staticParametersContainerSchema = z13.object({
+  type: z13.literal("braintrust.staticParameters"),
   schema: staticParametersSchema,
-  source: z12.null()
+  source: z13.null()
 });
-var serializedParametersContainerSchema = z12.union([
+var serializedParametersContainerSchema = z13.union([
   parametersContainerSchema,
   staticParametersContainerSchema,
   // keeping this type here since old versions of the SDK will still pass the unwrapped schema and we need to handle this in the app
   staticParametersSchema
 ]);
-var evaluatorDefinitionSchema = z12.object({
+var evaluatorDefinitionSchema = z13.object({
   parameters: serializedParametersContainerSchema.optional(),
-  scores: z12.array(z12.object({ name: z12.string() })).optional()
+  scores: z13.array(z13.object({ name: z13.string() })).optional()
 });
-var evaluatorDefinitionsSchema = z12.record(
-  z12.string(),
+var evaluatorDefinitionsSchema = z13.record(
+  z13.string(),
   evaluatorDefinitionSchema
 );
 
@@ -20083,6 +20195,7 @@ export {
   promptDefinitionToPromptData,
   promptDefinitionWithToolsSchema,
   registerOtelFlush,
+  registerSandbox,
   registerTemplatePlugin,
   renderMessage,
   renderPromptParams,

package/dist/instrumentation/index.js

@@ -5499,6 +5499,7 @@ function utf8ByteLength(value) {
 function now() {
   return (/* @__PURE__ */ new Date()).getTime();
 }
+var DEFAULT_FLUSH_BACKPRESSURE_BYTES = 10 * 1024 * 1024;
 var BACKGROUND_LOGGER_BASE_SLEEP_TIME_S = 1;
 var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
   apiConn;
@@ -5517,7 +5518,8 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
   queueDropLoggingPeriod = 60;
   failedPublishPayloadsDir = void 0;
   allPublishPayloadsDir = void 0;
-  flushChunkSize = 25;
+  _flushBackpressureBytes = DEFAULT_FLUSH_BACKPRESSURE_BYTES;
+  _pendingBytes = 0;
   _disabled = false;
   queueDropLoggingState = {
     numDropped: 0,
@@ -5557,11 +5559,16 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (!isNaN(queueDropLoggingPeriodEnv)) {
       this.queueDropLoggingPeriod = queueDropLoggingPeriodEnv;
     }
-    const flushChunkSizeEnv = Number(
-      isomorph_default.getEnv("BRAINTRUST_LOG_FLUSH_CHUNK_SIZE")
+    if (isomorph_default.getEnv("BRAINTRUST_LOG_FLUSH_CHUNK_SIZE")) {
+      console.warn(
+        "BRAINTRUST_LOG_FLUSH_CHUNK_SIZE is deprecated and no longer has any effect. Log flushing now sends all items at once and batches them automatically. This environment variable will be removed in a future major release."
+      );
+    }
+    const flushBackpressureBytesEnv = Number(
+      isomorph_default.getEnv("BRAINTRUST_FLUSH_BACKPRESSURE_BYTES")
     );
-    if (!isNaN(flushChunkSizeEnv) && flushChunkSizeEnv > 0) {
-      this.flushChunkSize = flushChunkSizeEnv;
+    if (!isNaN(flushBackpressureBytesEnv) && flushBackpressureBytesEnv > 0) {
+      this._flushBackpressureBytes = flushBackpressureBytesEnv;
     }
     const failedPublishPayloadsDirEnv = isomorph_default.getEnv(
       "BRAINTRUST_FAILED_PUBLISH_PAYLOADS_DIR"
@@ -5585,6 +5592,12 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
   setMaskingFunction(maskingFunction) {
     this.maskingFunction = maskingFunction;
   }
+  pendingFlushBytes() {
+    return this._pendingBytes;
+  }
+  flushBackpressureBytes() {
+    return this._flushBackpressureBytes;
+  }
   log(items) {
     if (this._disabled) {
       return;
@@ -5647,14 +5660,7 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (wrappedItems.length === 0) {
       return;
     }
-    const chunkSize = Math.max(1, Math.min(batchSize, this.flushChunkSize));
-    let index = 0;
-    while (index < wrappedItems.length) {
-      const chunk = wrappedItems.slice(index, index + chunkSize);
-      await this.flushWrappedItemsChunk(chunk, batchSize);
-      index += chunk.length;
-    }
-    wrappedItems.length = 0;
+    await this.flushWrappedItemsChunk(wrappedItems, batchSize);
     if (this.queue.length() > 0) {
       await this.flushOnce(args);
     }
@@ -5667,9 +5673,13 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (allItems.length === 0) {
       return;
     }
-    const allItemsWithMeta = allItems.map(
-      (item) => stringifyWithOverflowMeta(item)
-    );
+    let chunkBytes = 0;
+    const allItemsWithMeta = allItems.map((item) => {
+      const withMeta = stringifyWithOverflowMeta(item);
+      chunkBytes += withMeta.str.length;
+      return withMeta;
+    });
+    this._pendingBytes += chunkBytes;
     const maxRequestSizeResult = await this.getMaxRequestSize();
     const batches = batchItems({
       items: allItemsWithMeta,
@@ -5688,6 +5698,7 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
       })()
     );
     const results = await Promise.all(postPromises);
+    this._pendingBytes = Math.max(0, this._pendingBytes - chunkBytes);
     const failingResultErrors = results.map((r) => r.type === "success" ? void 0 : r.value).filter((r) => r !== void 0);
     if (failingResultErrors.length) {
       throw new AggregateError(

package/dist/instrumentation/index.mjs

@@ -5460,6 +5460,7 @@ function utf8ByteLength(value) {
 function now() {
   return (/* @__PURE__ */ new Date()).getTime();
 }
+var DEFAULT_FLUSH_BACKPRESSURE_BYTES = 10 * 1024 * 1024;
 var BACKGROUND_LOGGER_BASE_SLEEP_TIME_S = 1;
 var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
   apiConn;
@@ -5478,7 +5479,8 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
   queueDropLoggingPeriod = 60;
   failedPublishPayloadsDir = void 0;
   allPublishPayloadsDir = void 0;
-  flushChunkSize = 25;
+  _flushBackpressureBytes = DEFAULT_FLUSH_BACKPRESSURE_BYTES;
+  _pendingBytes = 0;
   _disabled = false;
   queueDropLoggingState = {
     numDropped: 0,
@@ -5518,11 +5520,16 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (!isNaN(queueDropLoggingPeriodEnv)) {
       this.queueDropLoggingPeriod = queueDropLoggingPeriodEnv;
     }
-    const flushChunkSizeEnv = Number(
-      isomorph_default.getEnv("BRAINTRUST_LOG_FLUSH_CHUNK_SIZE")
+    if (isomorph_default.getEnv("BRAINTRUST_LOG_FLUSH_CHUNK_SIZE")) {
+      console.warn(
+        "BRAINTRUST_LOG_FLUSH_CHUNK_SIZE is deprecated and no longer has any effect. Log flushing now sends all items at once and batches them automatically. This environment variable will be removed in a future major release."
+      );
+    }
+    const flushBackpressureBytesEnv = Number(
+      isomorph_default.getEnv("BRAINTRUST_FLUSH_BACKPRESSURE_BYTES")
     );
-    if (!isNaN(flushChunkSizeEnv) && flushChunkSizeEnv > 0) {
-      this.flushChunkSize = flushChunkSizeEnv;
+    if (!isNaN(flushBackpressureBytesEnv) && flushBackpressureBytesEnv > 0) {
+      this._flushBackpressureBytes = flushBackpressureBytesEnv;
     }
     const failedPublishPayloadsDirEnv = isomorph_default.getEnv(
       "BRAINTRUST_FAILED_PUBLISH_PAYLOADS_DIR"
@@ -5546,6 +5553,12 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
   setMaskingFunction(maskingFunction) {
     this.maskingFunction = maskingFunction;
   }
+  pendingFlushBytes() {
+    return this._pendingBytes;
+  }
+  flushBackpressureBytes() {
+    return this._flushBackpressureBytes;
+  }
   log(items) {
     if (this._disabled) {
       return;
@@ -5608,14 +5621,7 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (wrappedItems.length === 0) {
       return;
     }
-    const chunkSize = Math.max(1, Math.min(batchSize, this.flushChunkSize));
-    let index = 0;
-    while (index < wrappedItems.length) {
-      const chunk = wrappedItems.slice(index, index + chunkSize);
-      await this.flushWrappedItemsChunk(chunk, batchSize);
-      index += chunk.length;
-    }
-    wrappedItems.length = 0;
+    await this.flushWrappedItemsChunk(wrappedItems, batchSize);
     if (this.queue.length() > 0) {
       await this.flushOnce(args);
     }
@@ -5628,9 +5634,13 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
     if (allItems.length === 0) {
       return;
     }
-    const allItemsWithMeta = allItems.map(
-      (item) => stringifyWithOverflowMeta(item)
-    );
+    let chunkBytes = 0;
+    const allItemsWithMeta = allItems.map((item) => {
+      const withMeta = stringifyWithOverflowMeta(item);
+      chunkBytes += withMeta.str.length;
+      return withMeta;
+    });
+    this._pendingBytes += chunkBytes;
     const maxRequestSizeResult = await this.getMaxRequestSize();
     const batches = batchItems({
       items: allItemsWithMeta,
@@ -5649,6 +5659,7 @@ var HTTPBackgroundLogger = class _HTTPBackgroundLogger {
       })()
     );
     const results = await Promise.all(postPromises);
+    this._pendingBytes = Math.max(0, this._pendingBytes - chunkBytes);
     const failingResultErrors = results.map((r) => r.type === "success" ? void 0 : r.value).filter((r) => r !== void 0);
     if (failingResultErrors.length) {
       throw new AggregateError(

package/dist/workerd.d.mts

@@ -1,2 +1,2 @@
-export { AnyDataset, Attachment, AttachmentParams, AttachmentReference, BackgroundLoggerOpts, BaseAttachment, BaseExperiment, BaseMetadata, BraintrustMiddleware, BraintrustState, BraintrustStream, BraintrustStreamChunk, CachedSpanFetcher, ChatPrompt, CodeFunction, CodeOpts, CodePrompt, CommentEvent, CompiledPrompt, CompiledPromptParams, CompletionPrompt, ContextManager, ContextParentSpanIds, CreateProjectOpts, DEFAULT_FETCH_BATCH_SIZE, DEFAULT_MAX_REQUEST_SIZE, DataSummary, Dataset, DatasetRecord, DatasetSummary, DefaultMetadataType, DefaultPromptArgs, ERR_PERMALINK, EndSpanArgs, Eval, EvalCase, EvalHooks, EvalParameterSerializedSchema, EvalParameters, EvalResult, EvalResultWithSummary, EvalScorer, EvalScorerArgs, EvalTask, Evaluator, EvaluatorDef, EvaluatorDefinition, EvaluatorDefinitions, EvaluatorFile, EvaluatorManifest, Experiment, ExperimentLogFullArgs, ExperimentLogPartialArgs, ExperimentSummary, Exportable, ExternalAttachment, ExternalAttachmentParams, FailedHTTPResponse, FullInitDatasetOptions, FullInitOptions, FullLoginOptions, FunctionEvent, GetThreadOptions, IDGenerator, IdField, InitDatasetOptions, InitLoggerOptions, InitOptions, InputField, InstrumentationConfig, InvokeFunctionArgs, InvokeReturn, JSONAttachment, LEGACY_CACHED_HEADER, LOGS3_OVERFLOW_REFERENCE_TYPE, LazyValue, LoadPromptOptions, LogCommentFullArgs, LogFeedbackFullArgs, LogOptions, Logger, LoginInvalidOrgError, LoginOptions, Logs3OverflowInputRow, Logs3OverflowUpload, MetricSummary, NOOP_SPAN, NOOP_SPAN_PERMALINK, NoopSpan, ObjectFetcher, ObjectMetadata, OtherExperimentLogFields, ParametersSource, ParentExperimentIds, ParentProjectLogIds, Project, ProjectNameIdMap, PromiseUnless, Prompt, PromptBuilder, PromptContents, PromptDefinition, PromptDefinitionWithTools, PromptOpts, PromptRowWithId, ReadonlyAttachment, ReadonlyExperiment, Reporter, ReporterBody, ScoreSummary, ScorerBuilder, ScorerOpts, SerializedBraintrustState, SetCurrentArg, Span, SpanContext, SpanData, SpanFetcher, SpanImpl, StartSpanArgs, TemplateFormat, TemplateRenderer, TemplateRendererPlugin, TestBackgroundLogger, ToolBuilder, Trace, UUIDGenerator, WithTransactionId, X_CACHED_HEADER, _exportsForTestingOnly, _internalGetGlobalState, _internalIso, _internalSetInitialState, addAzureBlobHeaders, braintrustStreamChunkSchema, buildLocalSummary, configureInstrumentation, constructLogs3OverflowRequest, createFinalValuePassThroughStream, currentExperiment, currentLogger, currentSpan, deepCopyEvent, default, defaultErrorScoreHandler, deserializePlainStringAsJSON, devNullWritableStream, evaluatorDefinitionSchema, evaluatorDefinitionsSchema, flush, getContextManager, getIdGenerator, getPromptVersions, getSpanParentObject, getTemplateRenderer, graph, init, initDataset, initExperiment, initFunction, initLogger, invoke, isTemplateFormat, loadParameters, loadPrompt, log, logError, login, loginToState, logs3OverflowUploadSchema, newId, parseCachedHeader, parseTemplateFormat, permalink, pickLogs3OverflowObjectIds, projects, promptContentsSchema, promptDefinitionSchema, promptDefinitionToPromptData, promptDefinitionWithToolsSchema, registerOtelFlush, registerTemplatePlugin, renderMessage, renderPromptParams, renderTemplateContent, reportFailures, runEvaluator, setFetch, setMaskingFunction, spanComponentsToObjectId, startSpan, summarize, templateRegistry, toolFunctionDefinitionSchema, traceable, traced, updateSpan, uploadLogs3OverflowPayload, utf8ByteLength, withCurrent, withDataset, withExperiment, withLogger, withParent, wrapAISDK, wrapAISDKModel, wrapAgentClass, wrapAnthropic, wrapClaudeAgentSDK, wrapGoogleGenAI, wrapMastraAgent, wrapOpenAI, wrapOpenAIv4, wrapTraced, wrapVitest } from './browser.mjs';
+export { AnyDataset, Attachment, AttachmentParams, AttachmentReference, BackgroundLoggerOpts, BaseAttachment, BaseExperiment, BaseMetadata, BraintrustMiddleware, BraintrustState, BraintrustStream, BraintrustStreamChunk, CachedSpanFetcher, ChatPrompt, CodeFunction, CodeOpts, CodePrompt, CommentEvent, CompiledPrompt, CompiledPromptParams, CompletionPrompt, ContextManager, ContextParentSpanIds, CreateProjectOpts, DEFAULT_FETCH_BATCH_SIZE, DEFAULT_MAX_REQUEST_SIZE, DataSummary, Dataset, DatasetRecord, DatasetSummary, DefaultMetadataType, DefaultPromptArgs, ERR_PERMALINK, EndSpanArgs, Eval, EvalCase, EvalHooks, EvalParameterSerializedSchema, EvalParameters, EvalResult, EvalResultWithSummary, EvalScorer, EvalScorerArgs, EvalTask, Evaluator, EvaluatorDef, EvaluatorDefinition, EvaluatorDefinitions, EvaluatorFile, EvaluatorManifest, Experiment, ExperimentLogFullArgs, ExperimentLogPartialArgs, ExperimentSummary, Exportable, ExternalAttachment, ExternalAttachmentParams, FailedHTTPResponse, FullInitDatasetOptions, FullInitOptions, FullLoginOptions, FunctionEvent, GetThreadOptions, IDGenerator, IdField, InitDatasetOptions, InitLoggerOptions, InitOptions, InputField, InstrumentationConfig, InvokeFunctionArgs, InvokeReturn, JSONAttachment, LEGACY_CACHED_HEADER, LOGS3_OVERFLOW_REFERENCE_TYPE, LazyValue, LoadPromptOptions, LogCommentFullArgs, LogFeedbackFullArgs, LogOptions, Logger, LoginInvalidOrgError, LoginOptions, Logs3OverflowInputRow, Logs3OverflowUpload, MetricSummary, NOOP_SPAN, NOOP_SPAN_PERMALINK, NoopSpan, ObjectFetcher, ObjectMetadata, OtherExperimentLogFields, ParametersSource, ParentExperimentIds, ParentProjectLogIds, Project, ProjectNameIdMap, PromiseUnless, Prompt, PromptBuilder, PromptContents, PromptDefinition, PromptDefinitionWithTools, PromptOpts, PromptRowWithId, ReadonlyAttachment, ReadonlyExperiment, RegisterSandboxOptions, RegisterSandboxResult, Reporter, ReporterBody, SandboxConfig, ScoreSummary, ScorerBuilder, ScorerOpts, SerializedBraintrustState, SetCurrentArg, Span, SpanContext, SpanData, SpanFetcher, SpanImpl, StartSpanArgs, TemplateFormat, TemplateRenderer, TemplateRendererPlugin, TestBackgroundLogger, ToolBuilder, Trace, UUIDGenerator, WithTransactionId, X_CACHED_HEADER, _exportsForTestingOnly, _internalGetGlobalState, _internalIso, _internalSetInitialState, addAzureBlobHeaders, braintrustStreamChunkSchema, buildLocalSummary, configureInstrumentation, constructLogs3OverflowRequest, createFinalValuePassThroughStream, currentExperiment, currentLogger, currentSpan, deepCopyEvent, default, defaultErrorScoreHandler, deserializePlainStringAsJSON, devNullWritableStream, evaluatorDefinitionSchema, evaluatorDefinitionsSchema, flush, getContextManager, getIdGenerator, getPromptVersions, getSpanParentObject, getTemplateRenderer, graph, init, initDataset, initExperiment, initFunction, initLogger, invoke, isTemplateFormat, loadParameters, loadPrompt, log, logError, login, loginToState, logs3OverflowUploadSchema, newId, parseCachedHeader, parseTemplateFormat, permalink, pickLogs3OverflowObjectIds, projects, promptContentsSchema, promptDefinitionSchema, promptDefinitionToPromptData, promptDefinitionWithToolsSchema, registerOtelFlush, registerSandbox, registerTemplatePlugin, renderMessage, renderPromptParams, renderTemplateContent, reportFailures, runEvaluator, setFetch, setMaskingFunction, spanComponentsToObjectId, startSpan, summarize, templateRegistry, toolFunctionDefinitionSchema, traceable, traced, updateSpan, uploadLogs3OverflowPayload, utf8ByteLength, withCurrent, withDataset, withExperiment, withLogger, withParent, wrapAISDK, wrapAISDKModel, wrapAgentClass, wrapAnthropic, wrapClaudeAgentSDK, wrapGoogleGenAI, wrapMastraAgent, wrapOpenAI, wrapOpenAIv4, wrapTraced, wrapVitest } from './browser.mjs';
 import 'zod/v3';

package/dist/workerd.d.ts

@@ -1,2 +1,2 @@
-export { AnyDataset, Attachment, AttachmentParams, AttachmentReference, BackgroundLoggerOpts, BaseAttachment, BaseExperiment, BaseMetadata, BraintrustMiddleware, BraintrustState, BraintrustStream, BraintrustStreamChunk, CachedSpanFetcher, ChatPrompt, CodeFunction, CodeOpts, CodePrompt, CommentEvent, CompiledPrompt, CompiledPromptParams, CompletionPrompt, ContextManager, ContextParentSpanIds, CreateProjectOpts, DEFAULT_FETCH_BATCH_SIZE, DEFAULT_MAX_REQUEST_SIZE, DataSummary, Dataset, DatasetRecord, DatasetSummary, DefaultMetadataType, DefaultPromptArgs, ERR_PERMALINK, EndSpanArgs, Eval, EvalCase, EvalHooks, EvalParameterSerializedSchema, EvalParameters, EvalResult, EvalResultWithSummary, EvalScorer, EvalScorerArgs, EvalTask, Evaluator, EvaluatorDef, EvaluatorDefinition, EvaluatorDefinitions, EvaluatorFile, EvaluatorManifest, Experiment, ExperimentLogFullArgs, ExperimentLogPartialArgs, ExperimentSummary, Exportable, ExternalAttachment, ExternalAttachmentParams, FailedHTTPResponse, FullInitDatasetOptions, FullInitOptions, FullLoginOptions, FunctionEvent, GetThreadOptions, IDGenerator, IdField, InitDatasetOptions, InitLoggerOptions, InitOptions, InputField, InstrumentationConfig, InvokeFunctionArgs, InvokeReturn, JSONAttachment, LEGACY_CACHED_HEADER, LOGS3_OVERFLOW_REFERENCE_TYPE, LazyValue, LoadPromptOptions, LogCommentFullArgs, LogFeedbackFullArgs, LogOptions, Logger, LoginInvalidOrgError, LoginOptions, Logs3OverflowInputRow, Logs3OverflowUpload, MetricSummary, NOOP_SPAN, NOOP_SPAN_PERMALINK, NoopSpan, ObjectFetcher, ObjectMetadata, OtherExperimentLogFields, ParametersSource, ParentExperimentIds, ParentProjectLogIds, Project, ProjectNameIdMap, PromiseUnless, Prompt, PromptBuilder, PromptContents, PromptDefinition, PromptDefinitionWithTools, PromptOpts, PromptRowWithId, ReadonlyAttachment, ReadonlyExperiment, Reporter, ReporterBody, ScoreSummary, ScorerBuilder, ScorerOpts, SerializedBraintrustState, SetCurrentArg, Span, SpanContext, SpanData, SpanFetcher, SpanImpl, StartSpanArgs, TemplateFormat, TemplateRenderer, TemplateRendererPlugin, TestBackgroundLogger, ToolBuilder, Trace, UUIDGenerator, WithTransactionId, X_CACHED_HEADER, _exportsForTestingOnly, _internalGetGlobalState, _internalIso, _internalSetInitialState, addAzureBlobHeaders, braintrustStreamChunkSchema, buildLocalSummary, configureInstrumentation, constructLogs3OverflowRequest, createFinalValuePassThroughStream, currentExperiment, currentLogger, currentSpan, deepCopyEvent, default, defaultErrorScoreHandler, deserializePlainStringAsJSON, devNullWritableStream, evaluatorDefinitionSchema, evaluatorDefinitionsSchema, flush, getContextManager, getIdGenerator, getPromptVersions, getSpanParentObject, getTemplateRenderer, graph, init, initDataset, initExperiment, initFunction, initLogger, invoke, isTemplateFormat, loadParameters, loadPrompt, log, logError, login, loginToState, logs3OverflowUploadSchema, newId, parseCachedHeader, parseTemplateFormat, permalink, pickLogs3OverflowObjectIds, projects, promptContentsSchema, promptDefinitionSchema, promptDefinitionToPromptData, promptDefinitionWithToolsSchema, registerOtelFlush, registerTemplatePlugin, renderMessage, renderPromptParams, renderTemplateContent, reportFailures, runEvaluator, setFetch, setMaskingFunction, spanComponentsToObjectId, startSpan, summarize, templateRegistry, toolFunctionDefinitionSchema, traceable, traced, updateSpan, uploadLogs3OverflowPayload, utf8ByteLength, withCurrent, withDataset, withExperiment, withLogger, withParent, wrapAISDK, wrapAISDKModel, wrapAgentClass, wrapAnthropic, wrapClaudeAgentSDK, wrapGoogleGenAI, wrapMastraAgent, wrapOpenAI, wrapOpenAIv4, wrapTraced, wrapVitest } from './browser.js';
+export { AnyDataset, Attachment, AttachmentParams, AttachmentReference, BackgroundLoggerOpts, BaseAttachment, BaseExperiment, BaseMetadata, BraintrustMiddleware, BraintrustState, BraintrustStream, BraintrustStreamChunk, CachedSpanFetcher, ChatPrompt, CodeFunction, CodeOpts, CodePrompt, CommentEvent, CompiledPrompt, CompiledPromptParams, CompletionPrompt, ContextManager, ContextParentSpanIds, CreateProjectOpts, DEFAULT_FETCH_BATCH_SIZE, DEFAULT_MAX_REQUEST_SIZE, DataSummary, Dataset, DatasetRecord, DatasetSummary, DefaultMetadataType, DefaultPromptArgs, ERR_PERMALINK, EndSpanArgs, Eval, EvalCase, EvalHooks, EvalParameterSerializedSchema, EvalParameters, EvalResult, EvalResultWithSummary, EvalScorer, EvalScorerArgs, EvalTask, Evaluator, EvaluatorDef, EvaluatorDefinition, EvaluatorDefinitions, EvaluatorFile, EvaluatorManifest, Experiment, ExperimentLogFullArgs, ExperimentLogPartialArgs, ExperimentSummary, Exportable, ExternalAttachment, ExternalAttachmentParams, FailedHTTPResponse, FullInitDatasetOptions, FullInitOptions, FullLoginOptions, FunctionEvent, GetThreadOptions, IDGenerator, IdField, InitDatasetOptions, InitLoggerOptions, InitOptions, InputField, InstrumentationConfig, InvokeFunctionArgs, InvokeReturn, JSONAttachment, LEGACY_CACHED_HEADER, LOGS3_OVERFLOW_REFERENCE_TYPE, LazyValue, LoadPromptOptions, LogCommentFullArgs, LogFeedbackFullArgs, LogOptions, Logger, LoginInvalidOrgError, LoginOptions, Logs3OverflowInputRow, Logs3OverflowUpload, MetricSummary, NOOP_SPAN, NOOP_SPAN_PERMALINK, NoopSpan, ObjectFetcher, ObjectMetadata, OtherExperimentLogFields, ParametersSource, ParentExperimentIds, ParentProjectLogIds, Project, ProjectNameIdMap, PromiseUnless, Prompt, PromptBuilder, PromptContents, PromptDefinition, PromptDefinitionWithTools, PromptOpts, PromptRowWithId, ReadonlyAttachment, ReadonlyExperiment, RegisterSandboxOptions, RegisterSandboxResult, Reporter, ReporterBody, SandboxConfig, ScoreSummary, ScorerBuilder, ScorerOpts, SerializedBraintrustState, SetCurrentArg, Span, SpanContext, SpanData, SpanFetcher, SpanImpl, StartSpanArgs, TemplateFormat, TemplateRenderer, TemplateRendererPlugin, TestBackgroundLogger, ToolBuilder, Trace, UUIDGenerator, WithTransactionId, X_CACHED_HEADER, _exportsForTestingOnly, _internalGetGlobalState, _internalIso, _internalSetInitialState, addAzureBlobHeaders, braintrustStreamChunkSchema, buildLocalSummary, configureInstrumentation, constructLogs3OverflowRequest, createFinalValuePassThroughStream, currentExperiment, currentLogger, currentSpan, deepCopyEvent, default, defaultErrorScoreHandler, deserializePlainStringAsJSON, devNullWritableStream, evaluatorDefinitionSchema, evaluatorDefinitionsSchema, flush, getContextManager, getIdGenerator, getPromptVersions, getSpanParentObject, getTemplateRenderer, graph, init, initDataset, initExperiment, initFunction, initLogger, invoke, isTemplateFormat, loadParameters, loadPrompt, log, logError, login, loginToState, logs3OverflowUploadSchema, newId, parseCachedHeader, parseTemplateFormat, permalink, pickLogs3OverflowObjectIds, projects, promptContentsSchema, promptDefinitionSchema, promptDefinitionToPromptData, promptDefinitionWithToolsSchema, registerOtelFlush, registerSandbox, registerTemplatePlugin, renderMessage, renderPromptParams, renderTemplateContent, reportFailures, runEvaluator, setFetch, setMaskingFunction, spanComponentsToObjectId, startSpan, summarize, templateRegistry, toolFunctionDefinitionSchema, traceable, traced, updateSpan, uploadLogs3OverflowPayload, utf8ByteLength, withCurrent, withDataset, withExperiment, withLogger, withParent, wrapAISDK, wrapAISDKModel, wrapAgentClass, wrapAnthropic, wrapClaudeAgentSDK, wrapGoogleGenAI, wrapMastraAgent, wrapOpenAI, wrapOpenAIv4, wrapTraced, wrapVitest } from './browser.js';
 import 'zod/v3';