@draht/ai 2026.10.31 → 2026.11.5-1

package/dist/api/azure-openai-responses.js

@@ -85,12 +85,6 @@ export const stream = (model, context, options) => {
             stream.end();
         }
         catch (error) {
-            for (const block of output.content) {
-                delete block.index;
-                // Streaming scratch buffers are only used during parsing; never persist them.
-                delete block.partialJson;
-                delete block.customInput;
-            }
             output.stopReason = options?.signal?.aborted ? "aborted" : "error";
             output.errorMessage = formatAzureOpenAIError(error);
             stream.push({ type: "error", reason: output.stopReason, error: output });
@@ -152,7 +146,7 @@ function buildParams(model, context, options, deploymentName, grammarToolInputPr
         model: deploymentName,
         input: messages,
         stream: true,
-        prompt_cache_key: clampOpenAIPromptCacheKey(options?.sessionId),
+        prompt_cache_key: model.compat?.promptCacheKeyMode === "disabled" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),
         store: false,
     };
     if (options?.maxTokens) {

package/dist/api/azure-openai-responses.js.map

@@ -1 +1 @@
-{"version":3,"file":"azure-openai-responses.js","sourceRoot":"","sources":["../../src/api/azure-openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,QAAQ,CAAC;AAErC,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAElD,OAAO,EAAE,mBAAmB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrF,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC1D,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrG,OAAO,EAA6B,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,0BAA0B,CAAC;AAChH,OAAO,EAAE,gCAAgC,EAAE,MAAM,2BAA2B,CAAC;AAC7E,OAAO,EAAE,yBAAyB,EAAE,MAAM,0BAA0B,CAAC;AACrE,OAAO,EACN,wBAAwB,EACxB,qBAAqB,EACrB,uBAAuB,EACvB,sBAAsB,GACtB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAE9E,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC;AAC3F,uGAAuG;AACvG,MAAM,kCAAkC,GAAG,EAAE,CAAC;AAE9C,SAAS,sBAAsB,CAAC,KAAc;IAC7C,OAAO,mBAAmB,CAAC,sBAAsB,CAAC,KAAK,CAAC,EAAE,wBAAwB,CAAC,CAAC;AACrF,CAAC;AASD;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAA0E,CAC5F,KAAsC,EACtC,OAA0B,EAC1B,OAAqC,EACP,EAAE;IAChC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IACjD,MAAM,iBAAiB,GAAG,iBAAiB,CAAC,OAAO,EAAE,KAAK,CAAC,MAAM,EAAE,8BAA8B,CAAC,CAAC;IAEnG,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE;QACX,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;QAE7D,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,wBAA+B;YACpC,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,SAAS;YACrB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,6BAA6B;YAC7B,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,CAAC;YAC/B,IAAI,CAAC,MAAM,EAAE,CAAC;gBACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;YAC/D,CAAC;YACD,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;YACpD,MAAM,0BAA0B,GAAG,gCAAgC,CAClE,gBAAgB,CAAC,iBAAiB,CAAC,QAAQ,CAAC,EAC5C,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK,CACjD,CAAC;YACF,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,iBAAiB,EAAE,OAAO,EAAE,cAAc,EAAE,0BAA0B,CAAC,CAAC;YACxG,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAA2C,CAAC;YACtD,CAAC;YACD,MAAM,cAAc,GAAG;gBACtB,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACtD,GAAG,CAAC,OAAO,EAAE,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3E,UAAU,EAAE,CAAC;aACb,CAAC;YACF,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,oBAAoB,CAClE,GAAG,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,YAAY,EAAE,EACpE;gBACC,UAAU,EAAE,OAAO,EAAE,UAAU;gBAC/B,eAAe,EAAE,OAAO,EAAE,eAAe;gBACzC,MAAM,EAAE,OAAO,EAAE,MAAM;aACvB,CACD,CAAC;YACF,MAAM,OAAO,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,eAAe,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;YAC5G,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAEhD,MAAM,sBAAsB,CAAC,YAAY,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE;gBACjE,qBAAqB,EAAE,OAAO,EAAE,qBAAqB;gBACrD,0BAA0B;aAC1B,CAAC,CAAC;YAEH,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;gBACrC,MAAM,IAAI,KAAK,CAAC,2DAA2D,CAAC,CAAC;YAC9E,CAAC;YACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,YAAY,IAAI,2BAA2B,CAAC,CAAC;YACrE,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpC,OAAQ,KAA4B,CAAC,KAAK,CAAC;gBAC3C,8EAA8E;gBAC9E,OAAQ,KAAkC,CAAC,WAAW,CAAC;gBACvD,OAAQ,KAAmC,CAAC,WAAW,CAAC;YACzD,CAAC;YACD,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,sBAAsB,CAAC,KAAK,CAAC,CAAC;YACpD,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IACF,CAAC,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AACf,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAAkE,CAC1F,KAAsC,EACtC,OAA0B,EAC1B,OAA6B,EACC,EAAE;IAChC,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,CAAC;IAC/B,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,IAAI,GAAG;QACZ,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC;QACpD,UAAU,EAAE,OAAO,EAAE,UAAU;KACO,CAAC;IACxC,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvG,MAAM,eAAe,GAAG,gBAAgB,KAAK,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IAElF,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE;QAC7B,GAAG,IAAI;QACP,eAAe;KACuB,CAAC,CAAC;AAC1C,CAAC,CAAC;AAEF,SAAS,YAAY,CAAC,KAAsC,EAAE,MAAc,EAAE,OAAqC;IAClH,MAAM,OAAO,GAAG,EAAE,GAAG,kBAAkB,CAAC,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IAEtF,IAAI,OAAO,EAAE,OAAO,EAAE,CAAC;QACtB,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC;IACzC,CAAC;IAED,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAEnE,OAAO,IAAI,WAAW,CAAC;QACtB,MAAM;QACN,UAAU;QACV,uBAAuB,EAAE,IAAI;QAC7B,KAAK,EAAE,OAAO,EAAE,KAAK;QACrB,cAAc,EAAE,OAAO;QACvB,OAAO,EAAE,OAAO;KAChB,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CACnB,KAAsC,EACtC,OAA0B,EAC1B,OAAgD,EAChD,cAAsB,EACtB,0BAA0B,GAAgC,gCAAgC,CACzF,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,EAClC,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK,CACjD;IAED,MAAM,uBAAuB,GAAG,KAAK,CAAC,MAAM,EAAE,uBAAuB,IAAI,KAAK,CAAC;IAC/E,MAAM,kBAAkB,GAAG,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK,CAAC;IACrE,MAAM,eAAe,GAAG,sBAAsB,CAC7C,OAAO,CAAC,QAAQ,EAChB,uBAAuB,CAAC;QACvB,8BAA8B,EAAE,KAAK,CAAC,MAAM,EAAE,8BAA8B,IAAI,KAAK;QACrF,uBAAuB;QACvB,kBAAkB;KAClB,CAAC,CACF,CAAC;IACF,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,yBAAyB,EAAE;QACpF,0BAA0B;QAC1B,8BAA8B,EAAE,KAAK,CAAC,MAAM,EAAE,8BAA8B,IAAI,KAAK;QACrF,uBAAuB;QACvB,kBAAkB;QAClB,WAAW,EAAE;YACZ,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,IAAI;YAC5D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK;SAC7E;KACD,CAAC,CAAC;IAEH,MAAM,MAAM,GAAkC;QAC7C,KAAK,EAAE,cAAc;QACrB,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,gBAAgB,EAAE,yBAAyB,CAAC,OAAO,EAAE,SAAS,CAAC;QAC/D,KAAK,EAAE,KAAK;KACZ,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,kCAAkC,CAAC,CAAC;IAC5F,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,eAAe,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7C,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,eAAe,CAAC,YAAY,EAAE;YAClE,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,IAAI;YAC5D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK;SAC7E,CAAC,CAAC;IACJ,CAAC;IACD,IAAI,OAAO,EAAE,UAAU,KAAK,SAAS,EAAE,CAAC;QACvC,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,UAAU,CAAC;IACzC,CAAC;IAED,MAAM,eAAe,GAAG,OAAO,EAAE,eAAe,IAAI,CAAC,OAAO,EAAE,gBAAgB,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;IACvG,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,eAAe,EAAE,CAAC;YACrB,MAAM,MAAM,GAAG,OAAO,EAAE,eAAe;gBACtC,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,eAAe,CAAC;gBAChF,CAAC,CAAC,eAAe,CAAC;YACnB,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,MAAwD;gBAChE,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,KAAK,CAAC,gBAAgB,EAAE,GAAG,KAAK,IAAI,EAAE,CAAC;YACjD,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,CAAC,KAAK,CAAC,gBAAgB,EAAE,GAAG,IAAI,MAAM,CAAmD;aACjG,CAAC;QACH,CAAC;IACF,CAAC;IAED,+EAA+E;IAC/E,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,eAAe,IAAI,KAAK,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;IACvG,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACvC,CAAC;IAED,OAAO,MAAM,CAAC;AACf,CAAC","sourcesContent":["import { AzureOpenAI } from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { clampThinkingLevel } from \"../models.ts\";\nimport type { Api, AssistantMessage, Model, SimpleStreamOptions, StreamFunction, TranscriptContext } from \"../types.ts\";\nimport { formatProviderError, normalizeProviderError } from \"../utils/error-body.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord } from \"../utils/headers.ts\";\nimport { getIdentityHeaders } from \"../utils/identity.ts\";\nimport { retryProviderRequest } from \"../utils/provider-retry.ts\";\nimport { getDeclaredTools, resolveTranscript, resolveTranscriptTools } from \"../utils/transcript.ts\";\nimport { type AzureEndpointOptions, resolveAzureConfig, resolveDeploymentName } from \"./azure-openai-config.ts\";\nimport { createGrammarToolInputProperties } from \"./constrained-sampling.ts\";\nimport { clampOpenAIPromptCacheKey } from \"./openai-prompt-cache.ts\";\nimport {\n\tconvertResponsesMessages,\n\tconvertResponsesTools,\n\tgetResponsesToolSupport,\n\tprocessResponsesStream,\n} from \"./openai-responses-shared.ts\";\nimport { buildBaseOptions, resolveSamplingParams } from \"./simple-options.ts\";\n\nconst AZURE_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\", \"azure\"]);\n// OpenAI Responses rejects max_output_tokens below 16: https://github.com/badlogic/pi-mono/issues/6265\nconst OPENAI_RESPONSES_MIN_OUTPUT_TOKENS = 16;\n\nfunction formatAzureOpenAIError(error: unknown): string {\n\treturn formatProviderError(normalizeProviderError(error), \"Azure OpenAI API error\");\n}\n\n// Azure OpenAI Responses-specific options\nexport interface AzureOpenAIResponsesOptions extends AzureEndpointOptions {\n\treasoningEffort?: \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\ttoolChoice?: ResponseCreateParamsStreaming[\"tool_choice\"];\n\treasoningSummary?: \"auto\" | \"detailed\" | \"concise\" | null;\n}\n\n/**\n * Generate function for Azure OpenAI Responses API\n */\nexport const stream: StreamFunction<\"azure-openai-responses\", AzureOpenAIResponsesOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions?: AzureOpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\tconst normalizedContext = resolveTranscript(context, model.compat?.supportsMidConvoSystemMessages);\n\n\t// Start async processing\n\t(async () => {\n\t\tconst deploymentName = resolveDeploymentName(model, options);\n\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"azure-openai-responses\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"pending\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create Azure OpenAI client\n\t\t\tconst apiKey = options?.apiKey;\n\t\t\tif (!apiKey) {\n\t\t\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t\t\t}\n\t\t\tconst client = createClient(model, apiKey, options);\n\t\t\tconst grammarToolInputProperties = createGrammarToolInputProperties(\n\t\t\t\tgetDeclaredTools(normalizedContext.messages),\n\t\t\t\tmodel.compat?.supportsOpenAIGrammarTools ?? false,\n\t\t\t);\n\t\t\tlet params = buildParams(model, normalizedContext, options, deploymentName, grammarToolInputProperties);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst requestOptions = {\n\t\t\t\t...(options?.signal ? { signal: options.signal } : {}),\n\t\t\t\t...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),\n\t\t\t\tmaxRetries: 0,\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await retryProviderRequest(\n\t\t\t\t() => client.responses.create(params, requestOptions).withResponse(),\n\t\t\t\t{\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t},\n\t\t\t);\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tonProviderStreamEvent: options?.onProviderStreamEvent,\n\t\t\t\tgrammarToolInputProperties,\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"pending\") {\n\t\t\t\tthrow new Error(\"Azure OpenAI Responses stream ended without a stop reason\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage || \"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// Streaming scratch buffers are only used during parsing; never persist them.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t\tdelete (block as { customInput?: unknown }).customInput;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = formatAzureOpenAIError(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimple: StreamFunction<\"azure-openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey;\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = {\n\t\t...buildBaseOptions(model, context, options, apiKey),\n\t\ttoolChoice: options?.toolChoice,\n\t} satisfies AzureOpenAIResponsesOptions;\n\tconst clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;\n\tconst reasoningEffort = clampedReasoning === \"off\" ? undefined : clampedReasoning;\n\n\treturn stream(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies AzureOpenAIResponsesOptions);\n};\n\nfunction createClient(model: Model<\"azure-openai-responses\">, apiKey: string, options?: AzureOpenAIResponsesOptions) {\n\tconst headers = { ...getIdentityHeaders(model, options?.identity), ...model.headers };\n\n\tif (options?.headers) {\n\t\tObject.assign(headers, options.headers);\n\t}\n\n\tconst { baseUrl, apiVersion } = resolveAzureConfig(model, options);\n\n\treturn new AzureOpenAI({\n\t\tapiKey,\n\t\tapiVersion,\n\t\tdangerouslyAllowBrowser: true,\n\t\tfetch: options?.fetch,\n\t\tdefaultHeaders: headers,\n\t\tbaseURL: baseUrl,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions: AzureOpenAIResponsesOptions | undefined,\n\tdeploymentName: string,\n\tgrammarToolInputProperties: ReadonlyMap<string, string> = createGrammarToolInputProperties(\n\t\tgetDeclaredTools(context.messages),\n\t\tmodel.compat?.supportsOpenAIGrammarTools ?? false,\n\t),\n) {\n\tconst supportsAdditionalTools = model.compat?.supportsAdditionalTools ?? false;\n\tconst supportsToolSearch = model.compat?.supportsToolSearch ?? false;\n\tconst transcriptTools = resolveTranscriptTools(\n\t\tcontext.messages,\n\t\tgetResponsesToolSupport({\n\t\t\tsupportsMidConvoSystemMessages: model.compat?.supportsMidConvoSystemMessages ?? false,\n\t\t\tsupportsAdditionalTools,\n\t\t\tsupportsToolSearch,\n\t\t}),\n\t);\n\tconst messages = convertResponsesMessages(model, context, AZURE_TOOL_CALL_PROVIDERS, {\n\t\tgrammarToolInputProperties,\n\t\tsupportsMidConvoSystemMessages: model.compat?.supportsMidConvoSystemMessages ?? false,\n\t\tsupportsAdditionalTools,\n\t\tsupportsToolSearch,\n\t\ttoolOptions: {\n\t\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? true,\n\t\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\t},\n\t});\n\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: deploymentName,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: clampOpenAIPromptCacheKey(options?.sessionId),\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = Math.max(options.maxTokens, OPENAI_RESPONSES_MIN_OUTPUT_TOKENS);\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (transcriptTools.requestTools.length > 0) {\n\t\tparams.tools = convertResponsesTools(transcriptTools.requestTools, {\n\t\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? true,\n\t\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\t});\n\t}\n\tif (options?.toolChoice !== undefined) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tconst reasoningEffort = options?.reasoningEffort ?? (options?.reasoningSummary ? \"medium\" : undefined);\n\tif (model.reasoning) {\n\t\tif (reasoningEffort) {\n\t\t\tconst effort = options?.reasoningEffort\n\t\t\t\t? (model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort)\n\t\t\t\t: reasoningEffort;\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.thinkingLevelMap?.off !== null) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: (model.thinkingLevelMap?.off ?? \"none\") as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t};\n\t\t}\n\t}\n\n\t// Last so model and request sampling parameters override named request fields.\n\tconst samplingParams = resolveSamplingParams(model, reasoningEffort ?? \"off\", options?.samplingParams);\n\tif (samplingParams) {\n\t\tObject.assign(params, samplingParams);\n\t}\n\n\treturn params;\n}\n"]}
+{"version":3,"file":"azure-openai-responses.js","sourceRoot":"","sources":["../../src/api/azure-openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,QAAQ,CAAC;AAErC,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAElD,OAAO,EAAE,mBAAmB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrF,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC1D,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,gBAAgB,EAAE,iBAAiB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrG,OAAO,EAA6B,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,0BAA0B,CAAC;AAChH,OAAO,EAAE,gCAAgC,EAAE,MAAM,2BAA2B,CAAC;AAC7E,OAAO,EAAE,yBAAyB,EAAE,MAAM,0BAA0B,CAAC;AACrE,OAAO,EACN,wBAAwB,EACxB,qBAAqB,EACrB,uBAAuB,EACvB,sBAAsB,GACtB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAE9E,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC;AAC3F,uGAAuG;AACvG,MAAM,kCAAkC,GAAG,EAAE,CAAC;AAE9C,SAAS,sBAAsB,CAAC,KAAc;IAC7C,OAAO,mBAAmB,CAAC,sBAAsB,CAAC,KAAK,CAAC,EAAE,wBAAwB,CAAC,CAAC;AACrF,CAAC;AASD;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAA0E,CAC5F,KAAsC,EACtC,OAA0B,EAC1B,OAAqC,EACP,EAAE;IAChC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IACjD,MAAM,iBAAiB,GAAG,iBAAiB,CAAC,OAAO,EAAE,KAAK,CAAC,MAAM,EAAE,8BAA8B,CAAC,CAAC;IAEnG,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE;QACX,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;QAE7D,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,wBAA+B;YACpC,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,SAAS;YACrB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,6BAA6B;YAC7B,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,CAAC;YAC/B,IAAI,CAAC,MAAM,EAAE,CAAC;gBACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;YAC/D,CAAC;YACD,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;YACpD,MAAM,0BAA0B,GAAG,gCAAgC,CAClE,gBAAgB,CAAC,iBAAiB,CAAC,QAAQ,CAAC,EAC5C,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK,CACjD,CAAC;YACF,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,iBAAiB,EAAE,OAAO,EAAE,cAAc,EAAE,0BAA0B,CAAC,CAAC;YACxG,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAA2C,CAAC;YACtD,CAAC;YACD,MAAM,cAAc,GAAG;gBACtB,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACtD,GAAG,CAAC,OAAO,EAAE,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3E,UAAU,EAAE,CAAC;aACb,CAAC;YACF,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,oBAAoB,CAClE,GAAG,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,YAAY,EAAE,EACpE;gBACC,UAAU,EAAE,OAAO,EAAE,UAAU;gBAC/B,eAAe,EAAE,OAAO,EAAE,eAAe;gBACzC,MAAM,EAAE,OAAO,EAAE,MAAM;aACvB,CACD,CAAC;YACF,MAAM,OAAO,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,eAAe,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;YAC5G,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAEhD,MAAM,sBAAsB,CAAC,YAAY,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE;gBACjE,qBAAqB,EAAE,OAAO,EAAE,qBAAqB;gBACrD,0BAA0B;aAC1B,CAAC,CAAC;YAEH,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;gBACrC,MAAM,IAAI,KAAK,CAAC,2DAA2D,CAAC,CAAC;YAC9E,CAAC;YACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,YAAY,IAAI,2BAA2B,CAAC,CAAC;YACrE,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,sBAAsB,CAAC,KAAK,CAAC,CAAC;YACpD,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IACF,CAAC,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AACf,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAAkE,CAC1F,KAAsC,EACtC,OAA0B,EAC1B,OAA6B,EACC,EAAE;IAChC,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,CAAC;IAC/B,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,IAAI,GAAG;QACZ,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC;QACpD,UAAU,EAAE,OAAO,EAAE,UAAU;KACO,CAAC;IACxC,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvG,MAAM,eAAe,GAAG,gBAAgB,KAAK,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IAElF,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE;QAC7B,GAAG,IAAI;QACP,eAAe;KACuB,CAAC,CAAC;AAC1C,CAAC,CAAC;AAEF,SAAS,YAAY,CAAC,KAAsC,EAAE,MAAc,EAAE,OAAqC;IAClH,MAAM,OAAO,GAAG,EAAE,GAAG,kBAAkB,CAAC,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IAEtF,IAAI,OAAO,EAAE,OAAO,EAAE,CAAC;QACtB,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC;IACzC,CAAC;IAED,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAEnE,OAAO,IAAI,WAAW,CAAC;QACtB,MAAM;QACN,UAAU;QACV,uBAAuB,EAAE,IAAI;QAC7B,KAAK,EAAE,OAAO,EAAE,KAAK;QACrB,cAAc,EAAE,OAAO;QACvB,OAAO,EAAE,OAAO;KAChB,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CACnB,KAAsC,EACtC,OAA0B,EAC1B,OAAgD,EAChD,cAAsB,EACtB,0BAA0B,GAAgC,gCAAgC,CACzF,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,EAClC,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK,CACjD;IAED,MAAM,uBAAuB,GAAG,KAAK,CAAC,MAAM,EAAE,uBAAuB,IAAI,KAAK,CAAC;IAC/E,MAAM,kBAAkB,GAAG,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK,CAAC;IACrE,MAAM,eAAe,GAAG,sBAAsB,CAC7C,OAAO,CAAC,QAAQ,EAChB,uBAAuB,CAAC;QACvB,8BAA8B,EAAE,KAAK,CAAC,MAAM,EAAE,8BAA8B,IAAI,KAAK;QACrF,uBAAuB;QACvB,kBAAkB;KAClB,CAAC,CACF,CAAC;IACF,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,yBAAyB,EAAE;QACpF,0BAA0B;QAC1B,8BAA8B,EAAE,KAAK,CAAC,MAAM,EAAE,8BAA8B,IAAI,KAAK;QACrF,uBAAuB;QACvB,kBAAkB;QAClB,WAAW,EAAE;YACZ,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,IAAI;YAC5D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK;SAC7E;KACD,CAAC,CAAC;IAEH,MAAM,MAAM,GAAkC;QAC7C,KAAK,EAAE,cAAc;QACrB,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,gBAAgB,EACf,KAAK,CAAC,MAAM,EAAE,kBAAkB,KAAK,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,yBAAyB,CAAC,OAAO,EAAE,SAAS,CAAC;QAC5G,KAAK,EAAE,KAAK;KACZ,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,kCAAkC,CAAC,CAAC;IAC5F,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,eAAe,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7C,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,eAAe,CAAC,YAAY,EAAE;YAClE,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,IAAI;YAC5D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK;SAC7E,CAAC,CAAC;IACJ,CAAC;IACD,IAAI,OAAO,EAAE,UAAU,KAAK,SAAS,EAAE,CAAC;QACvC,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,UAAU,CAAC;IACzC,CAAC;IAED,MAAM,eAAe,GAAG,OAAO,EAAE,eAAe,IAAI,CAAC,OAAO,EAAE,gBAAgB,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;IACvG,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,eAAe,EAAE,CAAC;YACrB,MAAM,MAAM,GAAG,OAAO,EAAE,eAAe;gBACtC,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,eAAe,CAAC;gBAChF,CAAC,CAAC,eAAe,CAAC;YACnB,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,MAAwD;gBAChE,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,KAAK,CAAC,gBAAgB,EAAE,GAAG,KAAK,IAAI,EAAE,CAAC;YACjD,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,CAAC,KAAK,CAAC,gBAAgB,EAAE,GAAG,IAAI,MAAM,CAAmD;aACjG,CAAC;QACH,CAAC;IACF,CAAC;IAED,+EAA+E;IAC/E,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,eAAe,IAAI,KAAK,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;IACvG,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACvC,CAAC;IAED,OAAO,MAAM,CAAC;AACf,CAAC","sourcesContent":["import { AzureOpenAI } from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { clampThinkingLevel } from \"../models.ts\";\nimport type { Api, AssistantMessage, Model, SimpleStreamOptions, StreamFunction, TranscriptContext } from \"../types.ts\";\nimport { formatProviderError, normalizeProviderError } from \"../utils/error-body.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord } from \"../utils/headers.ts\";\nimport { getIdentityHeaders } from \"../utils/identity.ts\";\nimport { retryProviderRequest } from \"../utils/provider-retry.ts\";\nimport { getDeclaredTools, resolveTranscript, resolveTranscriptTools } from \"../utils/transcript.ts\";\nimport { type AzureEndpointOptions, resolveAzureConfig, resolveDeploymentName } from \"./azure-openai-config.ts\";\nimport { createGrammarToolInputProperties } from \"./constrained-sampling.ts\";\nimport { clampOpenAIPromptCacheKey } from \"./openai-prompt-cache.ts\";\nimport {\n\tconvertResponsesMessages,\n\tconvertResponsesTools,\n\tgetResponsesToolSupport,\n\tprocessResponsesStream,\n} from \"./openai-responses-shared.ts\";\nimport { buildBaseOptions, resolveSamplingParams } from \"./simple-options.ts\";\n\nconst AZURE_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\", \"azure\"]);\n// OpenAI Responses rejects max_output_tokens below 16: https://github.com/badlogic/pi-mono/issues/6265\nconst OPENAI_RESPONSES_MIN_OUTPUT_TOKENS = 16;\n\nfunction formatAzureOpenAIError(error: unknown): string {\n\treturn formatProviderError(normalizeProviderError(error), \"Azure OpenAI API error\");\n}\n\n// Azure OpenAI Responses-specific options\nexport interface AzureOpenAIResponsesOptions extends AzureEndpointOptions {\n\treasoningEffort?: \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\ttoolChoice?: ResponseCreateParamsStreaming[\"tool_choice\"];\n\treasoningSummary?: \"auto\" | \"detailed\" | \"concise\" | null;\n}\n\n/**\n * Generate function for Azure OpenAI Responses API\n */\nexport const stream: StreamFunction<\"azure-openai-responses\", AzureOpenAIResponsesOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions?: AzureOpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\tconst normalizedContext = resolveTranscript(context, model.compat?.supportsMidConvoSystemMessages);\n\n\t// Start async processing\n\t(async () => {\n\t\tconst deploymentName = resolveDeploymentName(model, options);\n\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"azure-openai-responses\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"pending\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create Azure OpenAI client\n\t\t\tconst apiKey = options?.apiKey;\n\t\t\tif (!apiKey) {\n\t\t\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t\t\t}\n\t\t\tconst client = createClient(model, apiKey, options);\n\t\t\tconst grammarToolInputProperties = createGrammarToolInputProperties(\n\t\t\t\tgetDeclaredTools(normalizedContext.messages),\n\t\t\t\tmodel.compat?.supportsOpenAIGrammarTools ?? false,\n\t\t\t);\n\t\t\tlet params = buildParams(model, normalizedContext, options, deploymentName, grammarToolInputProperties);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst requestOptions = {\n\t\t\t\t...(options?.signal ? { signal: options.signal } : {}),\n\t\t\t\t...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),\n\t\t\t\tmaxRetries: 0,\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await retryProviderRequest(\n\t\t\t\t() => client.responses.create(params, requestOptions).withResponse(),\n\t\t\t\t{\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t},\n\t\t\t);\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tonProviderStreamEvent: options?.onProviderStreamEvent,\n\t\t\t\tgrammarToolInputProperties,\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"pending\") {\n\t\t\t\tthrow new Error(\"Azure OpenAI Responses stream ended without a stop reason\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage || \"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = formatAzureOpenAIError(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimple: StreamFunction<\"azure-openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey;\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = {\n\t\t...buildBaseOptions(model, context, options, apiKey),\n\t\ttoolChoice: options?.toolChoice,\n\t} satisfies AzureOpenAIResponsesOptions;\n\tconst clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;\n\tconst reasoningEffort = clampedReasoning === \"off\" ? undefined : clampedReasoning;\n\n\treturn stream(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies AzureOpenAIResponsesOptions);\n};\n\nfunction createClient(model: Model<\"azure-openai-responses\">, apiKey: string, options?: AzureOpenAIResponsesOptions) {\n\tconst headers = { ...getIdentityHeaders(model, options?.identity), ...model.headers };\n\n\tif (options?.headers) {\n\t\tObject.assign(headers, options.headers);\n\t}\n\n\tconst { baseUrl, apiVersion } = resolveAzureConfig(model, options);\n\n\treturn new AzureOpenAI({\n\t\tapiKey,\n\t\tapiVersion,\n\t\tdangerouslyAllowBrowser: true,\n\t\tfetch: options?.fetch,\n\t\tdefaultHeaders: headers,\n\t\tbaseURL: baseUrl,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: TranscriptContext,\n\toptions: AzureOpenAIResponsesOptions | undefined,\n\tdeploymentName: string,\n\tgrammarToolInputProperties: ReadonlyMap<string, string> = createGrammarToolInputProperties(\n\t\tgetDeclaredTools(context.messages),\n\t\tmodel.compat?.supportsOpenAIGrammarTools ?? false,\n\t),\n) {\n\tconst supportsAdditionalTools = model.compat?.supportsAdditionalTools ?? false;\n\tconst supportsToolSearch = model.compat?.supportsToolSearch ?? false;\n\tconst transcriptTools = resolveTranscriptTools(\n\t\tcontext.messages,\n\t\tgetResponsesToolSupport({\n\t\t\tsupportsMidConvoSystemMessages: model.compat?.supportsMidConvoSystemMessages ?? false,\n\t\t\tsupportsAdditionalTools,\n\t\t\tsupportsToolSearch,\n\t\t}),\n\t);\n\tconst messages = convertResponsesMessages(model, context, AZURE_TOOL_CALL_PROVIDERS, {\n\t\tgrammarToolInputProperties,\n\t\tsupportsMidConvoSystemMessages: model.compat?.supportsMidConvoSystemMessages ?? false,\n\t\tsupportsAdditionalTools,\n\t\tsupportsToolSearch,\n\t\ttoolOptions: {\n\t\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? true,\n\t\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\t},\n\t});\n\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: deploymentName,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key:\n\t\t\tmodel.compat?.promptCacheKeyMode === \"disabled\" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = Math.max(options.maxTokens, OPENAI_RESPONSES_MIN_OUTPUT_TOKENS);\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (transcriptTools.requestTools.length > 0) {\n\t\tparams.tools = convertResponsesTools(transcriptTools.requestTools, {\n\t\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? true,\n\t\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\t});\n\t}\n\tif (options?.toolChoice !== undefined) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tconst reasoningEffort = options?.reasoningEffort ?? (options?.reasoningSummary ? \"medium\" : undefined);\n\tif (model.reasoning) {\n\t\tif (reasoningEffort) {\n\t\t\tconst effort = options?.reasoningEffort\n\t\t\t\t? (model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort)\n\t\t\t\t: reasoningEffort;\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.thinkingLevelMap?.off !== null) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: (model.thinkingLevelMap?.off ?? \"none\") as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t};\n\t\t}\n\t}\n\n\t// Last so model and request sampling parameters override named request fields.\n\tconst samplingParams = resolveSamplingParams(model, reasoningEffort ?? \"off\", options?.samplingParams);\n\tif (samplingParams) {\n\t\tObject.assign(params, samplingParams);\n\t}\n\n\treturn params;\n}\n"]}

package/dist/api/bedrock-converse-stream.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"bedrock-converse-stream.d.ts","sourceRoot":"","sources":["../../src/api/bedrock-converse-stream.ts"],"names":[],"mappings":"AA6BA,OAAO,KAAK,EAUX,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAEb,eAAe,EAEf,aAAa,EAIb,MAAM,aAAa,CAAC;AA2BrB,MAAM,MAAM,sBAAsB,GAAG,YAAY,GAAG,SAAS,CAAC;AAE9D,MAAM,WAAW,cAAe,SAAQ,aAAa;IACpD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IAEtE,SAAS,CAAC,EAAE,aAAa,CAAC;IAE1B,eAAe,CAAC,EAAE,eAAe,CAAC;IAElC,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;;;;;;;OASG;IACH,eAAe,CAAC,EAAE,sBAAsB,CAAC;IACzC;;;sGAGkG;IAClG,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACzC;;;;yGAIqG;IACrG,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAgBD,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,yBAAyB,EAAE,cAAc,CA6O5E,CAAC;AAkLF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,yBAAyB,EAAE,mBAAmB,CAiDvF,CAAC"}
+{"version":3,"file":"bedrock-converse-stream.d.ts","sourceRoot":"","sources":["../../src/api/bedrock-converse-stream.ts"],"names":[],"mappings":"AA8BA,OAAO,KAAK,EAUX,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAEb,eAAe,EACf,aAAa,EAGb,MAAM,aAAa,CAAC;AAgCrB,MAAM,MAAM,sBAAsB,GAAG,YAAY,GAAG,SAAS,CAAC;AAE9D,MAAM,WAAW,cAAe,SAAQ,aAAa;IACpD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IAEtE,SAAS,CAAC,EAAE,aAAa,CAAC;IAE1B,eAAe,CAAC,EAAE,eAAe,CAAC;IAElC,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;;;;;;;OASG;IACH,eAAe,CAAC,EAAE,sBAAsB,CAAC;IACzC;;;sGAGkG;IAClG,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACzC;;;;yGAIqG;IACrG,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAsBD,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,yBAAyB,EAAE,cAAc,CAqQ5E,CAAC;AAkLF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,yBAAyB,EAAE,mBAAmB,CAiDvF,CAAC"}

package/dist/api/bedrock-converse-stream.js

@@ -8,13 +8,13 @@ import { normalizeProviderError } from "../utils/error-body.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { providerHeadersToRecord } from "../utils/headers.js";
 import { getClientVersion } from "../utils/identity.js";
-import { parseStreamingJson, parseToolCallArguments } from "../utils/json-parse.js";
 import { resolveHttpProxyUrlForTarget } from "../utils/node-http-proxy.js";
 import { getProviderEnvValue } from "../utils/provider-env.js";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
+import { ToolCallArgumentsStream } from "../utils/streaming-json.js";
 import { getSystemMessageText } from "../utils/text.js";
 import { collapseSystemMessages, getCurrentTools, getInitialSystemMessage, withoutInitialSystemMessage, } from "../utils/transcript.js";
-import { getJsonSchemaToolParameters, resolveJsonSchemaStrictSampling } from "./constrained-sampling.js";
+import { getJsonSchemaToolParameters, isStrictToolLimitErrorMessage, prefersJsonSchemaStrictSampling, resolveJsonSchemaStrictSampling, } from "./constrained-sampling.js";
 import { adjustMaxTokensForThinking, buildBaseOptions, clampMaxTokensToContext, clampReasoning, } from "./simple-options.js";
 import { splitErrorToolResultImages, transformMessages } from "./transform-messages.js";
 const EMPTY_TEXT_PLACEHOLDER = "<empty>";
@@ -43,7 +43,7 @@ export const stream = (model, context, options = {}) => {
             stopReason: "pending",
             timestamp: Date.now(),
         };
-        const blocks = output.content;
+        const openBlocks = new Map();
         // A profile explicitly configured through pi's auth flow (the `profile`
         // option or scoped `AWS_PROFILE` on the stored credential's env) must win
         // over ambient AWS_ACCESS_KEY_ID/AWS_SECRET_ACCESS_KEY. The SDK default
@@ -145,24 +145,49 @@ export const stream = (model, context, options = {}) => {
             const inferenceMaxTokens = options.maxTokens ?? (isAnthropicClaudeModel(model) ? model.maxTokens : undefined);
             const initialSystemMessage = getInitialSystemMessage(normalizedContext.messages);
             const initialSystemPrompt = initialSystemMessage ? getSystemMessageText(initialSystemMessage) : undefined;
-            let commandInput = {
-                modelId: model.id,
-                messages: convertMessages(normalizedContext, model, cacheRetention, options.env),
-                system: buildSystemPrompt(initialSystemPrompt, model, cacheRetention, options.env),
-                inferenceConfig: {
-                    ...(inferenceMaxTokens !== undefined && { maxTokens: inferenceMaxTokens }),
-                    ...(options.temperature !== undefined && { temperature: options.temperature }),
-                },
-                toolConfig: convertToolConfig(getCurrentTools(normalizedContext.messages), options.toolChoice, supportsStrictMode),
-                additionalModelRequestFields: buildAdditionalModelRequestFields(model, options),
-                ...(options.requestMetadata !== undefined && { requestMetadata: options.requestMetadata }),
-            };
-            const nextCommandInput = await options?.onPayload?.(commandInput, model);
-            if (nextCommandInput !== undefined) {
-                commandInput = nextCommandInput;
+            let disablePreferredStrictTools = false;
+            let response;
+            for (;;) {
+                const tools = convertToolConfig(getCurrentTools(normalizedContext.messages), options.toolChoice, supportsStrictMode, disablePreferredStrictTools);
+                let commandInput = {
+                    modelId: model.id,
+                    messages: convertMessages(normalizedContext, model, cacheRetention, options.env),
+                    system: buildSystemPrompt(initialSystemPrompt, model, cacheRetention, options.env),
+                    inferenceConfig: {
+                        ...(inferenceMaxTokens !== undefined && { maxTokens: inferenceMaxTokens }),
+                        ...(options.temperature !== undefined && { temperature: options.temperature }),
+                    },
+                    toolConfig: tools.toolConfig,
+                    additionalModelRequestFields: buildAdditionalModelRequestFields(model, options),
+                    ...(options.requestMetadata !== undefined && { requestMetadata: options.requestMetadata }),
+                };
+                const nextCommandInput = await options?.onPayload?.(commandInput, model);
+                if (nextCommandInput !== undefined) {
+                    commandInput = nextCommandInput;
+                }
+                try {
+                    response = await client.send(new ConverseStreamCommand(commandInput), { abortSignal: options.signal });
+                    break;
+                }
+                catch (error) {
+                    // Bedrock checks the tool schemas before it starts the stream, so a retry never repeats streamed output.
+                    if (options.signal?.aborted ||
+                        disablePreferredStrictTools ||
+                        !tools.hasPreferredStrictTools ||
+                        !(error instanceof BedrockRuntimeServiceException) ||
+                        error.name !== "ValidationException" ||
+                        !isStrictToolLimitErrorMessage(error.message)) {
+                        throw error;
+                    }
+                    disablePreferredStrictTools = true;
+                    appendAssistantMessageDiagnostic(output, {
+                        type: "bedrock_strict_tools_fallback",
+                        timestamp: Date.now(),
+                        error: { message: error.message },
+                        details: { reason: "strict_tool_limits_exceeded" },
+                    });
+                }
             }
-            const command = new ConverseStreamCommand(commandInput);
-            const response = await client.send(command, { abortSignal: options.signal });
             responseRequestId = normalizeDiagnosticValue(response.$metadata.requestId);
             if (!observedRawResponse && response.$metadata.httpStatusCode !== undefined) {
                 const responseHeaders = {};
@@ -180,13 +205,13 @@ export const stream = (model, context, options = {}) => {
                     stream.push({ type: "start", partial: output });
                 }
                 else if (item.contentBlockStart) {
-                    handleContentBlockStart(item.contentBlockStart, blocks, output, stream);
+                    handleContentBlockStart(item.contentBlockStart, openBlocks, output, stream);
                 }
                 else if (item.contentBlockDelta) {
-                    handleContentBlockDelta(item.contentBlockDelta, blocks, output, stream);
+                    handleContentBlockDelta(item.contentBlockDelta, openBlocks, output, stream);
                 }
                 else if (item.contentBlockStop) {
-                    handleContentBlockStop(item.contentBlockStop, blocks, output, stream);
+                    handleContentBlockStop(item.contentBlockStop, openBlocks, output, stream);
                 }
                 else if (item.messageStop) {
                     output.rawStopReason = item.messageStop.stopReason;
@@ -225,15 +250,12 @@ export const stream = (model, context, options = {}) => {
                 throw new Error(output.errorMessage || "An unknown error occurred");
             }
             // A stream can settle without stopping every block, so finalize here too.
-            for (const block of output.content)
-                finalizeStreamingBlock(block);
+            flushOpenBlocks(openBlocks, output);
             stream.push({ type: "done", reason: output.stopReason, message: output });
             stream.end();
         }
         catch (error) {
-            for (const block of output.content) {
-                finalizeStreamingBlock(block);
-            }
+            flushOpenBlocks(openBlocks, output);
             output.stopReason = options.signal?.aborted ? "aborted" : "error";
             output.errorMessage = formatBedrockError(error);
             if (output.stopReason === "error") {
@@ -436,59 +458,60 @@ export const streamSimple = (model, context, options) => {
         thinkingBudgets: options.thinkingBudgets,
     });
 };
-function handleContentBlockStart(event, blocks, output, stream) {
-    const index = event.contentBlockIndex;
+function handleContentBlockStart(event, openBlocks, output, stream) {
     const start = event.start;
     if (start?.toolUse) {
-        const block = {
+        output.content.push({
             type: "toolCall",
             id: start.toolUse.toolUseId || "",
             name: start.toolUse.name || "",
             arguments: {},
-            partialJson: "",
-            index,
-        };
-        output.content.push(block);
-        stream.push({ type: "toolcall_start", contentIndex: blocks.length - 1, partial: output });
+        });
+        const contentIndex = output.content.length - 1;
+        openBlocks.set(event.contentBlockIndex, { contentIndex, arguments: new ToolCallArgumentsStream() });
+        stream.push({ type: "toolcall_start", contentIndex, partial: output });
     }
 }
-function handleContentBlockDelta(event, blocks, output, stream) {
+function handleContentBlockDelta(event, openBlocks, output, stream) {
     const contentBlockIndex = event.contentBlockIndex;
     const delta = event.delta;
-    let index = blocks.findIndex((b) => b.index === contentBlockIndex);
-    let block = blocks[index];
+    let open = openBlocks.get(contentBlockIndex);
+    let block = open === undefined ? undefined : output.content[open.contentIndex];
     if (delta?.text !== undefined) {
         // If no text block exists yet, create one, as `handleContentBlockStart` is not sent for text blocks
-        if (!block) {
-            const newBlock = { type: "text", text: "", index: contentBlockIndex };
-            output.content.push(newBlock);
-            index = blocks.length - 1;
-            block = blocks[index];
-            stream.push({ type: "text_start", contentIndex: index, partial: output });
+        if (!open) {
+            output.content.push({ type: "text", text: "" });
+            open = { contentIndex: output.content.length - 1 };
+            openBlocks.set(contentBlockIndex, open);
+            block = output.content[open.contentIndex];
+            stream.push({ type: "text_start", contentIndex: open.contentIndex, partial: output });
         }
-        if (block.type === "text") {
+        if (block?.type === "text") {
             block.text += delta.text;
-            stream.push({ type: "text_delta", contentIndex: index, delta: delta.text, partial: output });
+            stream.push({ type: "text_delta", contentIndex: open.contentIndex, delta: delta.text, partial: output });
         }
     }
-    else if (delta?.toolUse && block?.type === "toolCall") {
-        block.partialJson = (block.partialJson || "") + (delta.toolUse.input || "");
-        block.arguments = parseStreamingJson(block.partialJson);
-        stream.push({ type: "toolcall_delta", contentIndex: index, delta: delta.toolUse.input || "", partial: output });
+    else if (delta?.toolUse && block?.type === "toolCall" && open?.arguments !== undefined) {
+        block.arguments = open.arguments.append(delta.toolUse.input || "");
+        stream.push({
+            type: "toolcall_delta",
+            contentIndex: open.contentIndex,
+            delta: delta.toolUse.input || "",
+            partial: output,
+        });
     }
     else if (delta?.reasoningContent) {
-        let thinkingBlock = block;
-        let thinkingIndex = index;
-        if (!thinkingBlock) {
-            const newBlock = { type: "thinking", thinking: "", thinkingSignature: "", index: contentBlockIndex };
-            output.content.push(newBlock);
-            thinkingIndex = blocks.length - 1;
-            thinkingBlock = blocks[thinkingIndex];
-            stream.push({ type: "thinking_start", contentIndex: thinkingIndex, partial: output });
+        if (!open) {
+            output.content.push({ type: "thinking", thinking: "", thinkingSignature: "" });
+            open = { contentIndex: output.content.length - 1 };
+            openBlocks.set(contentBlockIndex, open);
+            block = output.content[open.contentIndex];
+            stream.push({ type: "thinking_start", contentIndex: open.contentIndex, partial: output });
         }
-        if (thinkingBlock?.type === "thinking") {
+        if (block?.type === "thinking") {
+            const thinkingIndex = open.contentIndex;
             if (delta.reasoningContent.text) {
-                thinkingBlock.thinking += delta.reasoningContent.text;
+                block.thinking += delta.reasoningContent.text;
                 stream.push({
                     type: "thinking_delta",
                     contentIndex: thinkingIndex,
@@ -498,18 +521,17 @@ function handleContentBlockDelta(event, blocks, output, stream) {
             }
             // `thinkingSignature` holds either an Anthropic signature or an opaque redacted
             // payload, never both: mixing them would corrupt whichever arrived first.
-            if (delta.reasoningContent.signature && !thinkingBlock.redacted) {
-                thinkingBlock.thinkingSignature =
-                    (thinkingBlock.thinkingSignature || "") + delta.reasoningContent.signature;
+            if (delta.reasoningContent.signature && !block.redacted) {
+                block.thinkingSignature = (block.thinkingSignature || "") + delta.reasoningContent.signature;
             }
             if (delta.reasoningContent.redactedContent?.length) {
                 // Encrypted reasoning from non-Anthropic models on Bedrock (e.g. OpenAI GPT-5.6).
                 // The payload is opaque, so keep it verbatim in `thinkingSignature` the way the
                 // Anthropic path stores redacted thinking, and replay it on the next turn.
-                if (!thinkingBlock.redacted) {
-                    thinkingBlock.redacted = true;
-                    thinkingBlock.thinkingSignature = "";
-                    thinkingBlock.thinking += REDACTED_THINKING_PLACEHOLDER;
+                if (!block.redacted) {
+                    block.redacted = true;
+                    block.thinkingSignature = "";
+                    block.thinking += REDACTED_THINKING_PLACEHOLDER;
                     stream.push({
                         type: "thinking_delta",
                         contentIndex: thinkingIndex,
@@ -517,32 +539,29 @@ function handleContentBlockDelta(event, blocks, output, stream) {
                         partial: output,
                     });
                 }
-                thinkingBlock.redactedChunks ??= [];
-                thinkingBlock.redactedChunks.push(delta.reasoningContent.redactedContent);
+                open.redactedChunks ??= [];
+                open.redactedChunks.push(delta.reasoningContent.redactedContent);
             }
         }
     }
 }
 /**
- * Encodes buffered encrypted reasoning into `thinkingSignature` and drops the scratch
- * buffer, which must never reach a persisted message: `Uint8Array` serializes to an
- * index-keyed object roughly ten times the size of the base64 payload.
+ * Encodes buffered encrypted reasoning into `thinkingSignature`. The chunks stay in the stream state, never on the
+ * block: `Uint8Array` is not JSON and serializes to an index-keyed object roughly ten times the size of the base64
+ * payload.
  */
-function flushRedactedContent(block) {
-    if (block.type !== "thinking" || !block.redactedChunks)
+function flushRedactedContent(open, output) {
+    const block = output.content[open.contentIndex];
+    if (block?.type !== "thinking" || !open.redactedChunks)
         return;
-    block.thinkingSignature = bytesToBase64(block.redactedChunks);
-    delete block.redactedChunks;
+    block.thinkingSignature = bytesToBase64(open.redactedChunks);
+    open.redactedChunks = undefined;
 }
-/**
- * Strips every streaming scratch field. Runs from the terminal paths as well as
- * `contentBlockStop`, because a stream can settle without stopping each block.
- */
-function finalizeStreamingBlock(block) {
-    delete block.index;
-    // partialJson is only a streaming scratch buffer; never persist it.
-    delete block.partialJson;
-    flushRedactedContent(block);
+/** Ends the stream state of blocks a stream settled without stopping. */
+function flushOpenBlocks(openBlocks, output) {
+    for (const open of openBlocks.values())
+        flushRedactedContent(open, output);
+    openBlocks.clear();
 }
 function handleMetadata(event, model, output) {
     if (event.usage) {
@@ -555,25 +574,24 @@ function handleMetadata(event, model, output) {
         calculateCost(model, output.usage);
     }
 }
-function handleContentBlockStop(event, blocks, output, stream) {
-    const index = blocks.findIndex((b) => b.index === event.contentBlockIndex);
-    const block = blocks[index];
-    if (!block)
+function handleContentBlockStop(event, openBlocks, output, stream) {
+    const open = openBlocks.get(event.contentBlockIndex);
+    if (!open)
         return;
-    delete block.index;
+    openBlocks.delete(event.contentBlockIndex);
+    const index = open.contentIndex;
+    const block = output.content[index];
     switch (block.type) {
         case "text":
             stream.push({ type: "text_end", contentIndex: index, content: block.text, partial: output });
             break;
         case "thinking":
-            flushRedactedContent(block);
+            flushRedactedContent(open, output);
             stream.push({ type: "thinking_end", contentIndex: index, content: block.thinking, partial: output });
             break;
         case "toolCall":
-            Object.assign(block, parseToolCallArguments(block.partialJson));
-            // Finalize in-place and strip the scratch buffer so replay only
-            // carries parsed arguments.
-            delete block.partialJson;
+            if (open.arguments !== undefined)
+                Object.assign(block, open.arguments.finish());
             stream.push({ type: "toolcall_end", contentIndex: index, toolCall: block, partial: output });
             break;
     }
@@ -924,13 +942,17 @@ function convertMessages(context, model, cacheRetention, env) {
     }
     return result;
 }
-function convertToolConfig(tools, toolChoice, supportsStrictMode) {
-    if (!tools?.length)
-        return undefined;
-    if (toolChoice === "none")
-        return undefined;
+function convertToolConfig(tools, toolChoice, supportsStrictMode, disablePreferredStrictTools) {
+    if (!tools?.length || toolChoice === "none")
+        return { toolConfig: undefined, hasPreferredStrictTools: false };
+    let hasPreferredStrictTools = false;
     const bedrockTools = tools.map((tool) => {
-        const strict = resolveJsonSchemaStrictSampling(tool, supportsStrictMode);
+        const preferred = prefersJsonSchemaStrictSampling(tool);
+        const strict = disablePreferredStrictTools && preferred
+            ? undefined
+            : resolveJsonSchemaStrictSampling(tool, supportsStrictMode);
+        if (strict === true && preferred)
+            hasPreferredStrictTools = true;
         return {
             toolSpec: {
                 name: tool.name,
@@ -953,7 +975,7 @@ function convertToolConfig(tools, toolChoice, supportsStrictMode) {
                 bedrockToolChoice = { tool: { name: toolChoice.name } };
             }
     }
-    return { tools: bedrockTools, toolChoice: bedrockToolChoice };
+    return { toolConfig: { tools: bedrockTools, toolChoice: bedrockToolChoice }, hasPreferredStrictTools };
 }
 function mapStopReason(reason) {
     switch (reason) {