@truefoundry/ai-gateway-sdk 0.0.0-stage → 0.0.2

package/dist/services/transformToProviderRequest.js

@@ -0,0 +1,268 @@
+import { ValidationError } from '../errors.js';
+import { AWS_BEDROCK_MANTLE, AZURE_OPEN_AI, BEDROCK, DEEPINFRA, GOOGLE_GEMINI, GOOGLE_VERTEX, isClaudeNativeStructuredOutputModel, MICROSOFT_FOUNDRY, MISTRAL_AI, OPEN_AI, OPENAI_MODELS_WITH_NATIVE_RESPONSE_FORMAT, OPENAI_MODELS_WITHOUT_NATIVE_RESPONSE_FORMAT, OPENROUTER, PERPLEXITY_AI, TOGETHER_AI, } from '../globals.js';
+import { getProviderConfig } from '../handlers/tfyUtils.js';
+import { logger } from '../logger.js';
+import ProviderConfigs from '../providers/index.js';
+import { isAnthropicModel as isMantleAnthropicModel } from '../providers/aws-bedrock-mantle/modelFamily.js';
+import { buildAnthropicOutputConfig, ResponseFormatSupport, universalResponseFormatRequestTransform, } from '../providers/shared/responseFormatUtils.js';
+import { getCacheInjectionMode, injectCacheControl, relocateBodyCacheControl, shouldStripCacheControl, stripCacheControl, } from '../providers/utils/cacheControlUtils.js';
+import { cloneWithSharedStrings } from '../utils/memoryOptimizationUtils.js';
+import { applyDefaultMaxTokensIfMissing, convertMidConversationSystemMessages, modelHasFeature, normalizeMaxTokensParam, normalizeReasoningEffortParam, normalizeSystemRoleMessages, } from '../utils/modelParamUtils.js';
+import { getModelSpecificParamsToExclude, getValue, resolveEffectiveModelName, transformUsingProviderConfig, } from '../utils/providerConfigTransform.js';
+export { transformUsingProviderConfig } from '../utils/providerConfigTransform.js';
+function toGetConfigArgs(params, providerOptions, callConfig, headerOptions, execution) {
+    return {
+        params: params,
+        providerOptions,
+        callConfig,
+        headerOptions,
+        execution,
+    };
+}
+function openAiSupportsNativeResponseFormat(model) {
+    if (OPENAI_MODELS_WITHOUT_NATIVE_RESPONSE_FORMAT.some(prefix => model.startsWith(prefix))) {
+        return false;
+    }
+    return OPENAI_MODELS_WITH_NATIVE_RESPONSE_FORMAT.some(prefix => model.startsWith(prefix));
+}
+function isAnthropicModelWithNativeStructuredOutput(model) {
+    return isClaudeNativeStructuredOutputModel(model);
+}
+function rejectsAnthropicOutputConfig(provider, model) {
+    if (provider === GOOGLE_VERTEX)
+        return model.startsWith('anthropic/');
+    if (provider === AWS_BEDROCK_MANTLE)
+        return isMantleAnthropicModel(model);
+    return false;
+}
+/**
+ * Converts response_format (OpenAI shape) to output_config for providers that
+ * use it natively (Anthropic direct API, Bedrock Converse API).
+ */
+function nativeOutputConfigHandler(requestBody) {
+    const outputConfig = buildAnthropicOutputConfig(requestBody.response_format);
+    const { response_format: _, ...rest } = requestBody;
+    return (outputConfig ? { ...rest, output_config: outputConfig } : rest);
+}
+/**
+ * Determines the structured-output support level and optional native handler
+ * for a given provider + model + response_format type combination.
+ *
+ * Resolution order:
+ *  1. Provider-specific hardcoded rules (OpenAI, Azure, Mistral, OpenRouter, Gemini, Vertex non-Anthropic)
+ *  2. Anthropic native output_config (direct API + Bedrock; Anthropic-format hosts that reject
+ *     output_config are excluded — see rejectsAnthropicOutputConfig)
+ *  3. Model JSON fallback (modelHasFeature 'structured_output'; same hosts excluded)
+ *  4. Default: NONE → tool conversion
+ */
+export function resolveResponseFormatSupport(provider, model, responseFormatType, modelEntity, execution) {
+    const NONE = { supportLevel: ResponseFormatSupport.NONE };
+    const FULL = { supportLevel: ResponseFormatSupport.FULL };
+    const PARTIAL = { supportLevel: ResponseFormatSupport.PARTIAL };
+    // 1. OpenAI / Azure — model-dependent
+    if ([OPEN_AI, AZURE_OPEN_AI, MICROSOFT_FOUNDRY].includes(provider)) {
+        const resolvedModel = provider === AZURE_OPEN_AI || provider === MICROSOFT_FOUNDRY
+            ? modelEntity?.manifest?.foundation_model || model
+            : model;
+        if (responseFormatType === 'json_object')
+            return FULL;
+        if (responseFormatType === 'json_schema') {
+            return openAiSupportsNativeResponseFormat(resolvedModel) ? FULL : NONE;
+        }
+        return NONE;
+    }
+    // 2. OpenAI-compatible providers — full passthrough
+    if (provider === MISTRAL_AI || provider === OPENROUTER || provider === TOGETHER_AI || provider === PERPLEXITY_AI)
+        return FULL;
+    // 3. DeepInfra — OpenAI-compatible; json_object is natively supported.
+    if (provider === DEEPINFRA && responseFormatType === 'json_object')
+        return FULL;
+    // 4. Gemini — PARTIAL (native when no tool conflicts, tool conversion otherwise)
+    if (provider === GOOGLE_GEMINI)
+        return PARTIAL;
+    // 5. Vertex — PARTIAL for non-Anthropic models (Gemini on Vertex)
+    if (provider === GOOGLE_VERTEX && !model.startsWith('anthropic/'))
+        return PARTIAL;
+    const hostRejectsOutputConfig = rejectsAnthropicOutputConfig(provider, model);
+    // 6. Anthropic native output_config (direct API + other hosts that accept it).
+    //    Bedrock Runtime is handled by rule 7 (Converse outputConfig).
+    if (!hostRejectsOutputConfig &&
+        provider !== BEDROCK &&
+        responseFormatType === 'json_schema' &&
+        isAnthropicModelWithNativeStructuredOutput(model)) {
+        return { supportLevel: ResponseFormatSupport.FULL, nativeHandler: nativeOutputConfigHandler };
+    }
+    // 7. Bedrock — convert response_format to Converse API's outputConfig.
+    //    The Converse API supports outputConfig.textFormat for structured output across all
+    //    model providers. Anthropic models with json_schema already returned in rule 6.
+    //    For json_object (no schema), the handler strips response_format cleanly since the
+    //    Converse API has no json_object equivalent.
+    //    CC→Responses must keep response_format so it becomes Responses text.format.
+    if (provider === BEDROCK && modelHasFeature(provider, model, 'structured_output', false)) {
+        if (execution?.openaiUseResponsesApi)
+            return FULL;
+        return { supportLevel: ResponseFormatSupport.FULL, nativeHandler: nativeOutputConfigHandler };
+    }
+    // 8. Model JSON fallback — trust the model config for providers with native passthrough.
+    //    Skip the hosts from rule 6: they strip response_format but don't support output_config,
+    //    so FULL without a nativeHandler would silently lose structured output.
+    if (!hostRejectsOutputConfig && modelHasFeature(provider, model, 'structured_output', false)) {
+        return FULL;
+    }
+    return NONE;
+}
+/**
+ * Transforms the request body to match the structure required by the AI provider.
+ * It also ensures the values for each parameter are within the minimum and maximum
+ * constraints defined in the provider's configuration. If a required parameter is missing,
+ * it assigns the default value from the provider's configuration.
+ */
+const transformToProviderRequestJSON = (providerOptions, params, endpoint, callConfig, headerOptions, execution, transformInput) => {
+    // Get the configuration for the specified provider
+    let providerConfig = getProviderConfig(providerOptions.providerAccount, providerOptions.model ?? undefined);
+    if (providerConfig.getConfig) {
+        providerConfig = providerConfig.getConfig(toGetConfigArgs(params, providerOptions, callConfig, headerOptions, execution))[endpoint];
+    }
+    else {
+        providerConfig = providerConfig[endpoint];
+    }
+    if (!providerConfig) {
+        throw new ValidationError(`${endpoint} is not supported by ${providerOptions.providerAccount.provider}`);
+    }
+    const cachedFromBody = relocateBodyCacheControl(params);
+    const promptCache = headerOptions.common?.promptCache;
+    if (!cachedFromBody && promptCache) {
+        let mode = getCacheInjectionMode(providerOptions.providerAccount.provider);
+        // Top-level cache_control is only read on the native messages shape; on
+        // chatComplete it would leak to the provider root, so use a block breakpoint.
+        if (mode === 'automatic' && endpoint !== 'messages') {
+            mode = 'explicit';
+        }
+        injectCacheControl(params, mode, promptCache);
+    }
+    if (shouldStripCacheControl(providerConfig, headerOptions, providerOptions, 
+    // Live resolveEffectiveModelName uses providerOptions.foundationModel for Bedrock ARNs.
+    headerOptions.aws?.awsBedrockModel || resolveEffectiveModelName(params, providerOptions))) {
+        stripCacheControl(params);
+    }
+    return transformUsingProviderConfig(providerConfig, params, transformInput);
+};
+const transformToProviderRequestFormData = (provider, params, endpoint, providerOptions, callConfig, headerOptions, execution, transformInput) => {
+    let providerConfig = ProviderConfigs[provider];
+    if (providerConfig.getConfig) {
+        providerConfig = providerConfig.getConfig(toGetConfigArgs(params, providerOptions, callConfig, headerOptions, execution))[endpoint];
+    }
+    else {
+        providerConfig = providerConfig[endpoint];
+    }
+    const formData = new FormData();
+    // check if there are model-specific parameters to exclude
+    let paramsToExclude = [];
+    const effectiveModelName = resolveEffectiveModelName(params, providerOptions);
+    if (effectiveModelName) {
+        paramsToExclude = getModelSpecificParamsToExclude(effectiveModelName);
+    }
+    for (const configParam in providerConfig) {
+        // skip parameters that are in the exclusion list for this model
+        if (paramsToExclude.includes(configParam)) {
+            continue;
+        }
+        let paramConfigs = providerConfig[configParam];
+        if (!Array.isArray(paramConfigs)) {
+            paramConfigs = [paramConfigs];
+        }
+        for (const paramConfig of paramConfigs) {
+            if (configParam in params) {
+                const value = getValue(configParam, params, paramConfig, transformInput);
+                const shouldRemove = typeof paramConfig.removeFromParams === 'function'
+                    ? paramConfig.removeFromParams(params)
+                    : paramConfig.removeFromParams;
+                if (shouldRemove) {
+                    continue;
+                }
+                formData.append(paramConfig.param, value);
+            }
+            else if (paramConfig && paramConfig.required && paramConfig.default !== undefined) {
+                formData.append(paramConfig.param, paramConfig.default);
+            }
+        }
+    }
+    return formData;
+};
+const transformToProviderRequestReadableStream = (provider, requestBody, requestHeaders, endpoint, providerOptions, callConfig, headerOptions, execution) => {
+    const providerConfig = ProviderConfigs[provider];
+    if (providerConfig.getConfig) {
+        return providerConfig
+            .getConfig(toGetConfigArgs(requestBody, providerOptions, callConfig, headerOptions, execution))
+            .requestTransforms[endpoint](requestBody, requestHeaders);
+    }
+    else {
+        return providerConfig.requestTransforms[endpoint](requestBody, requestHeaders);
+    }
+};
+/**
+ * Transforms the request parameters to the format expected by the provider.
+ *
+ * Retargeted off legacy flattened `Options` onto explicit SDK inputs (GATEWAY-3138 / P4.3).
+ */
+export const transformToProviderRequest = async (provider, params, requestBody, endpoint, requestHeaders, providerOptions, callConfig, headerOptions, execution) => {
+    if (requestBody instanceof FormData || requestBody instanceof ArrayBuffer)
+        return requestBody;
+    if (endpoint === 'tts-proxy' || endpoint === 'stt-proxy' || endpoint === 'video-proxy') {
+        return requestBody;
+    }
+    if (endpoint === 'proxy') {
+        return params;
+    }
+    const transformInput = { callConfig, headerOptions, providerOptions };
+    if (endpoint === 'uploadFile' && requestBody instanceof ReadableStream) {
+        return await transformToProviderRequestReadableStream(provider, requestBody, requestHeaders, endpoint, providerOptions, callConfig, headerOptions, execution);
+    }
+    const providerConfig = getProviderConfig(providerOptions.providerAccount, providerOptions.model ?? undefined);
+    const activeProviderConfig = providerConfig.getConfig
+        ? providerConfig.getConfig(toGetConfigArgs(params, providerOptions, callConfig, headerOptions, execution))
+        : providerConfig;
+    // # Memory Optimization Block
+    let transformedParams = { ...params };
+    if (params.messages) {
+        transformedParams.messages = cloneWithSharedStrings(params.messages);
+    }
+    if (params.tools) {
+        transformedParams.tools = structuredClone(params.tools);
+    }
+    if (params.response_format) {
+        transformedParams.response_format = structuredClone(params.response_format);
+    }
+    if (params.metadata && typeof params.metadata === 'object') {
+        transformedParams.metadata = structuredClone(params.metadata);
+    }
+    if (params.extra_body && typeof params.extra_body === 'object') {
+        transformedParams.extra_body = structuredClone(params.extra_body);
+    }
+    if (endpoint === 'chatComplete') {
+        normalizeMaxTokensParam(transformedParams, provider, providerOptions);
+        applyDefaultMaxTokensIfMissing(transformedParams, provider, providerOptions, headerOptions);
+        normalizeReasoningEffortParam(transformedParams, provider, providerOptions);
+        convertMidConversationSystemMessages(transformedParams, activeProviderConfig);
+    }
+    if (endpoint === 'messages' || endpoint === 'messagesCountTokens') {
+        normalizeSystemRoleMessages(transformedParams, provider, providerOptions);
+    }
+    // Apply universal response format transform for chatComplete before provider-specific transforms
+    if (endpoint === 'chatComplete' && transformedParams.response_format) {
+        const { supportLevel, nativeHandler } = resolveResponseFormatSupport(provider, transformedParams.model ?? '', transformedParams.response_format?.type ?? '', providerOptions.model, execution);
+        logger.info(`[ResponseFormat] Model: ${provider}/${transformedParams.model}, Support Level: ${supportLevel}`);
+        transformedParams = universalResponseFormatRequestTransform(transformedParams, supportLevel, nativeHandler);
+    }
+    if (activeProviderConfig.requestTransforms &&
+        typeof activeProviderConfig.requestTransforms[endpoint] === 'function') {
+        transformedParams = activeProviderConfig.requestTransforms[endpoint](transformedParams, requestHeaders);
+    }
+    const providerAPIConfig = providerConfig.api;
+    if (providerAPIConfig.transformToFormData &&
+        providerAPIConfig.transformToFormData({ gatewayRequestBody: transformedParams }))
+        return transformToProviderRequestFormData(provider, transformedParams, endpoint, providerOptions, callConfig, headerOptions, execution, transformInput);
+    return transformToProviderRequestJSON(providerOptions, transformedParams, endpoint, callConfig, headerOptions, execution, transformInput);
+};
+export default transformToProviderRequest;
+//# sourceMappingURL=transformToProviderRequest.js.map

package/dist/services/transformToProviderRequest.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"transformToProviderRequest.js","sourceRoot":"","sources":["../../src/services/transformToProviderRequest.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,eAAe,EAAE,MAAM,WAAW,CAAC;AAC5C,OAAO,EACL,kBAAkB,EAClB,aAAa,EACb,OAAO,EACP,SAAS,EACT,aAAa,EACb,aAAa,EACb,mCAAmC,EACnC,iBAAiB,EACjB,UAAU,EACV,OAAO,EACP,yCAAyC,EACzC,4CAA4C,EAC5C,UAAU,EACV,aAAa,EACb,WAAW,GACZ,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,iBAAiB,EAAE,MAAM,sBAAsB,CAAC;AACzD,OAAO,EAAE,MAAM,EAAE,MAAM,WAAW,CAAC;AACnC,OAAO,eAAe,MAAM,cAAc,CAAC;AAC3C,OAAO,EAAE,gBAAgB,IAAI,sBAAsB,EAAE,MAAM,6CAA6C,CAAC;AACzG,OAAO,EACL,0BAA0B,EAC1B,qBAAqB,EACrB,uCAAuC,GACxC,MAAM,yCAAyC,CAAC;AAEjD,OAAO,EACL,qBAAqB,EACrB,kBAAkB,EAClB,wBAAwB,EACxB,uBAAuB,EACvB,iBAAiB,GAClB,MAAM,sCAAsC,CAAC;AAI9C,OAAO,EAAE,sBAAsB,EAAE,MAAM,kCAAkC,CAAC;AAC1E,OAAO,EACL,8BAA8B,EAC9B,oCAAoC,EACpC,eAAe,EACf,uBAAuB,EACvB,6BAA6B,EAC7B,2BAA2B,GAC5B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,+BAA+B,EAC/B,QAAQ,EACR,yBAAyB,EACzB,4BAA4B,GAC7B,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EAAE,4BAA4B,EAAE,MAAM,kCAAkC,CAAC;AAOhF,SAAS,eAAe,CACtB,MAA+B,EAC/B,eAAgC,EAChC,UAA8B,EAC9B,aAAoC,EACpC,SAA8B;IAE9B,OAAO;QACL,MAAM,EAAE,MAAgB;QACxB,eAAe;QACf,UAAU;QACV,aAAa;QACb,SAAS;KACV,CAAC;AACJ,CAAC;AAED,SAAS,kCAAkC,CAAC,KAAa;IACvD,IAAI,4CAA4C,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;QAC1F,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,yCAAyC,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC;AAC5F,CAAC;AAED,SAAS,0CAA0C,CAAC,KAAa;IAC/D,OAAO,mCAAmC,CAAC,KAAK,CAAC,CAAC;AACpD,CAAC;AAED,SAAS,4BAA4B,CAAC,QAAgB,EAAE,KAAa;IACnE,IAAI,QAAQ,KAAK,aAAa;QAAE,OAAO,KAAK,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC;IACtE,IAAI,QAAQ,KAAK,kBAAkB;QAAE,OAAO,sBAAsB,CAAC,KAAK,CAAC,CAAC;IAC1E,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;;GAGG;AACH,SAAS,yBAAyB,CAAC,WAAmB;IACpD,MAAM,YAAY,GAAG,0BAA0B,CAAC,WAAW,CAAC,eAAe,CAAC,CAAC;IAC7E,MAAM,EAAE,eAAe,EAAE,CAAC,EAAE,GAAG,IAAI,EAAE,GAAG,WAAW,CAAC;IACpD,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,aAAa,EAAE,YAAY,EAAE,CAAC,CAAC,CAAC,IAAI,CAAW,CAAC;AACpF,CAAC;AAED;;;;;;;;;;GAUG;AACH,MAAM,UAAU,4BAA4B,CAC1C,QAAgB,EAChB,KAAa,EACb,kBAA0B,EAC1B,WAAsC,EACtC,SAA6D;IAE7D,MAAM,IAAI,GAA6B,EAAE,YAAY,EAAE,qBAAqB,CAAC,IAAI,EAAE,CAAC;IACpF,MAAM,IAAI,GAA6B,EAAE,YAAY,EAAE,qBAAqB,CAAC,IAAI,EAAE,CAAC;IACpF,MAAM,OAAO,GAA6B,EAAE,YAAY,EAAE,qBAAqB,CAAC,OAAO,EAAE,CAAC;IAE1F,sCAAsC;IACtC,IAAI,CAAC,OAAO,EAAE,aAAa,EAAE,iBAAiB,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,CAAC;QACnE,MAAM,aAAa,GACjB,QAAQ,KAAK,aAAa,IAAI,QAAQ,KAAK,iBAAiB;YAC1D,CAAC,CAAE,WAAW,EAAE,QAAqD,EAAE,gBAAgB,IAAI,KAAK;YAChG,CAAC,CAAC,KAAK,CAAC;QAEZ,IAAI,kBAAkB,KAAK,aAAa;YAAE,OAAO,IAAI,CAAC;QACtD,IAAI,kBAAkB,KAAK,aAAa,EAAE,CAAC;YACzC,OAAO,kCAAkC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;QACzE,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAED,oDAAoD;IACpD,IAAI,QAAQ,KAAK,UAAU,IAAI,QAAQ,KAAK,UAAU,IAAI,QAAQ,KAAK,WAAW,IAAI,QAAQ,KAAK,aAAa;QAC9G,OAAO,IAAI,CAAC;IAEd,uEAAuE;IACvE,IAAI,QAAQ,KAAK,SAAS,IAAI,kBAAkB,KAAK,aAAa;QAAE,OAAO,IAAI,CAAC;IAEhF,iFAAiF;IACjF,IAAI,QAAQ,KAAK,aAAa;QAAE,OAAO,OAAO,CAAC;IAE/C,kEAAkE;IAClE,IAAI,QAAQ,KAAK,aAAa,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,YAAY,CAAC;QAAE,OAAO,OAAO,CAAC;IAElF,MAAM,uBAAuB,GAAG,4BAA4B,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;IAE9E,+EAA+E;IAC/E,mEAAmE;IACnE,IACE,CAAC,uBAAuB;QACxB,QAAQ,KAAK,OAAO;QACpB,kBAAkB,KAAK,aAAa;QACpC,0CAA0C,CAAC,KAAK,CAAC,EACjD,CAAC;QACD,OAAO,EAAE,YAAY,EAAE,qBAAqB,CAAC,IAAI,EAAE,aAAa,EAAE,yBAAyB,EAAE,CAAC;IAChG,CAAC;IAED,uEAAuE;IACvE,wFAAwF;IACxF,oFAAoF;IACpF,uFAAuF;IACvF,iDAAiD;IACjD,iFAAiF;IACjF,IAAI,QAAQ,KAAK,OAAO,IAAI,eAAe,CAAC,QAAQ,EAAE,KAAK,EAAE,mBAAmB,EAAE,KAAK,CAAC,EAAE,CAAC;QACzF,IAAI,SAAS,EAAE,qBAAqB;YAAE,OAAO,IAAI,CAAC;QAClD,OAAO,EAAE,YAAY,EAAE,qBAAqB,CAAC,IAAI,EAAE,aAAa,EAAE,yBAAyB,EAAE,CAAC;IAChG,CAAC;IAED,yFAAyF;IACzF,6FAA6F;IAC7F,4EAA4E;IAC5E,IAAI,CAAC,uBAAuB,IAAI,eAAe,CAAC,QAAQ,EAAE,KAAK,EAAE,mBAAmB,EAAE,KAAK,CAAC,EAAE,CAAC;QAC7F,OAAO,IAAI,CAAC;IACd,CAAC;IAED,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;GAKG;AACH,MAAM,8BAA8B,GAAG,CACrC,eAAgC,EAChC,MAAc,EACd,QAAkB,EAClB,UAA8B,EAC9B,aAAoC,EACpC,SAAyC,EACzC,cAAsC,EACd,EAAE;IAC1B,mDAAmD;IACnD,IAAI,cAAc,GAAG,iBAAiB,CAAC,eAAe,CAAC,eAAe,EAAE,eAAe,CAAC,KAAK,IAAI,SAAS,CAAC,CAAC;IAC5G,IAAI,cAAc,CAAC,SAAS,EAAE,CAAC;QAC7B,cAAc,GAAG,cAAc,CAAC,SAAS,CACvC,eAAe,CAAC,MAAM,EAAE,eAAe,EAAE,UAAU,EAAE,aAAa,EAAE,SAAS,CAAC,CAC/E,CAAC,QAAQ,CAAC,CAAC;IACd,CAAC;SAAM,CAAC;QACN,cAAc,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAC;IAC5C,CAAC;IAED,IAAI,CAAC,cAAc,EAAE,CAAC;QACpB,MAAM,IAAI,eAAe,CAAC,GAAG,QAAQ,wBAAwB,eAAe,CAAC,eAAe,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC3G,CAAC;IAED,MAAM,cAAc,GAAG,wBAAwB,CAAC,MAAM,CAAC,CAAC;IACxD,MAAM,WAAW,GAAG,aAAa,CAAC,MAAM,EAAE,WAAW,CAAC;IACtD,IAAI,CAAC,cAAc,IAAI,WAAW,EAAE,CAAC;QACnC,IAAI,IAAI,GAAG,qBAAqB,CAAC,eAAe,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC;QAC3E,wEAAwE;QACxE,8EAA8E;QAC9E,IAAI,IAAI,KAAK,WAAW,IAAI,QAAQ,KAAK,UAAU,EAAE,CAAC;YACpD,IAAI,GAAG,UAAU,CAAC;QACpB,CAAC;QACD,kBAAkB,CAAC,MAAM,EAAE,IAAI,EAAE,WAAW,CAAC,CAAC;IAChD,CAAC;IAED,IACE,uBAAuB,CACrB,cAAc,EACd,aAAa,EACb,eAAe;IACf,wFAAwF;IACxF,aAAa,CAAC,GAAG,EAAE,eAAe,IAAI,yBAAyB,CAAC,MAAM,EAAE,eAAe,CAAC,CACzF,EACD,CAAC;QACD,iBAAiB,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;IAED,OAAO,4BAA4B,CAAC,cAAc,EAAE,MAAM,EAAE,cAAc,CAAC,CAAC;AAC9E,CAAC,CAAC;AAEF,MAAM,kCAAkC,GAAG,CACzC,QAAgB,EAChB,MAAc,EACd,QAAkB,EAClB,eAAgC,EAChC,UAA8B,EAC9B,aAAoC,EACpC,SAAyC,EACzC,cAAsC,EAC5B,EAAE;IACZ,IAAI,cAAc,GAAG,eAAe,CAAC,QAAQ,CAAC,CAAC;IAC/C,IAAI,cAAc,CAAC,SAAS,EAAE,CAAC;QAC7B,cAAc,GAAG,cAAc,CAAC,SAAS,CACvC,eAAe,CAAC,MAAM,EAAE,eAAe,EAAE,UAAU,EAAE,aAAa,EAAE,SAAS,CAAC,CAC/E,CAAC,QAAQ,CAAC,CAAC;IACd,CAAC;SAAM,CAAC;QACN,cAAc,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAC;IAC5C,CAAC;IACD,MAAM,QAAQ,GAAG,IAAI,QAAQ,EAAE,CAAC;IAEhC,0DAA0D;IAC1D,IAAI,eAAe,GAAa,EAAE,CAAC;IACnC,MAAM,kBAAkB,GAAG,yBAAyB,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC9E,IAAI,kBAAkB,EAAE,CAAC;QACvB,eAAe,GAAG,+BAA+B,CAAC,kBAAkB,CAAC,CAAC;IACxE,CAAC;IAED,KAAK,MAAM,WAAW,IAAI,cAAc,EAAE,CAAC;QACzC,gEAAgE;QAChE,IAAI,eAAe,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;YAC1C,SAAS;QACX,CAAC;QAED,IAAI,YAAY,GAAG,cAAc,CAAC,WAAW,CAAC,CAAC;QAC/C,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;YACjC,YAAY,GAAG,CAAC,YAAY,CAAC,CAAC;QAChC,CAAC;QACD,KAAK,MAAM,WAAW,IAAI,YAAY,EAAE,CAAC;YACvC,IAAI,WAAW,IAAI,MAAM,EAAE,CAAC;gBAC1B,MAAM,KAAK,GAAG,QAAQ,CAAC,WAAW,EAAE,MAAM,EAAE,WAAW,EAAE,cAAc,CAAC,CAAC;gBACzE,MAAM,YAAY,GAChB,OAAO,WAAW,CAAC,gBAAgB,KAAK,UAAU;oBAChD,CAAC,CAAC,WAAW,CAAC,gBAAgB,CAAC,MAAM,CAAC;oBACtC,CAAC,CAAC,WAAW,CAAC,gBAAgB,CAAC;gBACnC,IAAI,YAAY,EAAE,CAAC;oBACjB,SAAS;gBACX,CAAC;gBAED,QAAQ,CAAC,MAAM,CAAC,WAAW,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAC5C,CAAC;iBAAM,IAAI,WAAW,IAAI,WAAW,CAAC,QAAQ,IAAI,WAAW,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;gBACpF,QAAQ,CAAC,MAAM,CAAC,WAAW,CAAC,KAAK,EAAE,WAAW,CAAC,OAAO,CAAC,CAAC;YAC1D,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC,CAAC;AAEF,MAAM,wCAAwC,GAAG,CAC/C,QAAgB,EAChB,WAA2B,EAC3B,cAAsC,EACtC,QAAkB,EAClB,eAAgC,EAChC,UAA8B,EAC9B,aAAoC,EACpC,SAAyC,EACzC,EAAE;IACF,MAAM,cAAc,GAAG,eAAe,CAAC,QAAQ,CAAC,CAAC;IACjD,IAAI,cAAc,CAAC,SAAS,EAAE,CAAC;QAC7B,OAAO,cAAc;aAClB,SAAS,CAAC,eAAe,CAAC,WAAW,EAAE,eAAe,EAAE,UAAU,EAAE,aAAa,EAAE,SAAS,CAAC,CAAC;aAC9F,iBAAiB,CAAC,QAAQ,CAAC,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC;IAC9D,CAAC;SAAM,CAAC;QACN,OAAO,cAAc,CAAC,iBAAiB,CAAC,QAAQ,CAAC,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC;IACjF,CAAC;AACH,CAAC,CAAC;AAEF;;;;GAIG;AACH,MAAM,CAAC,MAAM,0BAA0B,GAAG,KAAK,EAC7C,QAAgB,EAChB,MAAc,EACd,WAA2E,EAC3E,QAAkB,EAClB,cAAsC,EACtC,eAAgC,EAChC,UAA8B,EAC9B,aAAoC,EACpC,SAA8B,EAC9B,EAAE;IACF,IAAI,WAAW,YAAY,QAAQ,IAAI,WAAW,YAAY,WAAW;QAAE,OAAO,WAAW,CAAC;IAE9F,IAAI,QAAQ,KAAK,WAAW,IAAI,QAAQ,KAAK,WAAW,IAAI,QAAQ,KAAK,aAAa,EAAE,CAAC;QACvF,OAAO,WAAW,CAAC;IACrB,CAAC;IAED,IAAI,QAAQ,KAAK,OAAO,EAAE,CAAC;QACzB,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,MAAM,cAAc,GAA2B,EAAE,UAAU,EAAE,aAAa,EAAE,eAAe,EAAE,CAAC;IAE9F,IAAI,QAAQ,KAAK,YAAY,IAAI,WAAW,YAAY,cAAc,EAAE,CAAC;QACvE,OAAO,MAAM,wCAAwC,CACnD,QAAQ,EACR,WAAW,EACX,cAAc,EACd,QAAQ,EACR,eAAe,EACf,UAAU,EACV,aAAa,EACb,SAAS,CACV,CAAC;IACJ,CAAC;IAED,MAAM,cAAc,GAAG,iBAAiB,CAAC,eAAe,CAAC,eAAe,EAAE,eAAe,CAAC,KAAK,IAAI,SAAS,CAAC,CAAC;IAC9G,MAAM,oBAAoB,GAAG,cAAc,CAAC,SAAS;QACnD,CAAC,CAAC,cAAc,CAAC,SAAS,CAAC,eAAe,CAAC,MAAM,EAAE,eAAe,EAAE,UAAU,EAAE,aAAa,EAAE,SAAS,CAAC,CAAC;QAC1G,CAAC,CAAC,cAAc,CAAC;IACnB,8BAA8B;IAC9B,IAAI,iBAAiB,GAAW,EAAE,GAAG,MAAM,EAAE,CAAC;IAC9C,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;QACpB,iBAAiB,CAAC,QAAQ,GAAG,sBAAsB,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACvE,CAAC;IACD,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;QACjB,iBAAiB,CAAC,KAAK,GAAG,eAAe,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAC1D,CAAC;IACD,IAAI,MAAM,CAAC,eAAe,EAAE,CAAC;QAC3B,iBAAiB,CAAC,eAAe,GAAG,eAAe,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC;IAC9E,CAAC;IACD,IAAI,MAAM,CAAC,QAAQ,IAAI,OAAO,MAAM,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC3D,iBAAiB,CAAC,QAAQ,GAAG,eAAe,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IAChE,CAAC;IACD,IAAK,MAAc,CAAC,UAAU,IAAI,OAAQ,MAAc,CAAC,UAAU,KAAK,QAAQ,EAAE,CAAC;QAChF,iBAAyB,CAAC,UAAU,GAAG,eAAe,CAAE,MAAc,CAAC,UAAU,CAAC,CAAC;IACtF,CAAC;IAED,IAAI,QAAQ,KAAK,cAAc,EAAE,CAAC;QAChC,uBAAuB,CAAC,iBAAiB,EAAE,QAAQ,EAAE,eAAe,CAAC,CAAC;QACtE,8BAA8B,CAAC,iBAAiB,EAAE,QAAQ,EAAE,eAAe,EAAE,aAAa,CAAC,CAAC;QAC5F,6BAA6B,CAAC,iBAAiB,EAAE,QAAQ,EAAE,eAAe,CAAC,CAAC;QAC5E,oCAAoC,CAAC,iBAAiB,EAAE,oBAAoB,CAAC,CAAC;IAChF,CAAC;IAED,IAAI,QAAQ,KAAK,UAAU,IAAI,QAAQ,KAAK,qBAAqB,EAAE,CAAC;QAClE,2BAA2B,CAAC,iBAAiB,EAAE,QAAQ,EAAE,eAAe,CAAC,CAAC;IAC5E,CAAC;IAED,iGAAiG;IACjG,IAAI,QAAQ,KAAK,cAAc,IAAI,iBAAiB,CAAC,eAAe,EAAE,CAAC;QACrE,MAAM,EAAE,YAAY,EAAE,aAAa,EAAE,GAAG,4BAA4B,CAClE,QAAQ,EACR,iBAAiB,CAAC,KAAK,IAAI,EAAE,EAC7B,iBAAiB,CAAC,eAAe,EAAE,IAAI,IAAI,EAAE,EAC7C,eAAe,CAAC,KAAK,EACrB,SAAS,CACV,CAAC;QAEF,MAAM,CAAC,IAAI,CAAC,2BAA2B,QAAQ,IAAI,iBAAiB,CAAC,KAAK,oBAAoB,YAAY,EAAE,CAAC,CAAC;QAE9G,iBAAiB,GAAG,uCAAuC,CAAC,iBAAiB,EAAE,YAAY,EAAE,aAAa,CAAC,CAAC;IAC9G,CAAC;IAED,IACE,oBAAoB,CAAC,iBAAiB;QACtC,OAAO,oBAAoB,CAAC,iBAAiB,CAAC,QAAQ,CAAC,KAAK,UAAU,EACtE,CAAC;QACD,iBAAiB,GAAG,oBAAoB,CAAC,iBAAiB,CAAC,QAAQ,CAAC,CAAC,iBAAiB,EAAE,cAAc,CAAC,CAAC;IAC1G,CAAC;IAED,MAAM,iBAAiB,GAAG,cAAc,CAAC,GAAG,CAAC;IAC7C,IACE,iBAAiB,CAAC,mBAAmB;QACrC,iBAAiB,CAAC,mBAAmB,CAAC,EAAE,kBAAkB,EAAE,iBAAiB,EAAE,CAAC;QAEhF,OAAO,kCAAkC,CACvC,QAAQ,EACR,iBAA2B,EAC3B,QAAQ,EACR,eAAe,EACf,UAAU,EACV,aAAa,EACb,SAAS,EACT,cAAc,CACf,CAAC;IACJ,OAAO,8BAA8B,CACnC,eAAe,EACf,iBAA2B,EAC3B,QAAQ,EACR,UAAU,EACV,aAAa,EACb,SAAS,EACT,cAAc,CACf,CAAC;AACJ,CAAC,CAAC;AAEF,eAAe,0BAA0B,CAAC"}

package/dist/stream/firstChunkGate.d.ts

@@ -0,0 +1,31 @@
+/**
+ * P5.6 — First-chunk gate (TTFT + optional in-band validators).
+ * Context-free port of live `src/handlers/firstChunkGate.ts`.
+ * Compose around existing SSE/AWS pumps — not a fourth pump.
+ */
+export type FirstChunkGateDecision = {
+    action: 'continue';
+} | {
+    action: 'resolve';
+} | {
+    action: 'reject';
+    error: Error;
+};
+export type FirstChunkGateChunkValidator = (chunk: string | Uint8Array) => FirstChunkGateDecision;
+export interface FirstChunkGateOptions {
+    validateFirstChunk?: FirstChunkGateChunkValidator;
+}
+export interface FirstChunkGate {
+    promise: Promise<void>;
+    /** false only when this chunk caused a reject — caller must not write this chunk. */
+    onChunk: (chunk: string | Uint8Array) => boolean;
+    onFirstChunk: () => void;
+    onError: (error: Error) => void;
+    isSettled: () => boolean;
+    dispose: () => void;
+}
+/**
+ * First-chunk gate: default behavior resolves on first chunk (empty or non-empty).
+ * Optional validator can override settle behavior and decide resolve/reject/continue.
+ */
+export declare function createFirstChunkGateFromSignal(signal: AbortSignal | undefined, timeoutMs: number | undefined, ttftTimer: ReturnType<typeof setTimeout> | undefined, options?: FirstChunkGateOptions): FirstChunkGate;

package/dist/stream/firstChunkGate.js

@@ -0,0 +1,86 @@
+/**
+ * P5.6 — First-chunk gate (TTFT + optional in-band validators).
+ * Context-free port of live `src/handlers/firstChunkGate.ts`.
+ * Compose around existing SSE/AWS pumps — not a fourth pump.
+ */
+import { TTFTTimeoutError } from '../errors.js';
+/**
+ * First-chunk gate: default behavior resolves on first chunk (empty or non-empty).
+ * Optional validator can override settle behavior and decide resolve/reject/continue.
+ */
+export function createFirstChunkGateFromSignal(signal, timeoutMs, ttftTimer, options) {
+    let settled = false;
+    let resolveFn;
+    let rejectFn;
+    const promise = new Promise((resolve, reject) => {
+        resolveFn = resolve;
+        rejectFn = reject;
+    });
+    const settleResolve = () => {
+        if (ttftTimer)
+            clearTimeout(ttftTimer);
+        if (settled)
+            return;
+        settled = true;
+        resolveFn();
+    };
+    const settleReject = (err) => {
+        if (ttftTimer)
+            clearTimeout(ttftTimer);
+        if (settled)
+            return;
+        settled = true;
+        rejectFn(err);
+    };
+    const abortHandler = () => {
+        const ms = timeoutMs ?? 0;
+        settleReject(new TTFTTimeoutError(`TTFT timeout: no first chunk within ${ms}ms`, ms));
+    };
+    const onChunk = (chunk) => {
+        if (settled)
+            return true;
+        if (!options?.validateFirstChunk) {
+            settleResolve();
+            return true;
+        }
+        try {
+            const decision = options.validateFirstChunk(chunk);
+            if (decision.action === 'resolve') {
+                settleResolve();
+                return true;
+            }
+            if (decision.action === 'reject') {
+                settleReject(decision.error);
+                return false;
+            }
+            if (decision.action === 'continue') {
+                if (ttftTimer)
+                    clearTimeout(ttftTimer);
+            }
+            return true;
+        }
+        catch (error) {
+            settleReject(error instanceof Error ? error : new Error(String(error)));
+            return false;
+        }
+    };
+    if (signal) {
+        if (signal.aborted) {
+            abortHandler();
+        }
+        else {
+            signal.addEventListener('abort', abortHandler, { once: true });
+        }
+    }
+    return {
+        promise,
+        onChunk,
+        onFirstChunk: settleResolve,
+        onError: settleReject,
+        isSettled: () => settled,
+        dispose: () => {
+            signal?.removeEventListener('abort', abortHandler);
+        },
+    };
+}
+//# sourceMappingURL=firstChunkGate.js.map

package/dist/stream/firstChunkGate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"firstChunkGate.js","sourceRoot":"","sources":["../../src/stream/firstChunkGate.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,gBAAgB,EAAE,MAAM,WAAW,CAAC;AAuB7C;;;GAGG;AACH,MAAM,UAAU,8BAA8B,CAC5C,MAA+B,EAC/B,SAA6B,EAC7B,SAAoD,EACpD,OAA+B;IAE/B,IAAI,OAAO,GAAG,KAAK,CAAC;IACpB,IAAI,SAAsB,CAAC;IAC3B,IAAI,QAA6B,CAAC;IAClC,MAAM,OAAO,GAAG,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACpD,SAAS,GAAG,OAAO,CAAC;QACpB,QAAQ,GAAG,MAAM,CAAC;IACpB,CAAC,CAAC,CAAC;IAEH,MAAM,aAAa,GAAG,GAAG,EAAE;QACzB,IAAI,SAAS;YAAE,YAAY,CAAC,SAAS,CAAC,CAAC;QACvC,IAAI,OAAO;YAAE,OAAO;QACpB,OAAO,GAAG,IAAI,CAAC;QACf,SAAS,EAAE,CAAC;IACd,CAAC,CAAC;IAEF,MAAM,YAAY,GAAG,CAAC,GAAU,EAAE,EAAE;QAClC,IAAI,SAAS;YAAE,YAAY,CAAC,SAAS,CAAC,CAAC;QACvC,IAAI,OAAO;YAAE,OAAO;QACpB,OAAO,GAAG,IAAI,CAAC;QACf,QAAQ,CAAC,GAAG,CAAC,CAAC;IAChB,CAAC,CAAC;IAEF,MAAM,YAAY,GAAG,GAAG,EAAE;QACxB,MAAM,EAAE,GAAG,SAAS,IAAI,CAAC,CAAC;QAC1B,YAAY,CAAC,IAAI,gBAAgB,CAAC,uCAAuC,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;IACxF,CAAC,CAAC;IAEF,MAAM,OAAO,GAAG,CAAC,KAA0B,EAAW,EAAE;QACtD,IAAI,OAAO;YAAE,OAAO,IAAI,CAAC;QACzB,IAAI,CAAC,OAAO,EAAE,kBAAkB,EAAE,CAAC;YACjC,aAAa,EAAE,CAAC;YAChB,OAAO,IAAI,CAAC;QACd,CAAC;QACD,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,OAAO,CAAC,kBAAkB,CAAC,KAAK,CAAC,CAAC;YACnD,IAAI,QAAQ,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;gBAClC,aAAa,EAAE,CAAC;gBAChB,OAAO,IAAI,CAAC;YACd,CAAC;YACD,IAAI,QAAQ,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;gBACjC,YAAY,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;gBAC7B,OAAO,KAAK,CAAC;YACf,CAAC;YACD,IAAI,QAAQ,CAAC,MAAM,KAAK,UAAU,EAAE,CAAC;gBACnC,IAAI,SAAS;oBAAE,YAAY,CAAC,SAAS,CAAC,CAAC;YACzC,CAAC;YACD,OAAO,IAAI,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,YAAY,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;YACxE,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC,CAAC;IAEF,IAAI,MAAM,EAAE,CAAC;QACX,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YACnB,YAAY,EAAE,CAAC;QACjB,CAAC;aAAM,CAAC;YACN,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;QACjE,CAAC;IACH,CAAC;IAED,OAAO;QACL,OAAO;QACP,OAAO;QACP,YAAY,EAAE,aAAa;QAC3B,OAAO,EAAE,YAAY;QACrB,SAAS,EAAE,GAAG,EAAE,CAAC,OAAO;QACxB,OAAO,EAAE,GAAG,EAAE;YACZ,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;QACrD,CAAC;KACF,CAAC;AACJ,CAAC"}

package/dist/stream/firstChunkGateValidators.d.ts

@@ -0,0 +1,23 @@
+/**
+ * P5.6 — In-band first-chunk validators (Anthropic overload, Responses errors).
+ * Context-free port of live `src/handlers/firstChunkGateUtils.ts`.
+ * String-SSE oriented; skip on Bedrock binary / yieldRawBytes paths.
+ */
+import { FirstChunkGateChunkValidator, FirstChunkGateDecision } from './firstChunkGate.js';
+export declare const ANTHROPIC_STREAM_OVERLOADED_ERROR_MESSAGE = "Anthropic stream overloaded";
+export declare const DETAIL_LESS_FAILURE_MESSAGE = "Provider reported a failure in the response stream without a reason";
+export declare const EMPTY_STREAM_FAILURE_MESSAGE = "Provider closed the response stream before producing any output";
+export declare function createAnthropicFirstMeaningfulChunkValidator(): FirstChunkGateChunkValidator;
+export type ResponsesInBandErrorValidator = FirstChunkGateChunkValidator & {
+    onStreamEnd: () => FirstChunkGateDecision;
+};
+/**
+ * A provider can answer 200 OK, send `response.created`, and only then report a failure in-band.
+ * Holding until generation starts lets the load balancer fall back instead of handing the client
+ * a broken 200 stream.
+ *
+ * Azure splits that failure across two frames — `response.failed` with `error: null`, then an
+ * `error` frame with the reason — so a detail-less failure holds one more frame rather than
+ * falling back with an unknown cause.
+ */
+export declare function createResponsesInBandErrorValidator(): ResponsesInBandErrorValidator;

package/dist/stream/firstChunkGateValidators.js

@@ -0,0 +1,174 @@
+/**
+ * P5.6 — In-band first-chunk validators (Anthropic overload, Responses errors).
+ * Context-free port of live `src/handlers/firstChunkGateUtils.ts`.
+ * String-SSE oriented; skip on Bedrock binary / yieldRawBytes paths.
+ */
+import { APIError, ProviderStreamError } from '../errors.js';
+import { logger } from '../logger.js';
+import { getSseDataLines, getSseDataPayload } from '../utils/sseUtils.js';
+export const ANTHROPIC_STREAM_OVERLOADED_ERROR_MESSAGE = 'Anthropic stream overloaded';
+const RESPONSES_PRE_OUTPUT_EVENT_TYPES = new Set(['response.created', 'response.in_progress', 'response.queued']);
+const RESPONSES_FAILURE_EVENT_TYPES = new Set(['response.failed', 'error']);
+const MAX_HELD_FRAMES = 4;
+export const DETAIL_LESS_FAILURE_MESSAGE = 'Provider reported a failure in the response stream without a reason';
+export const EMPTY_STREAM_FAILURE_MESSAGE = 'Provider closed the response stream before producing any output';
+const IN_BAND_ERROR_STATUS_BY_TOKEN = {
+    too_many_requests: 429,
+    rate_limit_exceeded: 429,
+    no_capacity: 429,
+    insufficient_quota: 429,
+    // Observed from OpenAI alongside insufficient_quota; another account may still have credit.
+    credit_balance_exhausted: 429,
+    server_error: 500,
+    api_error: 500,
+    overloaded_error: 503,
+    authentication_error: 401,
+    permission_error: 403,
+    permission_denied: 403,
+    not_found_error: 404,
+    invalid_request_error: 400,
+    invalid_prompt: 400,
+    context_length_exceeded: 400,
+    content_filter: 400,
+};
+function isAnthropicStreamOverloadTransformedChunk(chunk) {
+    if (!chunk.includes('overloaded_error'))
+        return false;
+    for (const payload of getSseDataLines(chunk)) {
+        try {
+            if (JSON.parse(payload)?.choices?.[0]?.finish_reason === 'overloaded_error') {
+                return true;
+            }
+        }
+        catch {
+            // ignore non-JSON / [DONE]
+        }
+    }
+    return false;
+}
+export function createAnthropicFirstMeaningfulChunkValidator() {
+    let firstMeaningfulChunkSeen = false;
+    return (chunk) => {
+        if (firstMeaningfulChunkSeen) {
+            return { action: 'continue' };
+        }
+        const isEmpty = typeof chunk === 'string' ? !chunk.length : !chunk || chunk.byteLength === 0;
+        if (isEmpty) {
+            return { action: 'continue' };
+        }
+        firstMeaningfulChunkSeen = true;
+        if (typeof chunk === 'string' && isAnthropicStreamOverloadTransformedChunk(chunk)) {
+            return {
+                action: 'reject',
+                error: new APIError(ANTHROPIC_STREAM_OVERLOADED_ERROR_MESSAGE, { statusCode: 503 }),
+            };
+        }
+        return { action: 'resolve' };
+    };
+}
+function statusForInBandError(errorType, errorCode) {
+    // Code first: when a provider sends both, it is the more specific of the two.
+    for (const token of [errorCode, errorType]) {
+        const status = token ? IN_BAND_ERROR_STATUS_BY_TOKEN[token.toLowerCase()] : undefined;
+        if (status !== undefined)
+            return status;
+    }
+    // 502 so an unknown failure still falls back; the log is how this table grows.
+    logger.warn('Unrecognized in-band Responses stream error type/code; mapping to 502', { errorType, errorCode });
+    return 502;
+}
+function extractFailureDetail(event) {
+    const nested = event.response?.error ?? event.error;
+    return {
+        type: nested?.type ?? null,
+        code: nested?.code ?? event.code ?? null,
+        message: nested?.message ?? event.message ?? DETAIL_LESS_FAILURE_MESSAGE,
+    };
+}
+function buildInBandError({ type, code, message }) {
+    // A span carries only a message, so the type/code are folded into it. Clients read them from
+    // providerErrorType/Code instead, so providerMessage keeps the provider's wording unchanged.
+    const detail = [type, code].filter(Boolean).join('/');
+    return new ProviderStreamError(detail ? `${message} (${detail})` : message, {
+        statusCode: statusForInBandError(type, code),
+        providerErrorType: type,
+        providerErrorCode: code,
+        providerMessage: message,
+    });
+}
+/** Azure never sent the follow-up detail, so we fail with a status that still falls back. */
+function buildDetailLessError() {
+    return new ProviderStreamError(DETAIL_LESS_FAILURE_MESSAGE, { statusCode: 502 });
+}
+function classifyResponsesFrame(frame) {
+    const payload = getSseDataPayload(frame);
+    if (payload === null || payload.length === 0) {
+        return { kind: 'hold' };
+    }
+    let event;
+    try {
+        event = JSON.parse(payload);
+    }
+    catch {
+        // `[DONE]`, or a provider streaming something else entirely: not ours to hold.
+        return { kind: 'generation' };
+    }
+    const eventType = typeof event.type === 'string' ? event.type : '';
+    if (RESPONSES_PRE_OUTPUT_EVENT_TYPES.has(eventType)) {
+        return { kind: 'hold' };
+    }
+    if (RESPONSES_FAILURE_EVENT_TYPES.has(eventType)) {
+        const detail = extractFailureDetail(event);
+        return detail.type || detail.code ? { kind: 'failure', detail } : { kind: 'failureAwaitingDetail' };
+    }
+    // Unrecognized events resolve, so an unfamiliar stream shape can never stall the gate.
+    return { kind: 'generation' };
+}
+/**
+ * A provider can answer 200 OK, send `response.created`, and only then report a failure in-band.
+ * Holding until generation starts lets the load balancer fall back instead of handing the client
+ * a broken 200 stream.
+ *
+ * Azure splits that failure across two frames — `response.failed` with `error: null`, then an
+ * `error` frame with the reason — so a detail-less failure holds one more frame rather than
+ * falling back with an unknown cause.
+ */
+export function createResponsesInBandErrorValidator() {
+    let heldFrames = 0;
+    let awaitingFailureDetail = false;
+    const validator = (chunk) => {
+        if (typeof chunk !== 'string') {
+            return { action: 'resolve' };
+        }
+        if (!chunk.length) {
+            return { action: 'continue' };
+        }
+        // Relies on readStream delivering one SSE frame per chunk; a chunk carrying several frames
+        // would have its data lines concatenated here and read as generation.
+        const verdict = classifyResponsesFrame(chunk);
+        if (verdict.kind === 'failure') {
+            return { action: 'reject', error: buildInBandError(verdict.detail) };
+        }
+        if (verdict.kind === 'generation') {
+            // Generation after an announced failure means the detail is never coming.
+            if (awaitingFailureDetail)
+                return { action: 'reject', error: buildDetailLessError() };
+            return { action: 'resolve' };
+        }
+        if (verdict.kind === 'failureAwaitingDetail') {
+            awaitingFailureDetail = true;
+        }
+        if (++heldFrames > MAX_HELD_FRAMES) {
+            return awaitingFailureDetail ? { action: 'reject', error: buildDetailLessError() } : { action: 'resolve' };
+        }
+        return { action: 'continue' };
+    };
+    validator.onStreamEnd = () => ({
+        action: 'reject',
+        error: awaitingFailureDetail
+            ? buildDetailLessError()
+            : new ProviderStreamError(EMPTY_STREAM_FAILURE_MESSAGE, { statusCode: 502 }),
+    });
+    return validator;
+}
+//# sourceMappingURL=firstChunkGateValidators.js.map

package/dist/stream/firstChunkGateValidators.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"firstChunkGateValidators.js","sourceRoot":"","sources":["../../src/stream/firstChunkGateValidators.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,QAAQ,EAAE,mBAAmB,EAAE,MAAM,WAAW,CAAC;AAC1D,OAAO,EAAE,MAAM,EAAE,MAAM,WAAW,CAAC;AACnC,OAAO,EAAE,eAAe,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAGvE,MAAM,CAAC,MAAM,yCAAyC,GAAG,6BAA6B,CAAC;AAEvF,MAAM,gCAAgC,GAAG,IAAI,GAAG,CAAC,CAAC,kBAAkB,EAAE,sBAAsB,EAAE,iBAAiB,CAAC,CAAC,CAAC;AAElH,MAAM,6BAA6B,GAAG,IAAI,GAAG,CAAC,CAAC,iBAAiB,EAAE,OAAO,CAAC,CAAC,CAAC;AAE5E,MAAM,eAAe,GAAG,CAAC,CAAC;AAE1B,MAAM,CAAC,MAAM,2BAA2B,GAAG,qEAAqE,CAAC;AACjH,MAAM,CAAC,MAAM,4BAA4B,GAAG,iEAAiE,CAAC;AAE9G,MAAM,6BAA6B,GAA2B;IAC5D,iBAAiB,EAAE,GAAG;IACtB,mBAAmB,EAAE,GAAG;IACxB,WAAW,EAAE,GAAG;IAChB,kBAAkB,EAAE,GAAG;IACvB,4FAA4F;IAC5F,wBAAwB,EAAE,GAAG;IAC7B,YAAY,EAAE,GAAG;IACjB,SAAS,EAAE,GAAG;IACd,gBAAgB,EAAE,GAAG;IACrB,oBAAoB,EAAE,GAAG;IACzB,gBAAgB,EAAE,GAAG;IACrB,iBAAiB,EAAE,GAAG;IACtB,eAAe,EAAE,GAAG;IACpB,qBAAqB,EAAE,GAAG;IAC1B,cAAc,EAAE,GAAG;IACnB,uBAAuB,EAAE,GAAG;IAC5B,cAAc,EAAE,GAAG;CACpB,CAAC;AAEF,SAAS,yCAAyC,CAAC,KAAa;IAC9D,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,kBAAkB,CAAC;QAAE,OAAO,KAAK,CAAC;IACtD,KAAK,MAAM,OAAO,IAAI,eAAe,CAAC,KAAK,CAAC,EAAE,CAAC;QAC7C,IAAI,CAAC;YACH,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,aAAa,KAAK,kBAAkB,EAAE,CAAC;gBAC5E,OAAO,IAAI,CAAC;YACd,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,2BAA2B;QAC7B,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,UAAU,4CAA4C;IAC1D,IAAI,wBAAwB,GAAG,KAAK,CAAC;IACrC,OAAO,CAAC,KAA0B,EAA0B,EAAE;QAC5D,IAAI,wBAAwB,EAAE,CAAC;YAC7B,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;QAChC,CAAC;QAED,MAAM,OAAO,GAAG,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,KAAK,CAAC,UAAU,KAAK,CAAC,CAAC;QAC7F,IAAI,OAAO,EAAE,CAAC;YACZ,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;QAChC,CAAC;QAED,wBAAwB,GAAG,IAAI,CAAC;QAChC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,yCAAyC,CAAC,KAAK,CAAC,EAAE,CAAC;YAClF,OAAO;gBACL,MAAM,EAAE,QAAQ;gBAChB,KAAK,EAAE,IAAI,QAAQ,CAAC,yCAAyC,EAAE,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC;aACpF,CAAC;QACJ,CAAC;QAED,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;IAC/B,CAAC,CAAC;AACJ,CAAC;AAaD,SAAS,oBAAoB,CAAC,SAAwB,EAAE,SAAwB;IAC9E,8EAA8E;IAC9E,KAAK,MAAM,KAAK,IAAI,CAAC,SAAS,EAAE,SAAS,CAAC,EAAE,CAAC;QAC3C,MAAM,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,6BAA6B,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACtF,IAAI,MAAM,KAAK,SAAS;YAAE,OAAO,MAAM,CAAC;IAC1C,CAAC;IAED,+EAA+E;IAC/E,MAAM,CAAC,IAAI,CAAC,uEAAuE,EAAE,EAAE,SAAS,EAAE,SAAS,EAAE,CAAC,CAAC;IAC/G,OAAO,GAAG,CAAC;AACb,CAAC;AAID,SAAS,oBAAoB,CAAC,KAA2B;IACvD,MAAM,MAAM,GAAG,KAAK,CAAC,QAAQ,EAAE,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC;IACpD,OAAO;QACL,IAAI,EAAE,MAAM,EAAE,IAAI,IAAI,IAAI;QAC1B,IAAI,EAAE,MAAM,EAAE,IAAI,IAAI,KAAK,CAAC,IAAI,IAAI,IAAI;QACxC,OAAO,EAAE,MAAM,EAAE,OAAO,IAAI,KAAK,CAAC,OAAO,IAAI,2BAA2B;KACzE,CAAC;AACJ,CAAC;AAED,SAAS,gBAAgB,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAiB;IAC9D,6FAA6F;IAC7F,6FAA6F;IAC7F,MAAM,MAAM,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAEtD,OAAO,IAAI,mBAAmB,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,OAAO,KAAK,MAAM,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE;QAC1E,UAAU,EAAE,oBAAoB,CAAC,IAAI,EAAE,IAAI,CAAC;QAC5C,iBAAiB,EAAE,IAAI;QACvB,iBAAiB,EAAE,IAAI;QACvB,eAAe,EAAE,OAAO;KACzB,CAAC,CAAC;AACL,CAAC;AAED,6FAA6F;AAC7F,SAAS,oBAAoB;IAC3B,OAAO,IAAI,mBAAmB,CAAC,2BAA2B,EAAE,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC,CAAC;AACnF,CAAC;AAYD,SAAS,sBAAsB,CAAC,KAAa;IAC3C,MAAM,OAAO,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;IACzC,IAAI,OAAO,KAAK,IAAI,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7C,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC;IAC1B,CAAC;IAED,IAAI,KAA2B,CAAC;IAChC,IAAI,CAAC;QACH,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAyB,CAAC;IACtD,CAAC;IAAC,MAAM,CAAC;QACP,+EAA+E;QAC/E,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC;IAChC,CAAC;IAED,MAAM,SAAS,GAAG,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;IACnE,IAAI,gCAAgC,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC;QACpD,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC;IAC1B,CAAC;IACD,IAAI,6BAA6B,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC;QACjD,MAAM,MAAM,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;QAC3C,OAAO,MAAM,CAAC,IAAI,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,uBAAuB,EAAE,CAAC;IACtG,CAAC;IAED,uFAAuF;IACvF,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC;AAChC,CAAC;AAMD;;;;;;;;GAQG;AACH,MAAM,UAAU,mCAAmC;IACjD,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,qBAAqB,GAAG,KAAK,CAAC;IAElC,MAAM,SAAS,GAAG,CAAC,KAA0B,EAA0B,EAAE;QACvE,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC9B,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;QAC/B,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC;YAClB,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;QAChC,CAAC;QAED,2FAA2F;QAC3F,sEAAsE;QACtE,MAAM,OAAO,GAAG,sBAAsB,CAAC,KAAK,CAAC,CAAC;QAC9C,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YAC/B,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,gBAAgB,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QACvE,CAAC;QACD,IAAI,OAAO,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;YAClC,0EAA0E;YAC1E,IAAI,qBAAqB;gBAAE,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,oBAAoB,EAAE,EAAE,CAAC;YACtF,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;QAC/B,CAAC;QACD,IAAI,OAAO,CAAC,IAAI,KAAK,uBAAuB,EAAE,CAAC;YAC7C,qBAAqB,GAAG,IAAI,CAAC;QAC/B,CAAC;QAED,IAAI,EAAE,UAAU,GAAG,eAAe,EAAE,CAAC;YACnC,OAAO,qBAAqB,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,oBAAoB,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;QAC7G,CAAC;QACD,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;IAChC,CAAC,CAAC;IAEF,SAAS,CAAC,WAAW,GAAG,GAA2B,EAAE,CAAC,CAAC;QACrD,MAAM,EAAE,QAAQ;QAChB,KAAK,EAAE,qBAAqB;YAC1B,CAAC,CAAC,oBAAoB,EAAE;YACxB,CAAC,CAAC,IAAI,mBAAmB,CAAC,4BAA4B,EAAE,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC;KAC/E,CAAC,CAAC;IAEH,OAAO,SAAS,CAAC;AACnB,CAAC"}

package/dist/stream/index.d.ts

@@ -0,0 +1,19 @@
+export { createFirstChunkGateFromSignal } from './firstChunkGate.js';
+export type { FirstChunkGate, FirstChunkGateChunkValidator, FirstChunkGateDecision, FirstChunkGateOptions, } from './firstChunkGate.js';
+export { createAnthropicFirstMeaningfulChunkValidator, createResponsesInBandErrorValidator, } from './firstChunkGateValidators.js';
+export type { ResponsesInBandErrorValidator } from './firstChunkGateValidators.js';
+export { formatInternalMetadataSseFrame, stripInternalMetadataSseFrames } from './internalMetadataSse.js';
+export { pumpProviderAwsEventStream } from './pumpProviderAwsEventStream.js';
+export { pumpProviderSseStream } from './pumpProviderSseStream.js';
+export { readAwsEventFrames } from './readAwsEventFrames.js';
+export { readSseFrames } from './readSseFrames.js';
+export { resolvePumpAbortSignal } from './resolvePumpAbortSignal.js';
+export type { ResolvePumpAbortSignalInput } from './resolvePumpAbortSignal.js';
+export { mapToMidStreamError, resolveStreamEndReason } from './resolveStreamEndReason.js';
+export type { ResolveStreamEndReasonInput } from './resolveStreamEndReason.js';
+export { shouldReadAwsEventStream } from './shouldReadAwsEventStream.js';
+export type { ShouldReadAwsEventStreamInput } from './shouldReadAwsEventStream.js';
+export { shouldRequireAzureStreamSleep } from './shouldRequireAzureStreamSleep.js';
+export { getStreamModeSplitPattern } from './splitPatterns.js';
+export type { GetStreamModeSplitPatternInput } from './splitPatterns.js';
+export type { AwsEventFrame, PumpProviderAwsEventStreamOptions, PumpProviderBaseOptions, PumpProviderSseOptions, PumpResult, ReadAwsEventFramesOptions, ReadAwsEventFramesResult, ReadSseFramesOptions, ReadSseFramesResult, SplitPattern, SseCommentMode, StreamTransformFn, } from './types.js';