@librechat/agents 3.4.6 → 3.5.0

package/dist/cjs/llm/openrouter/index.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.cjs","names":["ChatOpenAI","emitStreamChunkCallback"],"sources":["../../../../src/llm/openrouter/index.ts"],"sourcesContent":["import type {\n  ChatOpenAICallOptions,\n  OpenAIChatInput,\n  OpenAIClient,\n} from '@langchain/openai';\nimport type { CallbackManagerForLLMRun } from '@langchain/core/callbacks/manager';\nimport type { ChatGenerationChunk } from '@langchain/core/outputs';\nimport type { BaseMessage } from '@langchain/core/messages';\nimport type { SeenScalarMetadata } from '@/llm/openai/streamMetadata';\nimport type { PromptCacheTtl } from '@/messages/cache';\nimport { dropRepeatedScalarMetadata } from '@/llm/openai/streamMetadata';\nimport { ChatOpenAI, emitStreamChunkCallback } from '@/llm/openai';\n\nexport type OpenRouterReasoningEffort =\n  | 'xhigh'\n  | 'high'\n  | 'medium'\n  | 'low'\n  | 'minimal'\n  | 'none';\n\nexport interface OpenRouterReasoning {\n  effort?: OpenRouterReasoningEffort;\n  max_tokens?: number;\n  exclude?: boolean;\n  enabled?: boolean;\n}\n\nexport interface ChatOpenRouterCallOptions\n  extends Omit<ChatOpenAICallOptions, 'reasoning'> {\n  /** @deprecated Use `reasoning` object instead */\n  include_reasoning?: boolean;\n  reasoning?: OpenRouterReasoning;\n  modelKwargs?: OpenAIChatInput['modelKwargs'];\n  useResponsesApi?: boolean;\n  promptCache?: boolean;\n  /**\n   * Prompt-cache breakpoint TTL. Defaults to `'1h'` (extended cache) when\n   * `promptCache` is enabled; set `'5m'` for the legacy 5-minute behavior.\n   * OpenRouter forwards this to Claude upstreams (Anthropic / Bedrock / Vertex),\n   * which downgrade to 5m where the extended TTL isn't supported.\n   */\n  promptCacheTtl?: PromptCacheTtl;\n}\n\nexport type ChatOpenRouterInput = Partial<\n  ChatOpenRouterCallOptions & OpenAIChatInput\n> & {\n  /** Minimum delay in ms between visible streamed deltas (default 25; 0 disables). */\n  _lc_stream_delay?: number;\n};\n\n/** invocationParams return type extended with OpenRouter reasoning */\nexport type OpenRouterInvocationParams = Omit<\n  OpenAIClient.Chat.ChatCompletionCreateParams,\n  'messages'\n> & {\n  reasoning?: OpenRouterReasoning;\n  cache_control?: { type: 'ephemeral'; ttl?: '1h' };\n};\n\ntype InvocationParamsExtra = {\n  streaming?: boolean;\n};\n\ninterface OpenRouterReasoningTextDetail {\n  type: 'reasoning.text';\n  text?: string;\n  format?: string;\n  index?: number;\n}\n\ninterface OpenRouterReasoningEncryptedDetail {\n  type: 'reasoning.encrypted';\n  id?: string;\n  data?: string;\n  format?: string;\n  index?: number;\n}\n\ntype OpenRouterReasoningDetail =\n  | OpenRouterReasoningTextDetail\n  | OpenRouterReasoningEncryptedDetail;\n\nfunction isReasoningTextDetail(\n  value: unknown\n): value is OpenRouterReasoningTextDetail {\n  return (\n    typeof value === 'object' &&\n    value !== null &&\n    'type' in value &&\n    value.type === 'reasoning.text'\n  );\n}\n\nfunction isReasoningEncryptedDetail(\n  value: unknown\n): value is OpenRouterReasoningEncryptedDetail {\n  return (\n    typeof value === 'object' &&\n    value !== null &&\n    'type' in value &&\n    value.type === 'reasoning.encrypted'\n  );\n}\n\nfunction getReasoningDetails(value: unknown): OpenRouterReasoningDetail[] {\n  if (!Array.isArray(value)) {\n    return [];\n  }\n  return value.filter(\n    (detail): detail is OpenRouterReasoningDetail =>\n      isReasoningTextDetail(detail) || isReasoningEncryptedDetail(detail)\n  );\n}\n\nexport class ChatOpenRouter extends ChatOpenAI {\n  private openRouterReasoning?: OpenRouterReasoning;\n  /** @deprecated Use `reasoning` object instead */\n  private includeReasoning?: boolean;\n\n  constructor(_fields: ChatOpenRouterInput) {\n    const fieldsWithoutPromptCache: ChatOpenRouterInput = { ..._fields };\n    delete fieldsWithoutPromptCache.promptCache;\n    delete fieldsWithoutPromptCache.promptCacheTtl;\n\n    const {\n      include_reasoning,\n      reasoning: openRouterReasoning,\n      modelKwargs = {},\n      ...fields\n    } = fieldsWithoutPromptCache;\n\n    // Extract reasoning from modelKwargs if provided there (e.g., from LLMConfig)\n    const { reasoning: mkReasoning, ...restModelKwargs } = modelKwargs as {\n      reasoning?: OpenRouterReasoning;\n    } & Record<string, unknown>;\n    const mergedReasoning =\n      mkReasoning != null || openRouterReasoning != null\n        ? {\n          ...mkReasoning,\n          ...openRouterReasoning,\n        }\n        : undefined;\n    const runtimeReasoning =\n      mergedReasoning ??\n      (include_reasoning === true ? { enabled: true } : undefined);\n    const parentModelKwargs =\n      runtimeReasoning == null\n        ? restModelKwargs\n        : { ...restModelKwargs, reasoning: runtimeReasoning };\n\n    super({\n      ...fields,\n      modelKwargs: parentModelKwargs,\n      includeReasoningDetails: true,\n      convertReasoningDetailsToContent: true,\n      preserveToolCacheControl: true,\n      responsesPromptCache: _fields.promptCache,\n      responsesPromptCacheTtl: _fields.promptCacheTtl,\n    });\n\n    // Merge reasoning config: modelKwargs.reasoning < constructor reasoning\n    if (mergedReasoning != null) {\n      this.openRouterReasoning = mergedReasoning;\n    }\n\n    this.includeReasoning = include_reasoning;\n  }\n  static lc_name(): 'LibreChatOpenRouter' {\n    return 'LibreChatOpenRouter';\n  }\n\n  // OpenRouter widens OpenAI reasoning with extra effort levels ('xhigh' | 'none' | 'minimal').\n  override invocationParams(\n    options?: this['ParsedCallOptions'],\n    extra?: InvocationParamsExtra\n  ): OpenRouterInvocationParams {\n    type MutableParams = Omit<\n      OpenAIClient.Chat.ChatCompletionCreateParams,\n      'messages'\n    > & {\n      reasoning_effort?: string;\n      reasoning?: OpenRouterReasoning;\n      cache_control?: { type: 'ephemeral'; ttl?: '1h' };\n    };\n\n    const optionsWithDefaults = this._combineCallOptions(options);\n    const useResponsesApi = this._useResponsesApi(options);\n    const params = (\n      useResponsesApi\n        ? this.responses.invocationParams(optionsWithDefaults)\n        : this.completions.invocationParams(optionsWithDefaults, extra)\n    ) as MutableParams;\n\n    // Remove the OpenAI-native reasoning_effort that the parent sets;\n    // OpenRouter uses a `reasoning` object instead\n    delete params.reasoning_effort;\n\n    // Build the OpenRouter reasoning config\n    const reasoning = this.buildOpenRouterReasoning(optionsWithDefaults);\n    if (reasoning != null) {\n      params.reasoning = reasoning;\n    } else {\n      delete params.reasoning;\n    }\n\n    return params;\n  }\n\n  private buildOpenRouterReasoning(\n    options?: this['ParsedCallOptions']\n  ): OpenRouterReasoning | undefined {\n    let reasoning: OpenRouterReasoning | undefined;\n\n    // 1. Instance-level reasoning config (from constructor)\n    if (this.openRouterReasoning != null) {\n      reasoning = { ...this.openRouterReasoning };\n    }\n\n    // 2. LangChain-style reasoning params (from parent's `this.reasoning`)\n    const lcReasoning = this.getReasoningParams(options);\n    if (lcReasoning?.effort != null) {\n      reasoning = {\n        ...reasoning,\n        effort: lcReasoning.effort as OpenRouterReasoningEffort,\n      };\n    }\n\n    // 3. Call-level reasoning override\n    const callReasoning = (options as ChatOpenRouterCallOptions | undefined)\n      ?.reasoning;\n    if (callReasoning != null) {\n      reasoning = { ...reasoning, ...callReasoning };\n    }\n\n    // 4. Legacy include_reasoning backward compatibility\n    if (reasoning == null && this.includeReasoning === true) {\n      reasoning = { enabled: true };\n    }\n\n    return reasoning;\n  }\n\n  override async *_streamResponseChunks(\n    messages: BaseMessage[],\n    options: this['ParsedCallOptions'],\n    runManager?: CallbackManagerForLLMRun\n  ): AsyncGenerator<ChatGenerationChunk> {\n    const reasoningTextByIndex = new Map<\n      number,\n      OpenRouterReasoningTextDetail\n    >();\n    const reasoningEncryptedById = new Map<\n      string,\n      OpenRouterReasoningEncryptedDetail\n    >();\n    // Consume the raw (un-deduped) stream so `finish_reason` survives as the\n    // reasoning-flush signal below; de-duplicate after that, before emitting.\n    const seenScalarMetadata: SeenScalarMetadata = new Map();\n\n    for await (const generationChunk of super._streamRawResponseChunks(\n      messages,\n      options,\n      undefined\n    )) {\n      let currentReasoningText = '';\n      const reasoningDetails = getReasoningDetails(\n        generationChunk.message.additional_kwargs.reasoning_details\n      );\n\n      for (const detail of reasoningDetails) {\n        if (detail.type === 'reasoning.text') {\n          currentReasoningText += detail.text ?? '';\n          const index = detail.index ?? 0;\n          const existing = reasoningTextByIndex.get(index);\n          if (existing != null) {\n            existing.text = `${existing.text ?? ''}${detail.text ?? ''}`;\n            continue;\n          }\n          reasoningTextByIndex.set(index, {\n            ...detail,\n            text: detail.text ?? '',\n          });\n          continue;\n        }\n        if (detail.id != null) {\n          reasoningEncryptedById.set(detail.id, { ...detail });\n        }\n      }\n\n      if (\n        currentReasoningText.length > 0 &&\n        generationChunk.message.additional_kwargs.reasoning == null\n      ) {\n        generationChunk.message.additional_kwargs.reasoning =\n          currentReasoningText;\n      }\n\n      if (generationChunk.generationInfo?.finish_reason != null) {\n        const finalReasoningDetails = [\n          ...reasoningTextByIndex.values(),\n          ...reasoningEncryptedById.values(),\n        ];\n        if (finalReasoningDetails.length > 0) {\n          generationChunk.message.additional_kwargs.reasoning_details =\n            finalReasoningDetails;\n        } else {\n          delete generationChunk.message.additional_kwargs.reasoning_details;\n        }\n      } else {\n        delete generationChunk.message.additional_kwargs.reasoning_details;\n      }\n\n      dropRepeatedScalarMetadata(generationChunk, seenScalarMetadata);\n      await emitStreamChunkCallback(generationChunk, runManager);\n      yield generationChunk;\n    }\n  }\n}\n"],"mappings":";;;AAoFA,SAAS,sBACP,OACwC;CACxC,OACE,OAAO,UAAU,YACjB,UAAU,QACV,UAAU,SACV,MAAM,SAAS;AAEnB;AAEA,SAAS,2BACP,OAC6C;CAC7C,OACE,OAAO,UAAU,YACjB,UAAU,QACV,UAAU,SACV,MAAM,SAAS;AAEnB;AAEA,SAAS,oBAAoB,OAA6C;CACxE,IAAI,CAAC,MAAM,QAAQ,KAAK,GACtB,OAAO,CAAC;CAEV,OAAO,MAAM,QACV,WACC,sBAAsB,MAAM,KAAK,2BAA2B,MAAM,CACtE;AACF;AAEA,IAAa,iBAAb,cAAoCA,cAAAA,WAAW;CAC7C;;CAEA;CAEA,YAAY,SAA8B;EACxC,MAAM,2BAAgD,EAAE,GAAG,QAAQ;EACnE,OAAO,yBAAyB;EAChC,OAAO,yBAAyB;EAEhC,MAAM,EACJ,mBACA,WAAW,qBACX,cAAc,CAAC,GACf,GAAG,WACD;EAGJ,MAAM,EAAE,WAAW,aAAa,GAAG,oBAAoB;EAGvD,MAAM,kBACJ,eAAe,QAAQ,uBAAuB,OAC1C;GACA,GAAG;GACH,GAAG;EACL,IACE,KAAA;EACN,MAAM,mBACJ,oBACC,sBAAsB,OAAO,EAAE,SAAS,KAAK,IAAI,KAAA;EACpD,MAAM,oBACJ,oBAAoB,OAChB,kBACA;GAAE,GAAG;GAAiB,WAAW;EAAiB;EAExD,MAAM;GACJ,GAAG;GACH,aAAa;GACb,yBAAyB;GACzB,kCAAkC;GAClC,0BAA0B;GAC1B,sBAAsB,QAAQ;GAC9B,yBAAyB,QAAQ;EACnC,CAAC;EAGD,IAAI,mBAAmB,MACrB,KAAK,sBAAsB;EAG7B,KAAK,mBAAmB;CAC1B;CACA,OAAO,UAAiC;EACtC,OAAO;CACT;CAGA,iBACE,SACA,OAC4B;EAU5B,MAAM,sBAAsB,KAAK,oBAAoB,OAAO;EAE5D,MAAM,SADkB,KAAK,iBAAiB,OAE9B,IACV,KAAK,UAAU,iBAAiB,mBAAmB,IACnD,KAAK,YAAY,iBAAiB,qBAAqB,KAAK;EAKlE,OAAO,OAAO;EAGd,MAAM,YAAY,KAAK,yBAAyB,mBAAmB;EACnE,IAAI,aAAa,MACf,OAAO,YAAY;OAEnB,OAAO,OAAO;EAGhB,OAAO;CACT;CAEA,yBACE,SACiC;EACjC,IAAI;EAGJ,IAAI,KAAK,uBAAuB,MAC9B,YAAY,EAAE,GAAG,KAAK,oBAAoB;EAI5C,MAAM,cAAc,KAAK,mBAAmB,OAAO;EACnD,IAAI,aAAa,UAAU,MACzB,YAAY;GACV,GAAG;GACH,QAAQ,YAAY;EACtB;EAIF,MAAM,gBAAiB,SACnB;EACJ,IAAI,iBAAiB,MACnB,YAAY;GAAE,GAAG;GAAW,GAAG;EAAc;EAI/C,IAAI,aAAa,QAAQ,KAAK,qBAAqB,MACjD,YAAY,EAAE,SAAS,KAAK;EAG9B,OAAO;CACT;CAEA,OAAgB,sBACd,UACA,SACA,YACqC;EACrC,MAAM,uCAAuB,IAAI,IAG/B;EACF,MAAM,yCAAyB,IAAI,IAGjC;EAGF,MAAM,qCAAyC,IAAI,IAAI;EAEvD,WAAW,MAAM,mBAAmB,MAAM,yBACxC,UACA,SACA,KAAA,CACF,GAAG;GACD,IAAI,uBAAuB;GAC3B,MAAM,mBAAmB,oBACvB,gBAAgB,QAAQ,kBAAkB,iBAC5C;GAEA,KAAK,MAAM,UAAU,kBAAkB;IACrC,IAAI,OAAO,SAAS,kBAAkB;KACpC,wBAAwB,OAAO,QAAQ;KACvC,MAAM,QAAQ,OAAO,SAAS;KAC9B,MAAM,WAAW,qBAAqB,IAAI,KAAK;KAC/C,IAAI,YAAY,MAAM;MACpB,SAAS,OAAO,GAAG,SAAS,QAAQ,KAAK,OAAO,QAAQ;MACxD;KACF;KACA,qBAAqB,IAAI,OAAO;MAC9B,GAAG;MACH,MAAM,OAAO,QAAQ;KACvB,CAAC;KACD;IACF;IACA,IAAI,OAAO,MAAM,MACf,uBAAuB,IAAI,OAAO,IAAI,EAAE,GAAG,OAAO,CAAC;GAEvD;GAEA,IACE,qBAAqB,SAAS,KAC9B,gBAAgB,QAAQ,kBAAkB,aAAa,MAEvD,gBAAgB,QAAQ,kBAAkB,YACxC;GAGJ,IAAI,gBAAgB,gBAAgB,iBAAiB,MAAM;IACzD,MAAM,wBAAwB,CAC5B,GAAG,qBAAqB,OAAO,GAC/B,GAAG,uBAAuB,OAAO,CACnC;IACA,IAAI,sBAAsB,SAAS,GACjC,gBAAgB,QAAQ,kBAAkB,oBACxC;SAEF,OAAO,gBAAgB,QAAQ,kBAAkB;GAErD,OACE,OAAO,gBAAgB,QAAQ,kBAAkB;GAGnD,uBAAA,2BAA2B,iBAAiB,kBAAkB;GAC9D,MAAMC,cAAAA,wBAAwB,iBAAiB,UAAU;GACzD,MAAM;EACR;CACF;AACF"}
+{"version":3,"file":"index.cjs","names":["ChatOpenAI","emitStreamChunkCallback"],"sources":["../../../../src/llm/openrouter/index.ts"],"mappings":";;;AAoFA,SAAS,sBACP,OACwC;CACxC,OACE,OAAO,UAAU,YACjB,UAAU,QACV,UAAU,SACV,MAAM,SAAS;AAEnB;AAEA,SAAS,2BACP,OAC6C;CAC7C,OACE,OAAO,UAAU,YACjB,UAAU,QACV,UAAU,SACV,MAAM,SAAS;AAEnB;AAEA,SAAS,oBAAoB,OAA6C;CACxE,IAAI,CAAC,MAAM,QAAQ,KAAK,GACtB,OAAO,CAAC;CAEV,OAAO,MAAM,QACV,WACC,sBAAsB,MAAM,KAAK,2BAA2B,MAAM,CACtE;AACF;AAEA,IAAa,iBAAb,cAAoCA,cAAAA,WAAW;CAC7C;CAEA;CAEA,YAAY,SAA8B;EACxC,MAAM,2BAAgD,EAAE,GAAG,QAAQ;EACnE,OAAO,yBAAyB;EAChC,OAAO,yBAAyB;EAEhC,MAAM,EACJ,mBACA,WAAW,qBACX,cAAc,CAAC,GACf,GAAG,WACD;EAGJ,MAAM,EAAE,WAAW,aAAa,GAAG,oBAAoB;EAGvD,MAAM,kBACJ,eAAe,QAAQ,uBAAuB,OAC1C;GACA,GAAG;GACH,GAAG;EACL,IACE,KAAA;EACN,MAAM,mBACJ,oBACC,sBAAsB,OAAO,EAAE,SAAS,KAAK,IAAI,KAAA;EACpD,MAAM,oBACJ,oBAAoB,OAChB,kBACA;GAAE,GAAG;GAAiB,WAAW;EAAiB;EAExD,MAAM;GACJ,GAAG;GACH,aAAa;GACb,yBAAyB;GACzB,kCAAkC;GAClC,0BAA0B;GAC1B,sBAAsB,QAAQ;GAC9B,yBAAyB,QAAQ;EACnC,CAAC;EAGD,IAAI,mBAAmB,MACrB,KAAK,sBAAsB;EAG7B,KAAK,mBAAmB;CAC1B;CACA,OAAO,UAAiC;EACtC,OAAO;CACT;CAGA,iBACE,SACA,OAC4B;EAU5B,MAAM,sBAAsB,KAAK,oBAAoB,OAAO;EAE5D,MAAM,SADkB,KAAK,iBAAiB,OAE9B,IACV,KAAK,UAAU,iBAAiB,mBAAmB,IACnD,KAAK,YAAY,iBAAiB,qBAAqB,KAAK;EAKlE,OAAO,OAAO;EAGd,MAAM,YAAY,KAAK,yBAAyB,mBAAmB;EACnE,IAAI,aAAa,MACf,OAAO,YAAY;OAEnB,OAAO,OAAO;EAGhB,OAAO;CACT;CAEA,yBACE,SACiC;EACjC,IAAI;EAGJ,IAAI,KAAK,uBAAuB,MAC9B,YAAY,EAAE,GAAG,KAAK,oBAAoB;EAI5C,MAAM,cAAc,KAAK,mBAAmB,OAAO;EACnD,IAAI,aAAa,UAAU,MACzB,YAAY;GACV,GAAG;GACH,QAAQ,YAAY;EACtB;EAIF,MAAM,gBAAiB,SACnB;EACJ,IAAI,iBAAiB,MACnB,YAAY;GAAE,GAAG;GAAW,GAAG;EAAc;EAI/C,IAAI,aAAa,QAAQ,KAAK,qBAAqB,MACjD,YAAY,EAAE,SAAS,KAAK;EAG9B,OAAO;CACT;CAEA,OAAgB,sBACd,UACA,SACA,YACqC;EACrC,MAAM,uCAAuB,IAAI,IAG/B;EACF,MAAM,yCAAyB,IAAI,IAGjC;EAGF,MAAM,qCAAyC,IAAI,IAAI;EAEvD,WAAW,MAAM,mBAAmB,MAAM,yBACxC,UACA,SACA,KAAA,CACF,GAAG;GACD,IAAI,uBAAuB;GAC3B,MAAM,mBAAmB,oBACvB,gBAAgB,QAAQ,kBAAkB,iBAC5C;GAEA,KAAK,MAAM,UAAU,kBAAkB;IACrC,IAAI,OAAO,SAAS,kBAAkB;KACpC,wBAAwB,OAAO,QAAQ;KACvC,MAAM,QAAQ,OAAO,SAAS;KAC9B,MAAM,WAAW,qBAAqB,IAAI,KAAK;KAC/C,IAAI,YAAY,MAAM;MACpB,SAAS,OAAO,GAAG,SAAS,QAAQ,KAAK,OAAO,QAAQ;MACxD;KACF;KACA,qBAAqB,IAAI,OAAO;MAC9B,GAAG;MACH,MAAM,OAAO,QAAQ;KACvB,CAAC;KACD;IACF;IACA,IAAI,OAAO,MAAM,MACf,uBAAuB,IAAI,OAAO,IAAI,EAAE,GAAG,OAAO,CAAC;GAEvD;GAEA,IACE,qBAAqB,SAAS,KAC9B,gBAAgB,QAAQ,kBAAkB,aAAa,MAEvD,gBAAgB,QAAQ,kBAAkB,YACxC;GAGJ,IAAI,gBAAgB,gBAAgB,iBAAiB,MAAM;IACzD,MAAM,wBAAwB,CAC5B,GAAG,qBAAqB,OAAO,GAC/B,GAAG,uBAAuB,OAAO,CACnC;IACA,IAAI,sBAAsB,SAAS,GACjC,gBAAgB,QAAQ,kBAAkB,oBACxC;SAEF,OAAO,gBAAgB,QAAQ,kBAAkB;GAErD,OACE,OAAO,gBAAgB,QAAQ,kBAAkB;GAGnD,uBAAA,2BAA2B,iBAAiB,kBAAkB;GAC9D,MAAMC,cAAAA,wBAAwB,iBAAiB,UAAU;GACzD,MAAM;EACR;CACF;AACF"}

package/dist/cjs/llm/openrouter/toolCache.cjs

@@ -23,12 +23,6 @@ function markCacheControl(tool, ttl) {
 		cache_control: require_cache.buildAnthropicCacheControl(ttl)
 	};
 }
-/**
-* Drop any existing `cache_control` from a tool. Reused/caller-supplied tools
-* can carry a stale marker (e.g. from a prior `promptCacheTtl: '5m'` run); since
-* all tools serialize before system/messages, a leftover 5-minute marker ahead
-* of the resolved breakpoint would violate the longer-TTL-first ordering.
-*/
 function stripCacheControl(tool) {
 	if (tool.cache_control == null) return tool;
 	const { cache_control: _omit, ...rest } = tool;

package/dist/cjs/llm/openrouter/toolCache.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"toolCache.cjs","names":["_convertToOpenAITool","buildAnthropicCacheControl"],"sources":["../../../../src/llm/openrouter/toolCache.ts"],"sourcesContent":["import type { BindToolsInput } from '@langchain/core/language_models/chat_models';\nimport type { OpenAIClient } from '@langchain/openai';\nimport type { GraphTools } from '@/types';\nimport {\n  buildAnthropicCacheControl,\n  type PromptCacheTtl,\n} from '@/messages/cache';\nimport { _convertToOpenAITool } from '@/llm/openai';\n\ntype OpenRouterCacheControl = { type: 'ephemeral'; ttl?: '1h' };\n\ntype OpenRouterToolWithCacheControl = OpenAIClient.ChatCompletionTool & {\n  cache_control?: OpenRouterCacheControl;\n  defer_loading?: boolean;\n};\n\ntype ToolNameCandidate = {\n  name?: unknown;\n  function?: {\n    name?: unknown;\n  };\n  defer_loading?: unknown;\n};\n\nfunction getToolName(tool: unknown): string | undefined {\n  const candidate = tool as ToolNameCandidate;\n  if (typeof candidate.name === 'string') {\n    return candidate.name;\n  }\n  if (typeof candidate.function?.name === 'string') {\n    return candidate.function.name;\n  }\n  return undefined;\n}\n\nfunction hasDeferredMarker(tool: unknown): boolean {\n  return (tool as ToolNameCandidate).defer_loading === true;\n}\n\nfunction toOpenRouterTool(tool: unknown): OpenRouterToolWithCacheControl {\n  const converted = _convertToOpenAITool(\n    tool as BindToolsInput\n  ) as OpenRouterToolWithCacheControl;\n\n  if (hasDeferredMarker(tool)) {\n    return { ...converted, defer_loading: true };\n  }\n\n  return converted;\n}\n\nfunction markCacheControl(\n  tool: OpenRouterToolWithCacheControl,\n  ttl?: PromptCacheTtl\n): OpenRouterToolWithCacheControl {\n  return {\n    ...tool,\n    cache_control: buildAnthropicCacheControl(ttl),\n  };\n}\n\n/**\n * Drop any existing `cache_control` from a tool. Reused/caller-supplied tools\n * can carry a stale marker (e.g. from a prior `promptCacheTtl: '5m'` run); since\n * all tools serialize before system/messages, a leftover 5-minute marker ahead\n * of the resolved breakpoint would violate the longer-TTL-first ordering.\n */\nfunction stripCacheControl(\n  tool: OpenRouterToolWithCacheControl\n): OpenRouterToolWithCacheControl {\n  if (tool.cache_control == null) {\n    return tool;\n  }\n  const { cache_control: _omit, ...rest } = tool;\n  return rest;\n}\n\nexport function partitionAndMarkOpenRouterToolCache(\n  tools: GraphTools | undefined,\n  isDeferred: (toolName: string) => boolean,\n  ttl?: PromptCacheTtl\n): GraphTools | undefined {\n  if (tools == null || tools.length === 0) {\n    return tools;\n  }\n\n  const staticTools: OpenRouterToolWithCacheControl[] = [];\n  const deferredTools: OpenRouterToolWithCacheControl[] = [];\n\n  for (const tool of tools as readonly unknown[]) {\n    const converted = toOpenRouterTool(tool);\n    const name = getToolName(converted) ?? getToolName(tool);\n\n    if (name != null && isDeferred(name)) {\n      deferredTools.push(converted);\n      continue;\n    }\n\n    staticTools.push(converted);\n  }\n\n  // Deferred tools sit after the breakpoint but still before system/messages,\n  // so strip any stale marker off them.\n  for (let i = 0; i < deferredTools.length; i++) {\n    deferredTools[i] = stripCacheControl(deferredTools[i]);\n  }\n\n  if (staticTools.length === 0) {\n    return [...deferredTools] as GraphTools;\n  }\n\n  // Strip stale markers off the earlier static tools, then stamp only the last\n  // static tool with the resolved TTL (markCacheControl overwrites any marker).\n  for (let i = 0; i < staticTools.length - 1; i++) {\n    staticTools[i] = stripCacheControl(staticTools[i]);\n  }\n  staticTools[staticTools.length - 1] = markCacheControl(\n    staticTools[staticTools.length - 1],\n    ttl\n  );\n\n  return [...staticTools, ...deferredTools] as GraphTools;\n}\n"],"mappings":";;;AAwBA,SAAS,YAAY,MAAmC;CACtD,MAAM,YAAY;CAClB,IAAI,OAAO,UAAU,SAAS,UAC5B,OAAO,UAAU;CAEnB,IAAI,OAAO,UAAU,UAAU,SAAS,UACtC,OAAO,UAAU,SAAS;AAG9B;AAEA,SAAS,kBAAkB,MAAwB;CACjD,OAAQ,KAA2B,kBAAkB;AACvD;AAEA,SAAS,iBAAiB,MAA+C;CACvE,MAAM,YAAYA,cAAAA,qBAChB,IACF;CAEA,IAAI,kBAAkB,IAAI,GACxB,OAAO;EAAE,GAAG;EAAW,eAAe;CAAK;CAG7C,OAAO;AACT;AAEA,SAAS,iBACP,MACA,KACgC;CAChC,OAAO;EACL,GAAG;EACH,eAAeC,cAAAA,2BAA2B,GAAG;CAC/C;AACF;;;;;;;AAQA,SAAS,kBACP,MACgC;CAChC,IAAI,KAAK,iBAAiB,MACxB,OAAO;CAET,MAAM,EAAE,eAAe,OAAO,GAAG,SAAS;CAC1C,OAAO;AACT;AAEA,SAAgB,oCACd,OACA,YACA,KACwB;CACxB,IAAI,SAAS,QAAQ,MAAM,WAAW,GACpC,OAAO;CAGT,MAAM,cAAgD,CAAC;CACvD,MAAM,gBAAkD,CAAC;CAEzD,KAAK,MAAM,QAAQ,OAA6B;EAC9C,MAAM,YAAY,iBAAiB,IAAI;EACvC,MAAM,OAAO,YAAY,SAAS,KAAK,YAAY,IAAI;EAEvD,IAAI,QAAQ,QAAQ,WAAW,IAAI,GAAG;GACpC,cAAc,KAAK,SAAS;GAC5B;EACF;EAEA,YAAY,KAAK,SAAS;CAC5B;CAIA,KAAK,IAAI,IAAI,GAAG,IAAI,cAAc,QAAQ,KACxC,cAAc,KAAK,kBAAkB,cAAc,EAAE;CAGvD,IAAI,YAAY,WAAW,GACzB,OAAO,CAAC,GAAG,aAAa;CAK1B,KAAK,IAAI,IAAI,GAAG,IAAI,YAAY,SAAS,GAAG,KAC1C,YAAY,KAAK,kBAAkB,YAAY,EAAE;CAEnD,YAAY,YAAY,SAAS,KAAK,iBACpC,YAAY,YAAY,SAAS,IACjC,GACF;CAEA,OAAO,CAAC,GAAG,aAAa,GAAG,aAAa;AAC1C"}
+{"version":3,"file":"toolCache.cjs","names":["_convertToOpenAITool","buildAnthropicCacheControl"],"sources":["../../../../src/llm/openrouter/toolCache.ts"],"mappings":";;;AAwBA,SAAS,YAAY,MAAmC;CACtD,MAAM,YAAY;CAClB,IAAI,OAAO,UAAU,SAAS,UAC5B,OAAO,UAAU;CAEnB,IAAI,OAAO,UAAU,UAAU,SAAS,UACtC,OAAO,UAAU,SAAS;AAG9B;AAEA,SAAS,kBAAkB,MAAwB;CACjD,OAAQ,KAA2B,kBAAkB;AACvD;AAEA,SAAS,iBAAiB,MAA+C;CACvE,MAAM,YAAYA,cAAAA,qBAChB,IACF;CAEA,IAAI,kBAAkB,IAAI,GACxB,OAAO;EAAE,GAAG;EAAW,eAAe;CAAK;CAG7C,OAAO;AACT;AAEA,SAAS,iBACP,MACA,KACgC;CAChC,OAAO;EACL,GAAG;EACH,eAAeC,cAAAA,2BAA2B,GAAG;CAC/C;AACF;AAQA,SAAS,kBACP,MACgC;CAChC,IAAI,KAAK,iBAAiB,MACxB,OAAO;CAET,MAAM,EAAE,eAAe,OAAO,GAAG,SAAS;CAC1C,OAAO;AACT;AAEA,SAAgB,oCACd,OACA,YACA,KACwB;CACxB,IAAI,SAAS,QAAQ,MAAM,WAAW,GACpC,OAAO;CAGT,MAAM,cAAgD,CAAC;CACvD,MAAM,gBAAkD,CAAC;CAEzD,KAAK,MAAM,QAAQ,OAA6B;EAC9C,MAAM,YAAY,iBAAiB,IAAI;EACvC,MAAM,OAAO,YAAY,SAAS,KAAK,YAAY,IAAI;EAEvD,IAAI,QAAQ,QAAQ,WAAW,IAAI,GAAG;GACpC,cAAc,KAAK,SAAS;GAC5B;EACF;EAEA,YAAY,KAAK,SAAS;CAC5B;CAIA,KAAK,IAAI,IAAI,GAAG,IAAI,cAAc,QAAQ,KACxC,cAAc,KAAK,kBAAkB,cAAc,EAAE;CAGvD,IAAI,YAAY,WAAW,GACzB,OAAO,CAAC,GAAG,aAAa;CAK1B,KAAK,IAAI,IAAI,GAAG,IAAI,YAAY,SAAS,GAAG,KAC1C,YAAY,KAAK,kBAAkB,YAAY,EAAE;CAEnD,YAAY,YAAY,SAAS,KAAK,iBACpC,YAAY,YAAY,SAAS,IACjC,GACF;CAEA,OAAO,CAAC,GAAG,aAAa,GAAG,aAAa;AAC1C"}

package/dist/cjs/llm/preempt.cjs

@@ -2,29 +2,11 @@ require("../common/constants.cjs");
 require("../common/enum.cjs");
 require("../common/index.cjs");
 //#region src/llm/preempt.ts
-/**
-* Normalizes a host-supplied seal budget.
-*
-* Read in two places that interpret it differently — a numeric comparison in
-* the seal gate and an addition into the recursion limit — so a value like
-* `1.5` would permit two seals while reserving fractional headroom, `NaN`
-* would poison the recursion limit outright, and `Infinity` would remove both
-* bounds at once. Normalizing once keeps the two readings in agreement.
-*
-* `0` is honored as a deliberate "never seal"; anything not finite falls back
-* to the default rather than silently disabling the feature.
-*/
 function resolveMaxSeals(maxSeals) {
 	if (maxSeals == null || !Number.isFinite(maxSeals)) return 8;
 	const whole = Math.floor(maxSeals);
 	return whole > 0 ? whole : 0;
 }
-/**
-* True when the accumulated content carries at least one non-whitespace text
-* block. `thinking` / `reasoning` blocks deliberately do not count: they are
-* stripped or signed on several providers and cannot stand in for the visible
-* assistant turn a sealed sequence needs.
-*/
 function hasNonEmptyTextContent(content) {
 	if (typeof content === "string") return content.trim() !== "";
 	for (const block of content) {
@@ -34,19 +16,6 @@ function hasNonEmptyTextContent(content) {
 	}
 	return false;
 }
-/**
-* True while a Gemini server-side tool call is still awaiting its response.
-*
-* Google's server-side tools (Search, URL context) never populate
-* `tool_calls` or `tool_call_chunks` — those are derived from `functionCall`
-* parts only, while a server-side invocation arrives as a `toolCall` CONTENT
-* block and its result as a later `toolResponse` block. The tool-call gates
-* cannot see it, so sealing between the two would replay a model turn holding
-* an unanswered server-side call.
-*
-* Counted rather than id-matched: Google answers calls in order, and counting
-* stays correct when a part omits its id.
-*/
 function hasOpenGoogleServerToolCall(content) {
 	if (typeof content === "string") return false;
 	let calls = 0;
@@ -55,11 +24,6 @@ function hasOpenGoogleServerToolCall(content) {
 	else if (block.type === "toolResponse") responses += 1;
 	return calls > responses;
 }
-/**
-* Ids of Anthropic server-side tool calls whose paired result block is already
-* on the accumulated content. Those calls are answered and cannot be orphaned
-* by a seal.
-*/
 function settledServerToolCallIds(content) {
 	const settled = /* @__PURE__ */ new Set();
 	if (typeof content === "string") return settled;
@@ -70,7 +34,6 @@ function settledServerToolCallIds(content) {
 	}
 	return settled;
 }
-/** Tool calls still awaiting an answer, ignoring ones already settled. */
 function countOpenToolCalls(calls, settled) {
 	if (calls == null || calls.length === 0) return 0;
 	if (settled.size === 0) return calls.length;
@@ -78,46 +41,8 @@ function countOpenToolCalls(calls, settled) {
 	for (const call of calls) if (call.id == null || !settled.has(call.id)) open += 1;
 	return open;
 }
-/**
-* Cooperative mid-generation seal gate. Returns true ONLY when sealing here
-* yields a message sequence valid on EVERY supported provider:
-*  - non-whitespace TEXT content, so the FIRST injected user turn is preceded
-*    by a non-empty assistant turn — no empty-content 400s. Note this says
-*    nothing about adjacency AMONG several injected turns: a boundary that
-*    drains two steers emits two consecutive user messages, which strict
-*    providers reject. That is normalized at the provider-facing hop by
-*    `coalesceAdjacentUserTurns`, not here;
-*  - no tool call in flight, so no `tool_use` can be orphaned AND no eagerly
-*    prestarted execution can be stripped out from under the model.
-*
-* Anthropic's server-side tools need no check of their own. Every
-* `server_tool_use` content block also emits a `tool_call_chunk`
-* (`_makeMessageChunkFromAnthropicEvent`), and `concat` keeps that chunk on
-* the accumulated message for the remainder of the turn, so the tool-call
-* gates below already cover it. The practical consequence is worth stating
-* plainly: once a turn starts a web search it is no longer preemptible, and
-* a queued message waits for the ordinary tool boundary instead.
-*
-* Nothing is stripped and nothing is repaired: when the accumulated shape is
-* not already safe the stream simply runs on, and whatever the host queued
-* lands at the next tool boundary instead. Chunks accumulate monotonically
-* through `concat`, so an unsafe shape can never be observed at a seal point.
-*/
 function canSealPreempt(chunk) {
 	if (chunk == null) return false;
-	/**
-	* Anthropic's server-side tools run inside the provider, so unlike a client
-	* tool they never reach `ToolNode` and never produce a `PostToolBatch`
-	* boundary. `concat` also keeps their `server_tool_use` chunk on the
-	* accumulated message for the rest of the turn. Together that means a naive
-	* tool-call gate makes a turn permanently unsealable the moment a web
-	* search starts, with no later boundary to drain into — the queued message
-	* waits for the whole turn to finish rather than for the next tool step.
-	*
-	* So calls whose paired result block has already landed are treated as
-	* settled: they cannot be orphaned, because the provider has already
-	* answered them.
-	*/
 	const settled = settledServerToolCallIds(chunk.content);
 	if (countOpenToolCalls(chunk.tool_calls, settled) > 0) return false;
 	if (countOpenToolCalls(chunk.tool_call_chunks, settled) > 0) return false;

package/dist/cjs/llm/preempt.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"preempt.cjs","names":[],"sources":["../../../src/llm/preempt.ts"],"sourcesContent":["// src/llm/preempt.ts\nimport type { AIMessageChunk } from '@langchain/core/messages';\nimport { ContentTypes, DEFAULT_MAX_SEALS } from '@/common';\n\n/**\n * Normalizes a host-supplied seal budget.\n *\n * Read in two places that interpret it differently — a numeric comparison in\n * the seal gate and an addition into the recursion limit — so a value like\n * `1.5` would permit two seals while reserving fractional headroom, `NaN`\n * would poison the recursion limit outright, and `Infinity` would remove both\n * bounds at once. Normalizing once keeps the two readings in agreement.\n *\n * `0` is honored as a deliberate \"never seal\"; anything not finite falls back\n * to the default rather than silently disabling the feature.\n */\nexport function resolveMaxSeals(maxSeals: number | undefined): number {\n  if (maxSeals == null || !Number.isFinite(maxSeals)) {\n    return DEFAULT_MAX_SEALS;\n  }\n  const whole = Math.floor(maxSeals);\n  return whole > 0 ? whole : 0;\n}\n\n/**\n * True when the accumulated content carries at least one non-whitespace text\n * block. `thinking` / `reasoning` blocks deliberately do not count: they are\n * stripped or signed on several providers and cannot stand in for the visible\n * assistant turn a sealed sequence needs.\n */\nfunction hasNonEmptyTextContent(content: AIMessageChunk['content']): boolean {\n  if (typeof content === 'string') {\n    return content.trim() !== '';\n  }\n  for (const block of content) {\n    if (block.type !== ContentTypes.TEXT) {\n      continue;\n    }\n    const text = block[ContentTypes.TEXT];\n    if (typeof text === 'string' && text.trim() !== '') {\n      return true;\n    }\n  }\n  return false;\n}\n\n/**\n * True while a Gemini server-side tool call is still awaiting its response.\n *\n * Google's server-side tools (Search, URL context) never populate\n * `tool_calls` or `tool_call_chunks` — those are derived from `functionCall`\n * parts only, while a server-side invocation arrives as a `toolCall` CONTENT\n * block and its result as a later `toolResponse` block. The tool-call gates\n * cannot see it, so sealing between the two would replay a model turn holding\n * an unanswered server-side call.\n *\n * Counted rather than id-matched: Google answers calls in order, and counting\n * stays correct when a part omits its id.\n */\nfunction hasOpenGoogleServerToolCall(\n  content: AIMessageChunk['content']\n): boolean {\n  if (typeof content === 'string') {\n    return false;\n  }\n  let calls = 0;\n  let responses = 0;\n  for (const block of content) {\n    if (block.type === 'toolCall') {\n      calls += 1;\n    } else if (block.type === 'toolResponse') {\n      responses += 1;\n    }\n  }\n  return calls > responses;\n}\n\n/**\n * Ids of Anthropic server-side tool calls whose paired result block is already\n * on the accumulated content. Those calls are answered and cannot be orphaned\n * by a seal.\n */\nfunction settledServerToolCallIds(\n  content: AIMessageChunk['content']\n): Set<string> {\n  const settled = new Set<string>();\n  if (typeof content === 'string') {\n    return settled;\n  }\n  for (const block of content) {\n    if (block.type !== 'web_search_tool_result') {\n      continue;\n    }\n    const id = block.tool_use_id;\n    if (typeof id === 'string') {\n      settled.add(id);\n    }\n  }\n  return settled;\n}\n\n/** Tool calls still awaiting an answer, ignoring ones already settled. */\nfunction countOpenToolCalls(\n  calls: ReadonlyArray<{ id?: string }> | undefined,\n  settled: Set<string>\n): number {\n  if (calls == null || calls.length === 0) {\n    return 0;\n  }\n  if (settled.size === 0) {\n    return calls.length;\n  }\n  let open = 0;\n  for (const call of calls) {\n    if (call.id == null || !settled.has(call.id)) {\n      open += 1;\n    }\n  }\n  return open;\n}\n\n/**\n * Cooperative mid-generation seal gate. Returns true ONLY when sealing here\n * yields a message sequence valid on EVERY supported provider:\n *  - non-whitespace TEXT content, so the FIRST injected user turn is preceded\n *    by a non-empty assistant turn — no empty-content 400s. Note this says\n *    nothing about adjacency AMONG several injected turns: a boundary that\n *    drains two steers emits two consecutive user messages, which strict\n *    providers reject. That is normalized at the provider-facing hop by\n *    `coalesceAdjacentUserTurns`, not here;\n *  - no tool call in flight, so no `tool_use` can be orphaned AND no eagerly\n *    prestarted execution can be stripped out from under the model.\n *\n * Anthropic's server-side tools need no check of their own. Every\n * `server_tool_use` content block also emits a `tool_call_chunk`\n * (`_makeMessageChunkFromAnthropicEvent`), and `concat` keeps that chunk on\n * the accumulated message for the remainder of the turn, so the tool-call\n * gates below already cover it. The practical consequence is worth stating\n * plainly: once a turn starts a web search it is no longer preemptible, and\n * a queued message waits for the ordinary tool boundary instead.\n *\n * Nothing is stripped and nothing is repaired: when the accumulated shape is\n * not already safe the stream simply runs on, and whatever the host queued\n * lands at the next tool boundary instead. Chunks accumulate monotonically\n * through `concat`, so an unsafe shape can never be observed at a seal point.\n */\nexport function canSealPreempt(chunk: AIMessageChunk | undefined): boolean {\n  if (chunk == null) {\n    return false;\n  }\n  /**\n   * Anthropic's server-side tools run inside the provider, so unlike a client\n   * tool they never reach `ToolNode` and never produce a `PostToolBatch`\n   * boundary. `concat` also keeps their `server_tool_use` chunk on the\n   * accumulated message for the rest of the turn. Together that means a naive\n   * tool-call gate makes a turn permanently unsealable the moment a web\n   * search starts, with no later boundary to drain into — the queued message\n   * waits for the whole turn to finish rather than for the next tool step.\n   *\n   * So calls whose paired result block has already landed are treated as\n   * settled: they cannot be orphaned, because the provider has already\n   * answered them.\n   */\n  const settled = settledServerToolCallIds(chunk.content);\n  if (countOpenToolCalls(chunk.tool_calls, settled) > 0) {\n    return false;\n  }\n  if (countOpenToolCalls(chunk.tool_call_chunks, settled) > 0) {\n    return false;\n  }\n  if ((chunk.invalid_tool_calls?.length ?? 0) > 0) {\n    return false;\n  }\n  if (hasOpenGoogleServerToolCall(chunk.content)) {\n    return false;\n  }\n  return hasNonEmptyTextContent(chunk.content);\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAgBA,SAAgB,gBAAgB,UAAsC;CACpE,IAAI,YAAY,QAAQ,CAAC,OAAO,SAAS,QAAQ,GAC/C,OAAA;CAEF,MAAM,QAAQ,KAAK,MAAM,QAAQ;CACjC,OAAO,QAAQ,IAAI,QAAQ;AAC7B;;;;;;;AAQA,SAAS,uBAAuB,SAA6C;CAC3E,IAAI,OAAO,YAAY,UACrB,OAAO,QAAQ,KAAK,MAAM;CAE5B,KAAK,MAAM,SAAS,SAAS;EAC3B,IAAI,MAAM,SAAA,QACR;EAEF,MAAM,OAAO,MAAA;EACb,IAAI,OAAO,SAAS,YAAY,KAAK,KAAK,MAAM,IAC9C,OAAO;CAEX;CACA,OAAO;AACT;;;;;;;;;;;;;;AAeA,SAAS,4BACP,SACS;CACT,IAAI,OAAO,YAAY,UACrB,OAAO;CAET,IAAI,QAAQ;CACZ,IAAI,YAAY;CAChB,KAAK,MAAM,SAAS,SAClB,IAAI,MAAM,SAAS,YACjB,SAAS;MACJ,IAAI,MAAM,SAAS,gBACxB,aAAa;CAGjB,OAAO,QAAQ;AACjB;;;;;;AAOA,SAAS,yBACP,SACa;CACb,MAAM,0BAAU,IAAI,IAAY;CAChC,IAAI,OAAO,YAAY,UACrB,OAAO;CAET,KAAK,MAAM,SAAS,SAAS;EAC3B,IAAI,MAAM,SAAS,0BACjB;EAEF,MAAM,KAAK,MAAM;EACjB,IAAI,OAAO,OAAO,UAChB,QAAQ,IAAI,EAAE;CAElB;CACA,OAAO;AACT;;AAGA,SAAS,mBACP,OACA,SACQ;CACR,IAAI,SAAS,QAAQ,MAAM,WAAW,GACpC,OAAO;CAET,IAAI,QAAQ,SAAS,GACnB,OAAO,MAAM;CAEf,IAAI,OAAO;CACX,KAAK,MAAM,QAAQ,OACjB,IAAI,KAAK,MAAM,QAAQ,CAAC,QAAQ,IAAI,KAAK,EAAE,GACzC,QAAQ;CAGZ,OAAO;AACT;;;;;;;;;;;;;;;;;;;;;;;;;;AA2BA,SAAgB,eAAe,OAA4C;CACzE,IAAI,SAAS,MACX,OAAO;;;;;;;;;;;;;;CAeT,MAAM,UAAU,yBAAyB,MAAM,OAAO;CACtD,IAAI,mBAAmB,MAAM,YAAY,OAAO,IAAI,GAClD,OAAO;CAET,IAAI,mBAAmB,MAAM,kBAAkB,OAAO,IAAI,GACxD,OAAO;CAET,KAAK,MAAM,oBAAoB,UAAU,KAAK,GAC5C,OAAO;CAET,IAAI,4BAA4B,MAAM,OAAO,GAC3C,OAAO;CAET,OAAO,uBAAuB,MAAM,OAAO;AAC7C"}
+{"version":3,"file":"preempt.cjs","names":[],"sources":["../../../src/llm/preempt.ts"],"mappings":";;;;AAgBA,SAAgB,gBAAgB,UAAsC;CACpE,IAAI,YAAY,QAAQ,CAAC,OAAO,SAAS,QAAQ,GAC/C,OAAA;CAEF,MAAM,QAAQ,KAAK,MAAM,QAAQ;CACjC,OAAO,QAAQ,IAAI,QAAQ;AAC7B;AAQA,SAAS,uBAAuB,SAA6C;CAC3E,IAAI,OAAO,YAAY,UACrB,OAAO,QAAQ,KAAK,MAAM;CAE5B,KAAK,MAAM,SAAS,SAAS;EAC3B,IAAI,MAAM,SAAA,QACR;EAEF,MAAM,OAAO,MAAA;EACb,IAAI,OAAO,SAAS,YAAY,KAAK,KAAK,MAAM,IAC9C,OAAO;CAEX;CACA,OAAO;AACT;AAeA,SAAS,4BACP,SACS;CACT,IAAI,OAAO,YAAY,UACrB,OAAO;CAET,IAAI,QAAQ;CACZ,IAAI,YAAY;CAChB,KAAK,MAAM,SAAS,SAClB,IAAI,MAAM,SAAS,YACjB,SAAS;MACJ,IAAI,MAAM,SAAS,gBACxB,aAAa;CAGjB,OAAO,QAAQ;AACjB;AAOA,SAAS,yBACP,SACa;CACb,MAAM,0BAAU,IAAI,IAAY;CAChC,IAAI,OAAO,YAAY,UACrB,OAAO;CAET,KAAK,MAAM,SAAS,SAAS;EAC3B,IAAI,MAAM,SAAS,0BACjB;EAEF,MAAM,KAAK,MAAM;EACjB,IAAI,OAAO,OAAO,UAChB,QAAQ,IAAI,EAAE;CAElB;CACA,OAAO;AACT;AAGA,SAAS,mBACP,OACA,SACQ;CACR,IAAI,SAAS,QAAQ,MAAM,WAAW,GACpC,OAAO;CAET,IAAI,QAAQ,SAAS,GACnB,OAAO,MAAM;CAEf,IAAI,OAAO;CACX,KAAK,MAAM,QAAQ,OACjB,IAAI,KAAK,MAAM,QAAQ,CAAC,QAAQ,IAAI,KAAK,EAAE,GACzC,QAAQ;CAGZ,OAAO;AACT;AA2BA,SAAgB,eAAe,OAA4C;CACzE,IAAI,SAAS,MACX,OAAO;CAeT,MAAM,UAAU,yBAAyB,MAAM,OAAO;CACtD,IAAI,mBAAmB,MAAM,YAAY,OAAO,IAAI,GAClD,OAAO;CAET,IAAI,mBAAmB,MAAM,kBAAkB,OAAO,IAAI,GACxD,OAAO;CAET,KAAK,MAAM,oBAAoB,UAAU,KAAK,GAC5C,OAAO;CAET,IAAI,4BAA4B,MAAM,OAAO,GAC3C,OAAO;CAET,OAAO,uBAAuB,MAAM,OAAO;AAC7C"}

package/dist/cjs/llm/providers.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"providers.cjs","names":["ChatXAI","ChatOpenAI","AzureChatOpenAI","ChatVertexAI","ChatDeepSeek","CustomChatMistralAI","CustomAnthropic","ChatOpenRouter","CustomChatBedrockConverse","CustomChatGoogleGenerativeAI","ChatMoonshot"],"sources":["../../../src/llm/providers.ts"],"sourcesContent":["// src/llm/providers.ts\nimport { CustomChatMistralAI } from '@/llm/mistral';\nimport type {\n  ChatModelConstructorMap,\n  ProviderOptionsMap,\n  ChatModelMap,\n} from '@/types';\nimport {\n  AzureChatOpenAI,\n  ChatDeepSeek,\n  ChatMoonshot,\n  ChatOpenAI,\n  ChatXAI,\n} from '@/llm/openai';\nimport { CustomChatGoogleGenerativeAI } from '@/llm/google';\nimport { CustomChatBedrockConverse } from '@/llm/bedrock';\nimport { CustomAnthropic } from '@/llm/anthropic';\nimport { ChatOpenRouter } from '@/llm/openrouter';\nimport { ChatVertexAI } from '@/llm/vertexai';\nimport { Providers } from '@/common';\n\nexport const llmProviders: Partial<ChatModelConstructorMap> = {\n  [Providers.XAI]: ChatXAI,\n  [Providers.OPENAI]: ChatOpenAI,\n  [Providers.AZURE]: AzureChatOpenAI,\n  [Providers.VERTEXAI]: ChatVertexAI,\n  [Providers.DEEPSEEK]: ChatDeepSeek,\n  [Providers.MISTRALAI]: CustomChatMistralAI,\n  [Providers.MISTRAL]: CustomChatMistralAI,\n  [Providers.ANTHROPIC]: CustomAnthropic,\n  [Providers.OPENROUTER]: ChatOpenRouter,\n  [Providers.BEDROCK]: CustomChatBedrockConverse,\n  // [Providers.ANTHROPIC]: ChatAnthropic,\n  [Providers.GOOGLE]: CustomChatGoogleGenerativeAI,\n  [Providers.MOONSHOT]: ChatMoonshot,\n};\n\nexport const manualToolStreamProviders = new Set<Providers | string>([\n  Providers.ANTHROPIC,\n  Providers.BEDROCK,\n]);\n\nexport const getChatModelClass = <P extends Providers>(\n  provider: P\n): new (config: ProviderOptionsMap[P]) => ChatModelMap[P] => {\n  const ChatModelClass = llmProviders[provider];\n  if (!ChatModelClass) {\n    throw new Error(`Unsupported LLM provider: ${provider}`);\n  }\n\n  return ChatModelClass;\n};\n"],"mappings":";;;;;;;;;;AAqBA,MAAa,eAAiD;UAC3CA,gBAAAA;aACGC,gBAAAA;kBACDC,gBAAAA;eACGC,gBAAAA;eACAC,gBAAAA;gBACCC,gBAAAA;cACFA,gBAAAA;gBACEC,gBAAAA;iBACCC,gBAAAA;cACHC,gBAAAA;aAEDC,gBAAAA;eACEC,gBAAAA;AACxB;AAEA,MAAa,4BAA4B,IAAI,IAAwB,CAAA,aAAA,SAGrE,CAAC;AAED,MAAa,qBACX,aAC2D;CAC3D,MAAM,iBAAiB,aAAa;CACpC,IAAI,CAAC,gBACH,MAAM,IAAI,MAAM,6BAA6B,UAAU;CAGzD,OAAO;AACT"}
+{"version":3,"file":"providers.cjs","names":["ChatXAI","ChatOpenAI","AzureChatOpenAI","ChatVertexAI","ChatDeepSeek","CustomChatMistralAI","CustomAnthropic","ChatOpenRouter","CustomChatBedrockConverse","CustomChatGoogleGenerativeAI","ChatMoonshot"],"sources":["../../../src/llm/providers.ts"],"mappings":";;;;;;;;;;AAqBA,MAAa,eAAiD;UAC3CA,gBAAAA;aACGC,gBAAAA;kBACDC,gBAAAA;eACGC,gBAAAA;eACAC,gBAAAA;gBACCC,gBAAAA;cACFA,gBAAAA;gBACEC,gBAAAA;iBACCC,gBAAAA;cACHC,gBAAAA;aAEDC,gBAAAA;eACEC,gBAAAA;AACxB;AAEA,MAAa,4BAA4B,IAAI,IAAwB,CAAA,aAAA,SAGrE,CAAC;AAED,MAAa,qBACX,aAC2D;CAC3D,MAAM,iBAAiB,aAAa;CACpC,IAAI,CAAC,gBACH,MAAM,IAAI,MAAM,6BAA6B,UAAU;CAGzD,OAAO;AACT"}

package/dist/cjs/llm/request.cjs

@@ -1,11 +1,6 @@
 require("../common/enum.cjs");
 require("../common/index.cjs");
 //#region src/llm/request.ts
-/**
-* Returns true when the provider + clientOptions indicate extended thinking
-* is enabled.  Works across Anthropic (direct), Bedrock (additionalModelRequestFields),
-* and OpenAI-compat (modelKwargs.thinking).
-*/
 function isThinkingEnabled(provider, clientOptions) {
 	if (!clientOptions) return false;
 	if (provider === "anthropic" && clientOptions.thinking != null) return true;
@@ -13,11 +8,6 @@ function isThinkingEnabled(provider, clientOptions) {
 	if (provider === "openAI" && (clientOptions.modelKwargs?.thinking)?.type === "enabled") return true;
 	return false;
 }
-/**
-* Returns the correct key for setting max output tokens on the model
-* constructor options.  Google/Vertex use `maxOutputTokens`, all others
-* use `maxTokens`.
-*/
 function getMaxOutputTokensKey(provider) {
 	return provider === "google" || provider === "vertexai" ? "maxOutputTokens" : "maxTokens";
 }

package/dist/cjs/llm/request.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"request.cjs","names":[],"sources":["../../../src/llm/request.ts"],"sourcesContent":["import type * as t from '@/types';\nimport { Providers } from '@/common';\n\n/**\n * Returns true when the provider + clientOptions indicate extended thinking\n * is enabled.  Works across Anthropic (direct), Bedrock (additionalModelRequestFields),\n * and OpenAI-compat (modelKwargs.thinking).\n */\nexport function isThinkingEnabled(\n  provider: Providers,\n  clientOptions?: t.ClientOptions\n): boolean {\n  if (!clientOptions) return false;\n\n  if (\n    provider === Providers.ANTHROPIC &&\n    (clientOptions as t.AnthropicClientOptions).thinking != null\n  ) {\n    return true;\n  }\n\n  if (\n    provider === Providers.BEDROCK &&\n    (clientOptions as t.BedrockAnthropicInput).additionalModelRequestFields?.[\n      'thinking'\n    ] != null\n  ) {\n    return true;\n  }\n\n  if (\n    provider === Providers.OPENAI &&\n    (\n      (clientOptions as t.OpenAIClientOptions).modelKwargs\n        ?.thinking as t.AnthropicClientOptions['thinking']\n    )?.type === 'enabled'\n  ) {\n    return true;\n  }\n\n  return false;\n}\n\n/**\n * Returns the correct key for setting max output tokens on the model\n * constructor options.  Google/Vertex use `maxOutputTokens`, all others\n * use `maxTokens`.\n */\nexport function getMaxOutputTokensKey(\n  provider: Providers | string\n): 'maxOutputTokens' | 'maxTokens' {\n  return provider === Providers.GOOGLE || provider === Providers.VERTEXAI\n    ? 'maxOutputTokens'\n    : 'maxTokens';\n}\n"],"mappings":";;;;;;;;AAQA,SAAgB,kBACd,UACA,eACS;CACT,IAAI,CAAC,eAAe,OAAO;CAE3B,IACE,aAAA,eACC,cAA2C,YAAY,MAExD,OAAO;CAGT,IACE,aAAA,aACC,cAA0C,+BACzC,eACG,MAEL,OAAO;CAGT,IACE,aAAA,aAEG,cAAwC,aACrC,SAAA,EACH,SAAS,WAEZ,OAAO;CAGT,OAAO;AACT;;;;;;AAOA,SAAgB,sBACd,UACiC;CACjC,OAAO,aAAA,YAAiC,aAAA,aACpC,oBACA;AACN"}
+{"version":3,"file":"request.cjs","names":[],"sources":["../../../src/llm/request.ts"],"mappings":";;;AAQA,SAAgB,kBACd,UACA,eACS;CACT,IAAI,CAAC,eAAe,OAAO;CAE3B,IACE,aAAA,eACC,cAA2C,YAAY,MAExD,OAAO;CAGT,IACE,aAAA,aACC,cAA0C,+BACzC,eACG,MAEL,OAAO;CAGT,IACE,aAAA,aAEG,cAAwC,aACrC,SAAA,EACH,SAAS,WAEZ,OAAO;CAGT,OAAO;AACT;AAOA,SAAgB,sBACd,UACiC;CACjC,OAAO,aAAA,YAAiC,aAAA,aACpC,oBACA;AACN"}

package/dist/cjs/llm/stream/chunkAdapters.cjs

@@ -2,14 +2,6 @@ const require_smoother = require("./smoother.cjs");
 let _langchain_core_messages = require("@langchain/core/messages");
 let _langchain_core_outputs = require("@langchain/core/outputs");
 //#region src/llm/stream/chunkAdapters.ts
-/**
-* Rebuilds a generation chunk carrying one piece of a split plain-text chunk.
-* Unsplit pieces return the original chunk untouched, so disabled smoothing is
-* byte-identical to no smoothing. Usage metadata, additional kwargs and
-* response metadata survive only on the first piece: the aggregator merges
-* dicts by concatenating string fields, so replicating them across pieces
-* would duplicate reasoning text and scalar metadata once per piece.
-*/
 function cloneGenerationChunkPiece(chunk, piece) {
 	if (piece.isFirst && piece.isLast) return chunk;
 	const message = chunk.message;
@@ -29,15 +21,6 @@ const SPLITTABLE_TEXT_PART_KEYS = new Set([
 	"text",
 	"index"
 ]);
-/**
-* Returns the sole plain text part of a single-element complex content array
-* (the shape google-common emits for text deltas), or null when the part
-* carries anything beyond `type`/`text`/`index` (thought signatures, media,
-* function calls) and must not be sliced. The part must carry a numeric
-* `index`: it is the aggregator's list-merge key, and without it each split
-* piece would append as a separate content part instead of merging back
-* into one — index-less parts pace whole instead.
-*/
 function getSplittableTextPart(message, chunkText) {
 	const content = message.content;
 	if (!Array.isArray(content) || content.length !== 1) return null;
@@ -64,25 +47,6 @@ function clonePartGenerationChunkPiece(chunk, part, piece) {
 		}))
 	});
 }
-/**
-* Provider-agnostic classification of a `ChatGenerationChunk` for the
-* smoothing engine:
-* - splittable: plain visible text — string content equal to `chunk.text`, or
-*   a single plain text part (google-common's delta shape) — with no
-*   logprobs / finish_reason; sliced adaptively at the pacing cadence.
-* - atomic: any other text-bearing chunk (complex content arrays, logprobs,
-*   finish_reason, provider-specific reasoning payloads surfaced via
-*   `getAtomicText`) — paced as one piece, never split.
-* - passthrough: tool-call deltas, usage-only and metadata chunks — strict
-*   FIFO, zero delay.
-*/
-/**
-* google-common stamps `logprobs: { content: [] }` on every chunk, so only
-* logprobs that actually carry data may block splitting here. This is looser
-* than the OpenAI-family adapter in `llm/openai/index.ts`, which treats ANY
-* logprobs as atomic (its providers only attach logprobs when requested, and
-* the DeepSeek suite pins that contract) — do not unify the two predicates.
-*/
 function hasMeaningfulLogprobs(generationInfo) {
 	const logprobs = generationInfo?.logprobs;
 	if (logprobs == null) return false;
@@ -90,23 +54,9 @@ function hasMeaningfulLogprobs(generationInfo) {
 	if (Array.isArray(content)) return content.length > 0;
 	return true;
 }
-/** google-common reports the terminal reason camelCase (`finishReason`). */
 function hasFinishReason(generationInfo) {
 	return generationInfo?.finish_reason != null || generationInfo?.finishReason != null;
 }
-/**
-* Chunks that pair visible text with reasoning payloads in
-* `additional_kwargs` (Gemini thought summaries, reasoning_content deltas,
-* OpenRouter reasoning_details) must pace whole: split pieces would each
-* carry the same kwargs and downstream merging — the aggregator's dict merge
-* or OpenRouter's reasoning_details accumulation — duplicates them once per
-* piece.
-*/
-/**
-* Mixed text/tool-call deltas must never split: cloned pieces would each
-* carry the same tool_call_chunks and downstream accumulation would corrupt
-* the assembled tool arguments.
-*/
 function hasToolCallChunks(message) {
 	return (message.tool_call_chunks?.length ?? 0) > 0;
 }
@@ -116,12 +66,6 @@ function hasReasoningKwargs(message) {
 	if (Array.isArray(kwargs.reasoning_details) && kwargs.reasoning_details.length > 0) return true;
 	return kwargs.reasoning != null;
 }
-/**
-* Extracts the visible reasoning text a kwargs-borne delta contributes, so
-* reasoning-only chunks (Gemini thoughts, DeepSeek reasoning_content,
-* OpenAI reasoning summaries, OpenRouter reasoning_details) pace atomically
-* at the cadence instead of passing through unsmoothed.
-*/
 function getReasoningKwargsText(message) {
 	const kwargs = message.additional_kwargs;
 	if (typeof kwargs.reasoning_content === "string" && kwargs.reasoning_content !== "") return kwargs.reasoning_content;
@@ -170,12 +114,6 @@ function toGenerationSmoothItem(chunk, getAtomicText) {
 		emit: () => chunk
 	};
 }
-/**
-* Wraps a provider's raw chunk stream with adaptive smoothing and per-piece
-* `handleLLMNewToken` dispatch. The raw stream must NOT dispatch runManager
-* callbacks itself — callback-echo consumers would otherwise observe the
-* unsmoothed deltas.
-*/
 async function* smoothGenerationChunks({ chunks, delayMs, signal, runManager }) {
 	const source = (async function* () {
 		for await (const chunk of chunks) yield toGenerationSmoothItem(chunk, getReasoningKwargsText);

package/dist/cjs/llm/stream/chunkAdapters.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"chunkAdapters.cjs","names":["ChatGenerationChunk","AIMessageChunk","smoothStream"],"sources":["../../../../src/llm/stream/chunkAdapters.ts"],"sourcesContent":["import { AIMessageChunk } from '@langchain/core/messages';\nimport { ChatGenerationChunk } from '@langchain/core/outputs';\nimport type { CallbackManagerForLLMRun } from '@langchain/core/callbacks/manager';\nimport type { SmoothItem, SmoothPiece } from '@/llm/stream/smoother';\nimport { smoothStream } from '@/llm/stream/smoother';\n\n/**\n * Rebuilds a generation chunk carrying one piece of a split plain-text chunk.\n * Unsplit pieces return the original chunk untouched, so disabled smoothing is\n * byte-identical to no smoothing. Usage metadata, additional kwargs and\n * response metadata survive only on the first piece: the aggregator merges\n * dicts by concatenating string fields, so replicating them across pieces\n * would duplicate reasoning text and scalar metadata once per piece.\n */\nexport function cloneGenerationChunkPiece(\n  chunk: ChatGenerationChunk,\n  piece: SmoothPiece\n): ChatGenerationChunk {\n  if (piece.isFirst && piece.isLast) {\n    return chunk;\n  }\n  const message = chunk.message as AIMessageChunk;\n  return new ChatGenerationChunk({\n    text: piece.text,\n    generationInfo: piece.isFirst ? chunk.generationInfo : undefined,\n    message: new AIMessageChunk(\n      Object.assign({}, message, {\n        content: piece.text,\n        usage_metadata: piece.isFirst ? message.usage_metadata : undefined,\n        additional_kwargs: piece.isFirst ? message.additional_kwargs : {},\n        response_metadata: piece.isFirst ? message.response_metadata : {},\n      })\n    ),\n  });\n}\n\nconst SPLITTABLE_TEXT_PART_KEYS: ReadonlySet<string> = new Set([\n  'type',\n  'text',\n  'index',\n]);\n\n/**\n * Returns the sole plain text part of a single-element complex content array\n * (the shape google-common emits for text deltas), or null when the part\n * carries anything beyond `type`/`text`/`index` (thought signatures, media,\n * function calls) and must not be sliced. The part must carry a numeric\n * `index`: it is the aggregator's list-merge key, and without it each split\n * piece would append as a separate content part instead of merging back\n * into one — index-less parts pace whole instead.\n */\nfunction getSplittableTextPart(\n  message: AIMessageChunk,\n  chunkText: string\n): Record<string, unknown> | null {\n  const content = message.content;\n  if (!Array.isArray(content) || content.length !== 1) {\n    return null;\n  }\n  const part: unknown = content[0];\n  if (part == null || typeof part !== 'object') {\n    return null;\n  }\n  const record = part as Record<string, unknown>;\n  if (typeof record.text !== 'string' || record.text !== chunkText) {\n    return null;\n  }\n  if (record.type != null && record.type !== 'text') {\n    return null;\n  }\n  if (typeof record.index !== 'number') {\n    return null;\n  }\n  if (!Object.keys(record).every((key) => SPLITTABLE_TEXT_PART_KEYS.has(key))) {\n    return null;\n  }\n  return record;\n}\n\nfunction clonePartGenerationChunkPiece(\n  chunk: ChatGenerationChunk,\n  part: Record<string, unknown>,\n  piece: SmoothPiece\n): ChatGenerationChunk {\n  if (piece.isFirst && piece.isLast) {\n    return chunk;\n  }\n  const message = chunk.message as AIMessageChunk;\n  return new ChatGenerationChunk({\n    text: piece.text,\n    generationInfo: piece.isFirst ? chunk.generationInfo : undefined,\n    message: new AIMessageChunk(\n      Object.assign({}, message, {\n        content: [Object.assign({}, part, { text: piece.text })],\n        usage_metadata: piece.isFirst ? message.usage_metadata : undefined,\n        additional_kwargs: piece.isFirst ? message.additional_kwargs : {},\n        response_metadata: piece.isFirst ? message.response_metadata : {},\n      })\n    ),\n  });\n}\n\n/**\n * Provider-agnostic classification of a `ChatGenerationChunk` for the\n * smoothing engine:\n * - splittable: plain visible text — string content equal to `chunk.text`, or\n *   a single plain text part (google-common's delta shape) — with no\n *   logprobs / finish_reason; sliced adaptively at the pacing cadence.\n * - atomic: any other text-bearing chunk (complex content arrays, logprobs,\n *   finish_reason, provider-specific reasoning payloads surfaced via\n *   `getAtomicText`) — paced as one piece, never split.\n * - passthrough: tool-call deltas, usage-only and metadata chunks — strict\n *   FIFO, zero delay.\n */\n/**\n * google-common stamps `logprobs: { content: [] }` on every chunk, so only\n * logprobs that actually carry data may block splitting here. This is looser\n * than the OpenAI-family adapter in `llm/openai/index.ts`, which treats ANY\n * logprobs as atomic (its providers only attach logprobs when requested, and\n * the DeepSeek suite pins that contract) — do not unify the two predicates.\n */\nfunction hasMeaningfulLogprobs(\n  generationInfo: ChatGenerationChunk['generationInfo']\n): boolean {\n  const logprobs = generationInfo?.logprobs;\n  if (logprobs == null) {\n    return false;\n  }\n  const content = (logprobs as { content?: unknown }).content;\n  if (Array.isArray(content)) {\n    return content.length > 0;\n  }\n  return true;\n}\n\n/** google-common reports the terminal reason camelCase (`finishReason`). */\nfunction hasFinishReason(\n  generationInfo: ChatGenerationChunk['generationInfo']\n): boolean {\n  return (\n    generationInfo?.finish_reason != null || generationInfo?.finishReason != null\n  );\n}\n\n/**\n * Chunks that pair visible text with reasoning payloads in\n * `additional_kwargs` (Gemini thought summaries, reasoning_content deltas,\n * OpenRouter reasoning_details) must pace whole: split pieces would each\n * carry the same kwargs and downstream merging — the aggregator's dict merge\n * or OpenRouter's reasoning_details accumulation — duplicates them once per\n * piece.\n */\n/**\n * Mixed text/tool-call deltas must never split: cloned pieces would each\n * carry the same tool_call_chunks and downstream accumulation would corrupt\n * the assembled tool arguments.\n */\nexport function hasToolCallChunks(message: AIMessageChunk): boolean {\n  return (message.tool_call_chunks?.length ?? 0) > 0;\n}\n\nexport function hasReasoningKwargs(message: AIMessageChunk): boolean {\n  const kwargs = message.additional_kwargs;\n  if (\n    typeof kwargs.reasoning_content === 'string' &&\n    kwargs.reasoning_content !== ''\n  ) {\n    return true;\n  }\n  if (\n    Array.isArray(kwargs.reasoning_details) &&\n    kwargs.reasoning_details.length > 0\n  ) {\n    return true;\n  }\n  return kwargs.reasoning != null;\n}\n\n/**\n * Extracts the visible reasoning text a kwargs-borne delta contributes, so\n * reasoning-only chunks (Gemini thoughts, DeepSeek reasoning_content,\n * OpenAI reasoning summaries, OpenRouter reasoning_details) pace atomically\n * at the cadence instead of passing through unsmoothed.\n */\nexport function getReasoningKwargsText(message: AIMessageChunk): string {\n  const kwargs = message.additional_kwargs;\n  if (\n    typeof kwargs.reasoning_content === 'string' &&\n    kwargs.reasoning_content !== ''\n  ) {\n    return kwargs.reasoning_content;\n  }\n  const reasoning = kwargs.reasoning;\n  if (typeof reasoning === 'string' && reasoning !== '') {\n    return reasoning;\n  }\n  if (reasoning != null && typeof reasoning === 'object') {\n    const summaryText = (\n      reasoning as { summary?: { text?: unknown }[] }\n    ).summary?.[0]?.text;\n    if (typeof summaryText === 'string' && summaryText !== '') {\n      return summaryText;\n    }\n  }\n  const details = kwargs.reasoning_details;\n  if (Array.isArray(details)) {\n    let text = '';\n    for (const detail of details) {\n      if (detail != null && typeof detail === 'object') {\n        const detailText = (detail as { text?: unknown }).text;\n        if (typeof detailText === 'string') {\n          text += detailText;\n        }\n      }\n    }\n    if (text !== '') {\n      return text;\n    }\n  }\n  return '';\n}\n\nexport function toGenerationSmoothItem(\n  chunk: ChatGenerationChunk,\n  getAtomicText?: (message: AIMessageChunk) => string\n): SmoothItem<ChatGenerationChunk> {\n  const { message } = chunk;\n  const isMessageChunk = message instanceof AIMessageChunk;\n  const cleanGenerationInfo =\n    !hasMeaningfulLogprobs(chunk.generationInfo) &&\n    !hasFinishReason(chunk.generationInfo);\n  const splittable =\n    Boolean(chunk.text) &&\n    isMessageChunk &&\n    typeof message.content === 'string' &&\n    message.content === chunk.text &&\n    cleanGenerationInfo &&\n    !hasReasoningKwargs(message) &&\n    !hasToolCallChunks(message);\n\n  if (splittable) {\n    return {\n      text: chunk.text,\n      smooth: true,\n      emit: (piece) => cloneGenerationChunkPiece(chunk, piece),\n    };\n  }\n\n  const splittablePart =\n    Boolean(chunk.text) &&\n    isMessageChunk &&\n    cleanGenerationInfo &&\n    !hasReasoningKwargs(message) &&\n    !hasToolCallChunks(message)\n      ? getSplittableTextPart(message, chunk.text)\n      : null;\n  if (splittablePart != null) {\n    return {\n      text: chunk.text,\n      smooth: true,\n      emit: (piece) =>\n        clonePartGenerationChunkPiece(chunk, splittablePart, piece),\n    };\n  }\n\n  const pacedText =\n    chunk.text ||\n    (isMessageChunk && getAtomicText != null ? getAtomicText(message) : '');\n  if (pacedText !== '') {\n    return {\n      text: pacedText,\n      smooth: true,\n      atomic: true,\n      emit: () => chunk,\n    };\n  }\n\n  return { text: '', smooth: false, emit: () => chunk };\n}\n\n/**\n * Wraps a provider's raw chunk stream with adaptive smoothing and per-piece\n * `handleLLMNewToken` dispatch. The raw stream must NOT dispatch runManager\n * callbacks itself — callback-echo consumers would otherwise observe the\n * unsmoothed deltas.\n */\nexport async function* smoothGenerationChunks({\n  chunks,\n  delayMs,\n  signal,\n  runManager,\n}: {\n  chunks: AsyncGenerator<ChatGenerationChunk>;\n  delayMs: number;\n  signal?: AbortSignal;\n  runManager?: CallbackManagerForLLMRun;\n}): AsyncGenerator<ChatGenerationChunk> {\n  const source = (async function* (): AsyncGenerator<\n    SmoothItem<ChatGenerationChunk>\n    > {\n    for await (const chunk of chunks) {\n      yield toGenerationSmoothItem(chunk, getReasoningKwargsText);\n    }\n  })();\n\n  for await (const outputChunk of smoothStream({ source, delayMs, signal })) {\n    yield outputChunk;\n    await runManager?.handleLLMNewToken(\n      outputChunk.text || '',\n      undefined,\n      undefined,\n      undefined,\n      undefined,\n      { chunk: outputChunk }\n    );\n  }\n}\n"],"mappings":";;;;;;;;;;;;AAcA,SAAgB,0BACd,OACA,OACqB;CACrB,IAAI,MAAM,WAAW,MAAM,QACzB,OAAO;CAET,MAAM,UAAU,MAAM;CACtB,OAAO,IAAIA,wBAAAA,oBAAoB;EAC7B,MAAM,MAAM;EACZ,gBAAgB,MAAM,UAAU,MAAM,iBAAiB,KAAA;EACvD,SAAS,IAAIC,yBAAAA,eACX,OAAO,OAAO,CAAC,GAAG,SAAS;GACzB,SAAS,MAAM;GACf,gBAAgB,MAAM,UAAU,QAAQ,iBAAiB,KAAA;GACzD,mBAAmB,MAAM,UAAU,QAAQ,oBAAoB,CAAC;GAChE,mBAAmB,MAAM,UAAU,QAAQ,oBAAoB,CAAC;EAClE,CAAC,CACH;CACF,CAAC;AACH;AAEA,MAAM,4BAAiD,IAAI,IAAI;CAC7D;CACA;CACA;AACF,CAAC;;;;;;;;;;AAWD,SAAS,sBACP,SACA,WACgC;CAChC,MAAM,UAAU,QAAQ;CACxB,IAAI,CAAC,MAAM,QAAQ,OAAO,KAAK,QAAQ,WAAW,GAChD,OAAO;CAET,MAAM,OAAgB,QAAQ;CAC9B,IAAI,QAAQ,QAAQ,OAAO,SAAS,UAClC,OAAO;CAET,MAAM,SAAS;CACf,IAAI,OAAO,OAAO,SAAS,YAAY,OAAO,SAAS,WACrD,OAAO;CAET,IAAI,OAAO,QAAQ,QAAQ,OAAO,SAAS,QACzC,OAAO;CAET,IAAI,OAAO,OAAO,UAAU,UAC1B,OAAO;CAET,IAAI,CAAC,OAAO,KAAK,MAAM,CAAC,CAAC,OAAO,QAAQ,0BAA0B,IAAI,GAAG,CAAC,GACxE,OAAO;CAET,OAAO;AACT;AAEA,SAAS,8BACP,OACA,MACA,OACqB;CACrB,IAAI,MAAM,WAAW,MAAM,QACzB,OAAO;CAET,MAAM,UAAU,MAAM;CACtB,OAAO,IAAID,wBAAAA,oBAAoB;EAC7B,MAAM,MAAM;EACZ,gBAAgB,MAAM,UAAU,MAAM,iBAAiB,KAAA;EACvD,SAAS,IAAIC,yBAAAA,eACX,OAAO,OAAO,CAAC,GAAG,SAAS;GACzB,SAAS,CAAC,OAAO,OAAO,CAAC,GAAG,MAAM,EAAE,MAAM,MAAM,KAAK,CAAC,CAAC;GACvD,gBAAgB,MAAM,UAAU,QAAQ,iBAAiB,KAAA;GACzD,mBAAmB,MAAM,UAAU,QAAQ,oBAAoB,CAAC;GAChE,mBAAmB,MAAM,UAAU,QAAQ,oBAAoB,CAAC;EAClE,CAAC,CACH;CACF,CAAC;AACH;;;;;;;;;;;;;;;;;;;;AAqBA,SAAS,sBACP,gBACS;CACT,MAAM,WAAW,gBAAgB;CACjC,IAAI,YAAY,MACd,OAAO;CAET,MAAM,UAAW,SAAmC;CACpD,IAAI,MAAM,QAAQ,OAAO,GACvB,OAAO,QAAQ,SAAS;CAE1B,OAAO;AACT;;AAGA,SAAS,gBACP,gBACS;CACT,OACE,gBAAgB,iBAAiB,QAAQ,gBAAgB,gBAAgB;AAE7E;;;;;;;;;;;;;;AAeA,SAAgB,kBAAkB,SAAkC;CAClE,QAAQ,QAAQ,kBAAkB,UAAU,KAAK;AACnD;AAEA,SAAgB,mBAAmB,SAAkC;CACnE,MAAM,SAAS,QAAQ;CACvB,IACE,OAAO,OAAO,sBAAsB,YACpC,OAAO,sBAAsB,IAE7B,OAAO;CAET,IACE,MAAM,QAAQ,OAAO,iBAAiB,KACtC,OAAO,kBAAkB,SAAS,GAElC,OAAO;CAET,OAAO,OAAO,aAAa;AAC7B;;;;;;;AAQA,SAAgB,uBAAuB,SAAiC;CACtE,MAAM,SAAS,QAAQ;CACvB,IACE,OAAO,OAAO,sBAAsB,YACpC,OAAO,sBAAsB,IAE7B,OAAO,OAAO;CAEhB,MAAM,YAAY,OAAO;CACzB,IAAI,OAAO,cAAc,YAAY,cAAc,IACjD,OAAO;CAET,IAAI,aAAa,QAAQ,OAAO,cAAc,UAAU;EACtD,MAAM,cACJ,UACA,UAAU,EAAE,EAAE;EAChB,IAAI,OAAO,gBAAgB,YAAY,gBAAgB,IACrD,OAAO;CAEX;CACA,MAAM,UAAU,OAAO;CACvB,IAAI,MAAM,QAAQ,OAAO,GAAG;EAC1B,IAAI,OAAO;EACX,KAAK,MAAM,UAAU,SACnB,IAAI,UAAU,QAAQ,OAAO,WAAW,UAAU;GAChD,MAAM,aAAc,OAA8B;GAClD,IAAI,OAAO,eAAe,UACxB,QAAQ;EAEZ;EAEF,IAAI,SAAS,IACX,OAAO;CAEX;CACA,OAAO;AACT;AAEA,SAAgB,uBACd,OACA,eACiC;CACjC,MAAM,EAAE,YAAY;CACpB,MAAM,iBAAiB,mBAAmBA,yBAAAA;CAC1C,MAAM,sBACJ,CAAC,sBAAsB,MAAM,cAAc,KAC3C,CAAC,gBAAgB,MAAM,cAAc;CAUvC,IARE,QAAQ,MAAM,IAAI,KAClB,kBACA,OAAO,QAAQ,YAAY,YAC3B,QAAQ,YAAY,MAAM,QAC1B,uBACA,CAAC,mBAAmB,OAAO,KAC3B,CAAC,kBAAkB,OAAO,GAG1B,OAAO;EACL,MAAM,MAAM;EACZ,QAAQ;EACR,OAAO,UAAU,0BAA0B,OAAO,KAAK;CACzD;CAGF,MAAM,iBACJ,QAAQ,MAAM,IAAI,KAClB,kBACA,uBACA,CAAC,mBAAmB,OAAO,KAC3B,CAAC,kBAAkB,OAAO,IACtB,sBAAsB,SAAS,MAAM,IAAI,IACzC;CACN,IAAI,kBAAkB,MACpB,OAAO;EACL,MAAM,MAAM;EACZ,QAAQ;EACR,OAAO,UACL,8BAA8B,OAAO,gBAAgB,KAAK;CAC9D;CAGF,MAAM,YACJ,MAAM,SACL,kBAAkB,iBAAiB,OAAO,cAAc,OAAO,IAAI;CACtE,IAAI,cAAc,IAChB,OAAO;EACL,MAAM;EACN,QAAQ;EACR,QAAQ;EACR,YAAY;CACd;CAGF,OAAO;EAAE,MAAM;EAAI,QAAQ;EAAO,YAAY;CAAM;AACtD;;;;;;;AAQA,gBAAuB,uBAAuB,EAC5C,QACA,SACA,QACA,cAMsC;CACtC,MAAM,UAAU,mBAEZ;EACF,WAAW,MAAM,SAAS,QACxB,MAAM,uBAAuB,OAAO,sBAAsB;CAE9D,EAAA,CAAG;CAEH,WAAW,MAAM,eAAeC,iBAAAA,aAAa;EAAE;EAAQ;EAAS;CAAO,CAAC,GAAG;EACzE,MAAM;EACN,MAAM,YAAY,kBAChB,YAAY,QAAQ,IACpB,KAAA,GACA,KAAA,GACA,KAAA,GACA,KAAA,GACA,EAAE,OAAO,YAAY,CACvB;CACF;AACF"}
+{"version":3,"file":"chunkAdapters.cjs","names":["ChatGenerationChunk","AIMessageChunk","smoothStream"],"sources":["../../../../src/llm/stream/chunkAdapters.ts"],"mappings":";;;;AAcA,SAAgB,0BACd,OACA,OACqB;CACrB,IAAI,MAAM,WAAW,MAAM,QACzB,OAAO;CAET,MAAM,UAAU,MAAM;CACtB,OAAO,IAAIA,wBAAAA,oBAAoB;EAC7B,MAAM,MAAM;EACZ,gBAAgB,MAAM,UAAU,MAAM,iBAAiB,KAAA;EACvD,SAAS,IAAIC,yBAAAA,eACX,OAAO,OAAO,CAAC,GAAG,SAAS;GACzB,SAAS,MAAM;GACf,gBAAgB,MAAM,UAAU,QAAQ,iBAAiB,KAAA;GACzD,mBAAmB,MAAM,UAAU,QAAQ,oBAAoB,CAAC;GAChE,mBAAmB,MAAM,UAAU,QAAQ,oBAAoB,CAAC;EAClE,CAAC,CACH;CACF,CAAC;AACH;AAEA,MAAM,4BAAiD,IAAI,IAAI;CAC7D;CACA;CACA;AACF,CAAC;AAWD,SAAS,sBACP,SACA,WACgC;CAChC,MAAM,UAAU,QAAQ;CACxB,IAAI,CAAC,MAAM,QAAQ,OAAO,KAAK,QAAQ,WAAW,GAChD,OAAO;CAET,MAAM,OAAgB,QAAQ;CAC9B,IAAI,QAAQ,QAAQ,OAAO,SAAS,UAClC,OAAO;CAET,MAAM,SAAS;CACf,IAAI,OAAO,OAAO,SAAS,YAAY,OAAO,SAAS,WACrD,OAAO;CAET,IAAI,OAAO,QAAQ,QAAQ,OAAO,SAAS,QACzC,OAAO;CAET,IAAI,OAAO,OAAO,UAAU,UAC1B,OAAO;CAET,IAAI,CAAC,OAAO,KAAK,MAAM,CAAC,CAAC,OAAO,QAAQ,0BAA0B,IAAI,GAAG,CAAC,GACxE,OAAO;CAET,OAAO;AACT;AAEA,SAAS,8BACP,OACA,MACA,OACqB;CACrB,IAAI,MAAM,WAAW,MAAM,QACzB,OAAO;CAET,MAAM,UAAU,MAAM;CACtB,OAAO,IAAID,wBAAAA,oBAAoB;EAC7B,MAAM,MAAM;EACZ,gBAAgB,MAAM,UAAU,MAAM,iBAAiB,KAAA;EACvD,SAAS,IAAIC,yBAAAA,eACX,OAAO,OAAO,CAAC,GAAG,SAAS;GACzB,SAAS,CAAC,OAAO,OAAO,CAAC,GAAG,MAAM,EAAE,MAAM,MAAM,KAAK,CAAC,CAAC;GACvD,gBAAgB,MAAM,UAAU,QAAQ,iBAAiB,KAAA;GACzD,mBAAmB,MAAM,UAAU,QAAQ,oBAAoB,CAAC;GAChE,mBAAmB,MAAM,UAAU,QAAQ,oBAAoB,CAAC;EAClE,CAAC,CACH;CACF,CAAC;AACH;AAqBA,SAAS,sBACP,gBACS;CACT,MAAM,WAAW,gBAAgB;CACjC,IAAI,YAAY,MACd,OAAO;CAET,MAAM,UAAW,SAAmC;CACpD,IAAI,MAAM,QAAQ,OAAO,GACvB,OAAO,QAAQ,SAAS;CAE1B,OAAO;AACT;AAGA,SAAS,gBACP,gBACS;CACT,OACE,gBAAgB,iBAAiB,QAAQ,gBAAgB,gBAAgB;AAE7E;AAeA,SAAgB,kBAAkB,SAAkC;CAClE,QAAQ,QAAQ,kBAAkB,UAAU,KAAK;AACnD;AAEA,SAAgB,mBAAmB,SAAkC;CACnE,MAAM,SAAS,QAAQ;CACvB,IACE,OAAO,OAAO,sBAAsB,YACpC,OAAO,sBAAsB,IAE7B,OAAO;CAET,IACE,MAAM,QAAQ,OAAO,iBAAiB,KACtC,OAAO,kBAAkB,SAAS,GAElC,OAAO;CAET,OAAO,OAAO,aAAa;AAC7B;AAQA,SAAgB,uBAAuB,SAAiC;CACtE,MAAM,SAAS,QAAQ;CACvB,IACE,OAAO,OAAO,sBAAsB,YACpC,OAAO,sBAAsB,IAE7B,OAAO,OAAO;CAEhB,MAAM,YAAY,OAAO;CACzB,IAAI,OAAO,cAAc,YAAY,cAAc,IACjD,OAAO;CAET,IAAI,aAAa,QAAQ,OAAO,cAAc,UAAU;EACtD,MAAM,cACJ,UACA,UAAU,EAAE,EAAE;EAChB,IAAI,OAAO,gBAAgB,YAAY,gBAAgB,IACrD,OAAO;CAEX;CACA,MAAM,UAAU,OAAO;CACvB,IAAI,MAAM,QAAQ,OAAO,GAAG;EAC1B,IAAI,OAAO;EACX,KAAK,MAAM,UAAU,SACnB,IAAI,UAAU,QAAQ,OAAO,WAAW,UAAU;GAChD,MAAM,aAAc,OAA8B;GAClD,IAAI,OAAO,eAAe,UACxB,QAAQ;EAEZ;EAEF,IAAI,SAAS,IACX,OAAO;CAEX;CACA,OAAO;AACT;AAEA,SAAgB,uBACd,OACA,eACiC;CACjC,MAAM,EAAE,YAAY;CACpB,MAAM,iBAAiB,mBAAmBA,yBAAAA;CAC1C,MAAM,sBACJ,CAAC,sBAAsB,MAAM,cAAc,KAC3C,CAAC,gBAAgB,MAAM,cAAc;CAUvC,IARE,QAAQ,MAAM,IAAI,KAClB,kBACA,OAAO,QAAQ,YAAY,YAC3B,QAAQ,YAAY,MAAM,QAC1B,uBACA,CAAC,mBAAmB,OAAO,KAC3B,CAAC,kBAAkB,OAAO,GAG1B,OAAO;EACL,MAAM,MAAM;EACZ,QAAQ;EACR,OAAO,UAAU,0BAA0B,OAAO,KAAK;CACzD;CAGF,MAAM,iBACJ,QAAQ,MAAM,IAAI,KAClB,kBACA,uBACA,CAAC,mBAAmB,OAAO,KAC3B,CAAC,kBAAkB,OAAO,IACtB,sBAAsB,SAAS,MAAM,IAAI,IACzC;CACN,IAAI,kBAAkB,MACpB,OAAO;EACL,MAAM,MAAM;EACZ,QAAQ;EACR,OAAO,UACL,8BAA8B,OAAO,gBAAgB,KAAK;CAC9D;CAGF,MAAM,YACJ,MAAM,SACL,kBAAkB,iBAAiB,OAAO,cAAc,OAAO,IAAI;CACtE,IAAI,cAAc,IAChB,OAAO;EACL,MAAM;EACN,QAAQ;EACR,QAAQ;EACR,YAAY;CACd;CAGF,OAAO;EAAE,MAAM;EAAI,QAAQ;EAAO,YAAY;CAAM;AACtD;AAQA,gBAAuB,uBAAuB,EAC5C,QACA,SACA,QACA,cAMsC;CACtC,MAAM,UAAU,mBAEZ;EACF,WAAW,MAAM,SAAS,QACxB,MAAM,uBAAuB,OAAO,sBAAsB;CAE9D,EAAA,CAAG;CAEH,WAAW,MAAM,eAAeC,iBAAAA,aAAa;EAAE;EAAQ;EAAS;CAAO,CAAC,GAAG;EACzE,MAAM;EACN,MAAM,YAAY,kBAChB,YAAY,QAAQ,IACpB,KAAA,GACA,KAAA,GACA,KAAA,GACA,KAAA,GACA,EAAE,OAAO,YAAY,CACvB;CACF;AACF"}

package/dist/cjs/llm/stream/smoother.cjs

@@ -11,20 +11,7 @@ const STREAM_BOUNDARIES = new Set([
 	":"
 ]);
 const STREAM_ABORT_MESSAGE = "AbortError: User aborted the request.";
-/**
-* How long generator teardown waits for the background producer to observe a
-* consumer close before abandoning it. Well-behaved streams settle in
-* microseconds (the next enqueue throws); a stalled provider that ignores
-* aborts otherwise blocks teardown — and abort propagation — indefinitely.
-* An abandoned producer still self-terminates on its next enqueue attempt.
-*/
 const PRODUCER_CLOSE_GRACE_MS = 1e3;
-/**
-* Resolves a configured stream delay to its effective value (default 25ms;
-* 0 disables smoothing). Non-finite inputs (NaN from a malformed config
-* value, ±Infinity) normalize to the default rather than poisoning piece
-* arithmetic downstream.
-*/
 function resolveStreamDelay(delay) {
 	if (delay == null || !Number.isFinite(delay)) return 25;
 	return Math.max(0, delay);
@@ -38,27 +25,15 @@ function findStreamChunkBoundary(text, minSize) {
 	for (let position = minSize; position < scanEnd; position++) if (STREAM_BOUNDARIES.has(text[position])) return position + 1;
 	return scanEnd;
 }
-/**
-* Backlog-proportional piece sizing: emit enough per tick that the current
-* backlog drains in ~`targetLatencyMs`, so render lag stays pinned near the
-* target regardless of how fast the provider streams. Token-sized arrivals
-* never exceed the minimum piece, matching the legacy fixed-size splitter.
-*/
 function computeAdaptivePieceSize(bufferedTextLength, tickMs, targetLatencyMs = 250) {
 	if (bufferedTextLength <= 0) return 4;
 	if (tickMs <= 0 || targetLatencyMs <= 0) return bufferedTextLength;
 	return Math.max(4, Math.ceil(bufferedTextLength * tickMs / targetLatencyMs));
 }
-/**
-* A cadence, not an additive sleep: time the consumer already spent since the
-* last visible emission counts against the target delay, so slow downstream
-* handlers never compound latency.
-*/
 function getCadencedStreamDelay({ targetDelay, lastVisibleTextAt, now }) {
 	if (targetDelay <= 0 || lastVisibleTextAt == null) return 0;
 	return Math.max(0, targetDelay - (now - lastVisibleTextAt));
 }
-/** Abort-aware sleep that resolves (never rejects) on abort; callers re-check the signal. */
 async function waitForStreamDelay(delay, signal) {
 	if (delay <= 0 || isSignalAborted(signal)) return;
 	await new Promise((resolve) => {
@@ -76,23 +51,8 @@ async function waitForStreamDelay(delay, signal) {
 		if (isSignalAborted(signal)) onAbort();
 	});
 }
-/**
-* Bounded producer/consumer smoothing engine.
-*
-* The producer drains `source` eagerly into a bounded queue (the buffer is the
-* backlog measurement adaptive sizing needs); at capacity it parks, applying
-* backpressure to the underlying stream. The consumer emits paced pieces,
-* decrementing the text budget and waking the producer *before* each cadenced
-* sleep so the provider stream keeps being read during pacing.
-*
-* `delayMs <= 0` disables smoothing entirely: every item passes through FIFO,
-* unsplit and undelayed.
-*/
 async function* smoothStream({ source, delayMs, signal, abortUpstream }) {
 	if (!(delayMs > 0)) {
-		/** Disabled smoothing preserves fully lazy streaming: no background
-		* producer, no read-ahead — each provider chunk is pulled only when the
-		* consumer asks, exactly like the pre-engine pass-through paths. */
 		for await (const item of source) {
 			if (isSignalAborted(signal)) {
 				abortUpstream?.();
@@ -127,9 +87,6 @@ async function* smoothStream({ source, delayMs, signal, abortUpstream }) {
 	const hasQueuedItems = () => queuedItemIndex < queuedItems.length;
 	const getQueuedItemCount = () => queuedItems.length - queuedItemIndex;
 	const isQueueAtCapacity = () => getQueuedItemCount() >= 256 || bufferedTextLength >= 8192;
-	/** Abort-aware: a consumer parked on an empty queue must wake when the
-	* signal fires even if the provider stream never honors the abort — the
-	* loop's top-of-iteration check then throws the canonical error. */
 	const waitForNextItem = async () => {
 		if (hasQueuedItems() || producerState.done || producerState.failed || isSignalAborted(signal)) return;
 		await new Promise((resolve) => {
@@ -185,14 +142,6 @@ async function* smoothStream({ source, delayMs, signal, abortUpstream }) {
 		bufferedTextLength += textLength;
 		notifyConsumerForItem();
 	};
-	/**
-	* Oversized splittable items are segmented at admission so a single giant
-	* provider chunk cannot blow past the text budget: each segment re-checks
-	* capacity, so the producer parks mid-chunk once the buffer fills — the
-	* same bound the legacy split-before-enqueue queues enforced. The wrapped
-	* emit maps segment-local pieces back to chunk-global isFirst/isLast so
-	* provider clone contracts are unaffected.
-	*/
 	const enqueueSegmented = async (item) => {
 		if (!item.smooth || item.atomic === true || item.text.length <= 4096) {
 			await enqueue(item);
@@ -272,9 +221,6 @@ async function* smoothStream({ source, delayMs, signal, abortUpstream }) {
 				current = void 0;
 				continue;
 			}
-			/** Once the producer is done the backlog is final: drain it linearly
-			* across the remaining target window instead of letting the
-			* proportional formula decay geometrically and stretch the tail. */
 			if (producerState.done && drainTicksRemaining == null) drainTicksRemaining = Math.max(1, Math.floor(250 / delayMs));
 			const tickBudget = drainTicksRemaining != null ? Math.max(4, Math.ceil(bufferedTextLength / drainTicksRemaining)) : computeAdaptivePieceSize(bufferedTextLength, delayMs);
 			if (drainTicksRemaining != null && drainTicksRemaining > 1) drainTicksRemaining -= 1;
@@ -297,10 +243,6 @@ async function* smoothStream({ source, delayMs, signal, abortUpstream }) {
 				});
 				continue;
 			}
-			/** One cadence tick drains up to the adaptive budget ACROSS queued
-			* items, so token-sized provider deltas coalesce instead of costing a
-			* full tick each; passthrough items flush free mid-batch (FIFO), and
-			* atomic items end the batch to take their own tick. */
 			let consumed = 0;
 			while (consumed < tickBudget) {
 				if (isSignalAborted(signal)) throwAborted();

package/dist/cjs/llm/stream/smoother.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"smoother.cjs","names":["iterator"],"sources":["../../../../src/llm/stream/smoother.ts"],"sourcesContent":["export const DEFAULT_STREAM_DELAY = 25;\nexport const SMOOTH_TARGET_LATENCY_MS = 250;\nexport const MAX_STREAM_QUEUE_CHUNKS = 256;\nexport const MAX_STREAM_QUEUE_TEXT_CHARS = 8192;\nexport const MAX_SMOOTH_ITEM_SEGMENT_CHARS = 4096;\nexport const STREAM_CHUNK_MIN_SIZE = 4;\nexport const STREAM_BOUNDARIES: ReadonlySet<string> = new Set([\n  ' ',\n  '.',\n  ',',\n  '!',\n  '?',\n  ';',\n  ':',\n]);\n\nexport const STREAM_ABORT_MESSAGE = 'AbortError: User aborted the request.';\nexport const STREAM_PRODUCER_FAILURE = 'Stream producer failed.';\n\n/**\n * How long generator teardown waits for the background producer to observe a\n * consumer close before abandoning it. Well-behaved streams settle in\n * microseconds (the next enqueue throws); a stalled provider that ignores\n * aborts otherwise blocks teardown — and abort propagation — indefinitely.\n * An abandoned producer still self-terminates on its next enqueue attempt.\n */\nexport const PRODUCER_CLOSE_GRACE_MS = 1000;\n\n/**\n * Resolves a configured stream delay to its effective value (default 25ms;\n * 0 disables smoothing). Non-finite inputs (NaN from a malformed config\n * value, ±Infinity) normalize to the default rather than poisoning piece\n * arithmetic downstream.\n */\nexport function resolveStreamDelay(delay?: number): number {\n  if (delay == null || !Number.isFinite(delay)) {\n    return DEFAULT_STREAM_DELAY;\n  }\n  return Math.max(0, delay);\n}\n\nexport function isSignalAborted(signal?: AbortSignal): boolean {\n  return signal?.aborted === true;\n}\n\n/**\n * How far past the target size the word-boundary search may extend before\n * hard-cutting. Natural language hits a boundary within a few characters;\n * boundary-free runs (base64, minified data, long identifiers) must not\n * stretch a piece — or an admission segment — arbitrarily far past its\n * budget.\n */\nexport const STREAM_BOUNDARY_LOOKAHEAD_CHARS = 64;\n\nexport function findStreamChunkBoundary(\n  text: string,\n  minSize: number\n): number {\n  if (minSize >= text.length) {\n    return text.length;\n  }\n\n  const scanEnd = Math.min(\n    text.length,\n    minSize + STREAM_BOUNDARY_LOOKAHEAD_CHARS\n  );\n  for (let position = minSize; position < scanEnd; position++) {\n    if (STREAM_BOUNDARIES.has(text[position])) {\n      return position + 1;\n    }\n  }\n\n  return scanEnd;\n}\n\n/**\n * Backlog-proportional piece sizing: emit enough per tick that the current\n * backlog drains in ~`targetLatencyMs`, so render lag stays pinned near the\n * target regardless of how fast the provider streams. Token-sized arrivals\n * never exceed the minimum piece, matching the legacy fixed-size splitter.\n */\nexport function computeAdaptivePieceSize(\n  bufferedTextLength: number,\n  tickMs: number,\n  targetLatencyMs: number = SMOOTH_TARGET_LATENCY_MS\n): number {\n  if (bufferedTextLength <= 0) {\n    return STREAM_CHUNK_MIN_SIZE;\n  }\n  if (tickMs <= 0 || targetLatencyMs <= 0) {\n    return bufferedTextLength;\n  }\n  return Math.max(\n    STREAM_CHUNK_MIN_SIZE,\n    Math.ceil((bufferedTextLength * tickMs) / targetLatencyMs)\n  );\n}\n\n/**\n * A cadence, not an additive sleep: time the consumer already spent since the\n * last visible emission counts against the target delay, so slow downstream\n * handlers never compound latency.\n */\nexport function getCadencedStreamDelay({\n  targetDelay,\n  lastVisibleTextAt,\n  now,\n}: {\n  targetDelay: number;\n  lastVisibleTextAt?: number;\n  now: number;\n}): number {\n  if (targetDelay <= 0 || lastVisibleTextAt == null) {\n    return 0;\n  }\n  return Math.max(0, targetDelay - (now - lastVisibleTextAt));\n}\n\n/** Abort-aware sleep that resolves (never rejects) on abort; callers re-check the signal. */\nexport async function waitForStreamDelay(\n  delay: number,\n  signal?: AbortSignal\n): Promise<void> {\n  if (delay <= 0 || isSignalAborted(signal)) {\n    return;\n  }\n  await new Promise<void>((resolve) => {\n    const timeoutRef: { current?: ReturnType<typeof setTimeout> } = {};\n    const onAbort = (): void => {\n      if (timeoutRef.current) {\n        clearTimeout(timeoutRef.current);\n      }\n      signal?.removeEventListener('abort', onAbort);\n      resolve();\n    };\n    timeoutRef.current = setTimeout(() => {\n      signal?.removeEventListener('abort', onAbort);\n      resolve();\n    }, delay);\n    signal?.addEventListener('abort', onAbort, { once: true });\n    if (isSignalAborted(signal)) {\n      onAbort();\n    }\n  });\n}\n\nexport type SmoothPiece = {\n  text: string;\n  isFirst: boolean;\n  isLast: boolean;\n};\n\n/**\n * One classified unit of provider stream output.\n *\n * - `smooth: true` — visible text, paced at the configured cadence and (unless\n *   `atomic`) sliced adaptively at dequeue time.\n * - `atomic: true` — paced as a single piece, never split (text-bearing chunks\n *   whose metadata cannot survive slicing, e.g. logprobs / finish_reason).\n * - `smooth: false` — passthrough: tool-call deltas, usage-only, id-only and\n *   seal chunks. Zero delay, strict FIFO with the text around them.\n *\n * `emit` builds the provider-specific output for one piece; `isFirst` lets\n * providers keep usage_metadata on only the first piece of a split.\n */\nexport type SmoothItem<TEmit> = {\n  text: string;\n  smooth: boolean;\n  atomic?: boolean;\n  emit: (piece: SmoothPiece) => TEmit;\n};\n\ntype ProducerState = {\n  done: boolean;\n  failed: boolean;\n  error?: unknown;\n};\n\ntype QueuedSmoothItem<TEmit> = {\n  item: SmoothItem<TEmit>;\n  textLength: number;\n};\n\n/**\n * Bounded producer/consumer smoothing engine.\n *\n * The producer drains `source` eagerly into a bounded queue (the buffer is the\n * backlog measurement adaptive sizing needs); at capacity it parks, applying\n * backpressure to the underlying stream. The consumer emits paced pieces,\n * decrementing the text budget and waking the producer *before* each cadenced\n * sleep so the provider stream keeps being read during pacing.\n *\n * `delayMs <= 0` disables smoothing entirely: every item passes through FIFO,\n * unsplit and undelayed.\n */\nexport async function* smoothStream<TEmit>({\n  source,\n  delayMs,\n  signal,\n  abortUpstream,\n}: {\n  source: AsyncIterable<SmoothItem<TEmit>>;\n  delayMs: number;\n  signal?: AbortSignal;\n  abortUpstream?: () => void;\n}): AsyncGenerator<TEmit> {\n  if (!(delayMs > 0)) {\n    /** Disabled smoothing preserves fully lazy streaming: no background\n     * producer, no read-ahead — each provider chunk is pulled only when the\n     * consumer asks, exactly like the pre-engine pass-through paths. */\n    for await (const item of source) {\n      if (isSignalAborted(signal)) {\n        abortUpstream?.();\n        throw new Error(STREAM_ABORT_MESSAGE);\n      }\n      yield item.emit({ text: item.text, isFirst: true, isLast: true });\n    }\n    return;\n  }\n\n  const queuedItems: QueuedSmoothItem<TEmit>[] = [];\n  const producerState: ProducerState = { done: false, failed: false };\n  let queuedItemIndex = 0;\n  let bufferedTextLength = 0;\n  let consumerClosed = false;\n  let notifyConsumer: (() => void) | undefined;\n  let notifyProducer: (() => void) | undefined;\n\n  const notifyConsumerForItem = (): void => {\n    notifyConsumer?.();\n    notifyConsumer = undefined;\n  };\n\n  const notifyProducerForSpace = (): void => {\n    notifyProducer?.();\n    notifyProducer = undefined;\n  };\n\n  const hasQueuedItems = (): boolean => queuedItemIndex < queuedItems.length;\n\n  const getQueuedItemCount = (): number =>\n    queuedItems.length - queuedItemIndex;\n\n  const isQueueAtCapacity = (): boolean =>\n    getQueuedItemCount() >= MAX_STREAM_QUEUE_CHUNKS ||\n    bufferedTextLength >= MAX_STREAM_QUEUE_TEXT_CHARS;\n\n  /** Abort-aware: a consumer parked on an empty queue must wake when the\n   * signal fires even if the provider stream never honors the abort — the\n   * loop's top-of-iteration check then throws the canonical error. */\n  const waitForNextItem = async (): Promise<void> => {\n    if (\n      hasQueuedItems() ||\n      producerState.done ||\n      producerState.failed ||\n      isSignalAborted(signal)\n    ) {\n      return;\n    }\n    await new Promise<void>((resolve) => {\n      const onAbort = (): void => {\n        signal?.removeEventListener('abort', onAbort);\n        resolve();\n      };\n      notifyConsumer = (): void => {\n        signal?.removeEventListener('abort', onAbort);\n        resolve();\n      };\n      signal?.addEventListener('abort', onAbort, { once: true });\n      if (isSignalAborted(signal)) {\n        onAbort();\n      }\n    });\n  };\n\n  const waitForQueueSpace = async (): Promise<void> => {\n    while (\n      isQueueAtCapacity() &&\n      !consumerClosed &&\n      !isSignalAborted(signal)\n    ) {\n      await new Promise<void>((resolve) => {\n        const onAbort = (): void => {\n          signal?.removeEventListener('abort', onAbort);\n          resolve();\n        };\n        const onSpace = (): void => {\n          signal?.removeEventListener('abort', onAbort);\n          resolve();\n        };\n        notifyProducer = onSpace;\n        signal?.addEventListener('abort', onAbort, { once: true });\n        if (isSignalAborted(signal)) {\n          onAbort();\n        }\n      });\n    }\n  };\n\n  const dequeue = (): QueuedSmoothItem<TEmit> | undefined => {\n    if (!hasQueuedItems()) {\n      return undefined;\n    }\n    const queuedItem = queuedItems[queuedItemIndex];\n    queuedItemIndex++;\n    if (queuedItemIndex > 128 && queuedItemIndex * 2 >= queuedItems.length) {\n      queuedItems.splice(0, queuedItemIndex);\n      queuedItemIndex = 0;\n    }\n    return queuedItem;\n  };\n\n  const throwAborted = (): never => {\n    abortUpstream?.();\n    throw new Error(STREAM_ABORT_MESSAGE);\n  };\n\n  const enqueue = async (item: SmoothItem<TEmit>): Promise<void> => {\n    await waitForQueueSpace();\n    if (consumerClosed || isSignalAborted(signal)) {\n      throwAborted();\n    }\n    const textLength = item.smooth ? item.text.length : 0;\n    queuedItems.push({ item, textLength });\n    bufferedTextLength += textLength;\n    notifyConsumerForItem();\n  };\n\n  /**\n   * Oversized splittable items are segmented at admission so a single giant\n   * provider chunk cannot blow past the text budget: each segment re-checks\n   * capacity, so the producer parks mid-chunk once the buffer fills — the\n   * same bound the legacy split-before-enqueue queues enforced. The wrapped\n   * emit maps segment-local pieces back to chunk-global isFirst/isLast so\n   * provider clone contracts are unaffected.\n   */\n  const enqueueSegmented = async (item: SmoothItem<TEmit>): Promise<void> => {\n    if (\n      !item.smooth ||\n      item.atomic === true ||\n      item.text.length <= MAX_SMOOTH_ITEM_SEGMENT_CHARS\n    ) {\n      await enqueue(item);\n      return;\n    }\n\n    const segments: { start: number; end: number }[] = [];\n    let offset = 0;\n    while (offset < item.text.length) {\n      const end =\n        offset +\n        findStreamChunkBoundary(\n          item.text.slice(offset),\n          MAX_SMOOTH_ITEM_SEGMENT_CHARS\n        );\n      segments.push({ start: offset, end });\n      offset = end;\n    }\n\n    for (let i = 0; i < segments.length; i++) {\n      const isFirstSegment = i === 0;\n      const isLastSegment = i === segments.length - 1;\n      await enqueue({\n        text: item.text.slice(segments[i].start, segments[i].end),\n        smooth: true,\n        emit: (piece) =>\n          item.emit({\n            text: piece.text,\n            isFirst: isFirstSegment && piece.isFirst,\n            isLast: isLastSegment && piece.isLast,\n          }),\n      });\n    }\n  };\n\n  const producer = (async (): Promise<void> => {\n    try {\n      for await (const item of source) {\n        if (isSignalAborted(signal)) {\n          throwAborted();\n        }\n        await enqueueSegmented(item);\n      }\n    } catch (error) {\n      producerState.failed = true;\n      producerState.error = error;\n    } finally {\n      producerState.done = true;\n      notifyConsumerForItem();\n    }\n  })();\n\n  let hasEmittedText = false;\n  let lastVisibleTextAt: number | undefined;\n  let drainTicksRemaining: number | undefined;\n  let current: QueuedSmoothItem<TEmit> | undefined;\n  let headOffset = 0;\n  let keepStreaming = true;\n  try {\n    while (keepStreaming) {\n      if (isSignalAborted(signal)) {\n        throwAborted();\n      }\n\n      if (current == null) {\n        await waitForNextItem();\n        current = dequeue();\n        headOffset = 0;\n      }\n\n      if (current == null) {\n        if (producerState.failed) {\n          throw producerState.error ?? new Error(STREAM_PRODUCER_FAILURE);\n        }\n        if (producerState.done) {\n          keepStreaming = false;\n        }\n        continue;\n      }\n\n      const { item } = current;\n\n      if (!item.smooth) {\n        notifyProducerForSpace();\n        current = undefined;\n        yield item.emit({ text: item.text, isFirst: true, isLast: true });\n        continue;\n      }\n\n      if (item.text === '') {\n        bufferedTextLength = Math.max(\n          0,\n          bufferedTextLength - current.textLength\n        );\n        notifyProducerForSpace();\n        current = undefined;\n        continue;\n      }\n\n      /** Once the producer is done the backlog is final: drain it linearly\n       * across the remaining target window instead of letting the\n       * proportional formula decay geometrically and stretch the tail. */\n      if (producerState.done && drainTicksRemaining == null) {\n        drainTicksRemaining = Math.max(\n          1,\n          Math.floor(SMOOTH_TARGET_LATENCY_MS / delayMs)\n        );\n      }\n      const tickBudget =\n        drainTicksRemaining != null\n          ? Math.max(\n            STREAM_CHUNK_MIN_SIZE,\n            Math.ceil(bufferedTextLength / drainTicksRemaining)\n          )\n          : computeAdaptivePieceSize(bufferedTextLength, delayMs);\n      if (drainTicksRemaining != null && drainTicksRemaining > 1) {\n        drainTicksRemaining -= 1;\n      }\n\n      await waitForStreamDelay(\n        getCadencedStreamDelay({\n          targetDelay: hasEmittedText ? delayMs : 0,\n          lastVisibleTextAt,\n          now: Date.now(),\n        }),\n        signal\n      );\n      if (isSignalAborted(signal)) {\n        throwAborted();\n      }\n      hasEmittedText = true;\n      lastVisibleTextAt = Date.now();\n\n      if (item.atomic === true) {\n        bufferedTextLength = Math.max(\n          0,\n          bufferedTextLength - current.textLength\n        );\n        notifyProducerForSpace();\n        current = undefined;\n        yield item.emit({ text: item.text, isFirst: true, isLast: true });\n        continue;\n      }\n\n      /** One cadence tick drains up to the adaptive budget ACROSS queued\n       * items, so token-sized provider deltas coalesce instead of costing a\n       * full tick each; passthrough items flush free mid-batch (FIFO), and\n       * atomic items end the batch to take their own tick. */\n      let consumed = 0;\n      while (consumed < tickBudget) {\n        if (isSignalAborted(signal)) {\n          throwAborted();\n        }\n        if (current == null) {\n          if (!hasQueuedItems()) {\n            break;\n          }\n          current = dequeue();\n          headOffset = 0;\n          if (current == null) {\n            break;\n          }\n        }\n\n        const batchItem = current.item;\n        if (!batchItem.smooth) {\n          notifyProducerForSpace();\n          current = undefined;\n          yield batchItem.emit({\n            text: batchItem.text,\n            isFirst: true,\n            isLast: true,\n          });\n          continue;\n        }\n        if (batchItem.text === '') {\n          bufferedTextLength = Math.max(\n            0,\n            bufferedTextLength - current.textLength\n          );\n          notifyProducerForSpace();\n          current = undefined;\n          continue;\n        }\n        if (batchItem.atomic === true) {\n          break;\n        }\n\n        const remainingText = batchItem.text.slice(headOffset);\n        const pieceLength = findStreamChunkBoundary(\n          remainingText,\n          tickBudget - consumed\n        );\n        const pieceEnd = headOffset + pieceLength;\n        const piece = batchItem.text.slice(headOffset, pieceEnd);\n        const isFirst = headOffset === 0;\n        const isLast = pieceEnd === batchItem.text.length;\n\n        bufferedTextLength = Math.max(0, bufferedTextLength - piece.length);\n        notifyProducerForSpace();\n        if (isLast) {\n          current = undefined;\n        } else {\n          headOffset = pieceEnd;\n        }\n        consumed += piece.length;\n        yield batchItem.emit({ text: piece, isFirst, isLast });\n      }\n    }\n  } finally {\n    consumerClosed = true;\n    if (producerState.done) {\n      await producer;\n    } else {\n      abortUpstream?.();\n      notifyProducerForSpace();\n      const iterator = source as Partial<AsyncGenerator<SmoothItem<TEmit>>>;\n      const closing = iterator.return?.call(source, undefined as never);\n      if (closing != null) {\n        void closing.then(\n          () => undefined,\n          () => undefined\n        );\n      }\n      await Promise.race([\n        producer,\n        new Promise<void>((resolve) => {\n          const timeout = setTimeout(resolve, PRODUCER_CLOSE_GRACE_MS);\n          timeout.unref();\n        }),\n      ]);\n    }\n  }\n}\n"],"mappings":";AAAA,MAAa,uBAAuB;AAIpC,MAAa,gCAAgC;AAE7C,MAAa,oBAAyC,IAAI,IAAI;CAC5D;CACA;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,MAAa,uBAAuB;;;;;;;;AAUpC,MAAa,0BAA0B;;;;;;;AAQvC,SAAgB,mBAAmB,OAAwB;CACzD,IAAI,SAAS,QAAQ,CAAC,OAAO,SAAS,KAAK,GACzC,OAAA;CAEF,OAAO,KAAK,IAAI,GAAG,KAAK;AAC1B;AAEA,SAAgB,gBAAgB,QAA+B;CAC7D,OAAO,QAAQ,YAAY;AAC7B;AAWA,SAAgB,wBACd,MACA,SACQ;CACR,IAAI,WAAW,KAAK,QAClB,OAAO,KAAK;CAGd,MAAM,UAAU,KAAK,IACnB,KAAK,QACL,UAAA,EACF;CACA,KAAK,IAAI,WAAW,SAAS,WAAW,SAAS,YAC/C,IAAI,kBAAkB,IAAI,KAAK,SAAS,GACtC,OAAO,WAAW;CAItB,OAAO;AACT;;;;;;;AAQA,SAAgB,yBACd,oBACA,QACA,kBAAA,KACQ;CACR,IAAI,sBAAsB,GACxB,OAAA;CAEF,IAAI,UAAU,KAAK,mBAAmB,GACpC,OAAO;CAET,OAAO,KAAK,IAAA,GAEV,KAAK,KAAM,qBAAqB,SAAU,eAAe,CAC3D;AACF;;;;;;AAOA,SAAgB,uBAAuB,EACrC,aACA,mBACA,OAKS;CACT,IAAI,eAAe,KAAK,qBAAqB,MAC3C,OAAO;CAET,OAAO,KAAK,IAAI,GAAG,eAAe,MAAM,kBAAkB;AAC5D;;AAGA,eAAsB,mBACpB,OACA,QACe;CACf,IAAI,SAAS,KAAK,gBAAgB,MAAM,GACtC;CAEF,MAAM,IAAI,SAAe,YAAY;EACnC,MAAM,aAA0D,CAAC;EACjE,MAAM,gBAAsB;GAC1B,IAAI,WAAW,SACb,aAAa,WAAW,OAAO;GAEjC,QAAQ,oBAAoB,SAAS,OAAO;GAC5C,QAAQ;EACV;EACA,WAAW,UAAU,iBAAiB;GACpC,QAAQ,oBAAoB,SAAS,OAAO;GAC5C,QAAQ;EACV,GAAG,KAAK;EACR,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;EACzD,IAAI,gBAAgB,MAAM,GACxB,QAAQ;CAEZ,CAAC;AACH;;;;;;;;;;;;;AAmDA,gBAAuB,aAAoB,EACzC,QACA,SACA,QACA,iBAMwB;CACxB,IAAI,EAAE,UAAU,IAAI;;;;EAIlB,WAAW,MAAM,QAAQ,QAAQ;GAC/B,IAAI,gBAAgB,MAAM,GAAG;IAC3B,gBAAgB;IAChB,MAAM,IAAI,MAAM,oBAAoB;GACtC;GACA,MAAM,KAAK,KAAK;IAAE,MAAM,KAAK;IAAM,SAAS;IAAM,QAAQ;GAAK,CAAC;EAClE;EACA;CACF;CAEA,MAAM,cAAyC,CAAC;CAChD,MAAM,gBAA+B;EAAE,MAAM;EAAO,QAAQ;CAAM;CAClE,IAAI,kBAAkB;CACtB,IAAI,qBAAqB;CACzB,IAAI,iBAAiB;CACrB,IAAI;CACJ,IAAI;CAEJ,MAAM,8BAAoC;EACxC,iBAAiB;EACjB,iBAAiB,KAAA;CACnB;CAEA,MAAM,+BAAqC;EACzC,iBAAiB;EACjB,iBAAiB,KAAA;CACnB;CAEA,MAAM,uBAAgC,kBAAkB,YAAY;CAEpE,MAAM,2BACJ,YAAY,SAAS;CAEvB,MAAM,0BACJ,mBAAmB,KAAA,OACnB,sBAAA;;;;CAKF,MAAM,kBAAkB,YAA2B;EACjD,IACE,eAAe,KACf,cAAc,QACd,cAAc,UACd,gBAAgB,MAAM,GAEtB;EAEF,MAAM,IAAI,SAAe,YAAY;GACnC,MAAM,gBAAsB;IAC1B,QAAQ,oBAAoB,SAAS,OAAO;IAC5C,QAAQ;GACV;GACA,uBAA6B;IAC3B,QAAQ,oBAAoB,SAAS,OAAO;IAC5C,QAAQ;GACV;GACA,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;GACzD,IAAI,gBAAgB,MAAM,GACxB,QAAQ;EAEZ,CAAC;CACH;CAEA,MAAM,oBAAoB,YAA2B;EACnD,OACE,kBAAkB,KAClB,CAAC,kBACD,CAAC,gBAAgB,MAAM,GAEvB,MAAM,IAAI,SAAe,YAAY;GACnC,MAAM,gBAAsB;IAC1B,QAAQ,oBAAoB,SAAS,OAAO;IAC5C,QAAQ;GACV;GACA,MAAM,gBAAsB;IAC1B,QAAQ,oBAAoB,SAAS,OAAO;IAC5C,QAAQ;GACV;GACA,iBAAiB;GACjB,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;GACzD,IAAI,gBAAgB,MAAM,GACxB,QAAQ;EAEZ,CAAC;CAEL;CAEA,MAAM,gBAAqD;EACzD,IAAI,CAAC,eAAe,GAClB;EAEF,MAAM,aAAa,YAAY;EAC/B;EACA,IAAI,kBAAkB,OAAO,kBAAkB,KAAK,YAAY,QAAQ;GACtE,YAAY,OAAO,GAAG,eAAe;GACrC,kBAAkB;EACpB;EACA,OAAO;CACT;CAEA,MAAM,qBAA4B;EAChC,gBAAgB;EAChB,MAAM,IAAI,MAAM,oBAAoB;CACtC;CAEA,MAAM,UAAU,OAAO,SAA2C;EAChE,MAAM,kBAAkB;EACxB,IAAI,kBAAkB,gBAAgB,MAAM,GAC1C,aAAa;EAEf,MAAM,aAAa,KAAK,SAAS,KAAK,KAAK,SAAS;EACpD,YAAY,KAAK;GAAE;GAAM;EAAW,CAAC;EACrC,sBAAsB;EACtB,sBAAsB;CACxB;;;;;;;;;CAUA,MAAM,mBAAmB,OAAO,SAA2C;EACzE,IACE,CAAC,KAAK,UACN,KAAK,WAAW,QAChB,KAAK,KAAK,UAAA,MACV;GACA,MAAM,QAAQ,IAAI;GAClB;EACF;EAEA,MAAM,WAA6C,CAAC;EACpD,IAAI,SAAS;EACb,OAAO,SAAS,KAAK,KAAK,QAAQ;GAChC,MAAM,MACJ,SACA,wBACE,KAAK,KAAK,MAAM,MAAM,GACtB,6BACF;GACF,SAAS,KAAK;IAAE,OAAO;IAAQ;GAAI,CAAC;GACpC,SAAS;EACX;EAEA,KAAK,IAAI,IAAI,GAAG,IAAI,SAAS,QAAQ,KAAK;GACxC,MAAM,iBAAiB,MAAM;GAC7B,MAAM,gBAAgB,MAAM,SAAS,SAAS;GAC9C,MAAM,QAAQ;IACZ,MAAM,KAAK,KAAK,MAAM,SAAS,EAAE,CAAC,OAAO,SAAS,EAAE,CAAC,GAAG;IACxD,QAAQ;IACR,OAAO,UACL,KAAK,KAAK;KACR,MAAM,MAAM;KACZ,SAAS,kBAAkB,MAAM;KACjC,QAAQ,iBAAiB,MAAM;IACjC,CAAC;GACL,CAAC;EACH;CACF;CAEA,MAAM,YAAY,YAA2B;EAC3C,IAAI;GACF,WAAW,MAAM,QAAQ,QAAQ;IAC/B,IAAI,gBAAgB,MAAM,GACxB,aAAa;IAEf,MAAM,iBAAiB,IAAI;GAC7B;EACF,SAAS,OAAO;GACd,cAAc,SAAS;GACvB,cAAc,QAAQ;EACxB,UAAU;GACR,cAAc,OAAO;GACrB,sBAAsB;EACxB;CACF,EAAA,CAAG;CAEH,IAAI,iBAAiB;CACrB,IAAI;CACJ,IAAI;CACJ,IAAI;CACJ,IAAI,aAAa;CACjB,IAAI,gBAAgB;CACpB,IAAI;EACF,OAAO,eAAe;GACpB,IAAI,gBAAgB,MAAM,GACxB,aAAa;GAGf,IAAI,WAAW,MAAM;IACnB,MAAM,gBAAgB;IACtB,UAAU,QAAQ;IAClB,aAAa;GACf;GAEA,IAAI,WAAW,MAAM;IACnB,IAAI,cAAc,QAChB,MAAM,cAAc,yBAAS,IAAI,MAAA,yBAA6B;IAEhE,IAAI,cAAc,MAChB,gBAAgB;IAElB;GACF;GAEA,MAAM,EAAE,SAAS;GAEjB,IAAI,CAAC,KAAK,QAAQ;IAChB,uBAAuB;IACvB,UAAU,KAAA;IACV,MAAM,KAAK,KAAK;KAAE,MAAM,KAAK;KAAM,SAAS;KAAM,QAAQ;IAAK,CAAC;IAChE;GACF;GAEA,IAAI,KAAK,SAAS,IAAI;IACpB,qBAAqB,KAAK,IACxB,GACA,qBAAqB,QAAQ,UAC/B;IACA,uBAAuB;IACvB,UAAU,KAAA;IACV;GACF;;;;GAKA,IAAI,cAAc,QAAQ,uBAAuB,MAC/C,sBAAsB,KAAK,IACzB,GACA,KAAK,MAAA,MAAiC,OAAO,CAC/C;GAEF,MAAM,aACJ,uBAAuB,OACnB,KAAK,IAAA,GAEL,KAAK,KAAK,qBAAqB,mBAAmB,CACpD,IACE,yBAAyB,oBAAoB,OAAO;GAC1D,IAAI,uBAAuB,QAAQ,sBAAsB,GACvD,uBAAuB;GAGzB,MAAM,mBACJ,uBAAuB;IACrB,aAAa,iBAAiB,UAAU;IACxC;IACA,KAAK,KAAK,IAAI;GAChB,CAAC,GACD,MACF;GACA,IAAI,gBAAgB,MAAM,GACxB,aAAa;GAEf,iBAAiB;GACjB,oBAAoB,KAAK,IAAI;GAE7B,IAAI,KAAK,WAAW,MAAM;IACxB,qBAAqB,KAAK,IACxB,GACA,qBAAqB,QAAQ,UAC/B;IACA,uBAAuB;IACvB,UAAU,KAAA;IACV,MAAM,KAAK,KAAK;KAAE,MAAM,KAAK;KAAM,SAAS;KAAM,QAAQ;IAAK,CAAC;IAChE;GACF;;;;;GAMA,IAAI,WAAW;GACf,OAAO,WAAW,YAAY;IAC5B,IAAI,gBAAgB,MAAM,GACxB,aAAa;IAEf,IAAI,WAAW,MAAM;KACnB,IAAI,CAAC,eAAe,GAClB;KAEF,UAAU,QAAQ;KAClB,aAAa;KACb,IAAI,WAAW,MACb;IAEJ;IAEA,MAAM,YAAY,QAAQ;IAC1B,IAAI,CAAC,UAAU,QAAQ;KACrB,uBAAuB;KACvB,UAAU,KAAA;KACV,MAAM,UAAU,KAAK;MACnB,MAAM,UAAU;MAChB,SAAS;MACT,QAAQ;KACV,CAAC;KACD;IACF;IACA,IAAI,UAAU,SAAS,IAAI;KACzB,qBAAqB,KAAK,IACxB,GACA,qBAAqB,QAAQ,UAC/B;KACA,uBAAuB;KACvB,UAAU,KAAA;KACV;IACF;IACA,IAAI,UAAU,WAAW,MACvB;IAIF,MAAM,cAAc,wBADE,UAAU,KAAK,MAAM,UAE7B,GACZ,aAAa,QACf;IACA,MAAM,WAAW,aAAa;IAC9B,MAAM,QAAQ,UAAU,KAAK,MAAM,YAAY,QAAQ;IACvD,MAAM,UAAU,eAAe;IAC/B,MAAM,SAAS,aAAa,UAAU,KAAK;IAE3C,qBAAqB,KAAK,IAAI,GAAG,qBAAqB,MAAM,MAAM;IAClE,uBAAuB;IACvB,IAAI,QACF,UAAU,KAAA;SAEV,aAAa;IAEf,YAAY,MAAM;IAClB,MAAM,UAAU,KAAK;KAAE,MAAM;KAAO;KAAS;IAAO,CAAC;GACvD;EACF;CACF,UAAU;EACR,iBAAiB;EACjB,IAAI,cAAc,MAChB,MAAM;OACD;GACL,gBAAgB;GAChB,uBAAuB;GAEvB,MAAM,UAAUA,OAAS,QAAQ,KAAK,QAAQ,KAAA,CAAkB;GAChE,IAAI,WAAW,MACb,QAAa,WACL,KAAA,SACA,KAAA,CACR;GAEF,MAAM,QAAQ,KAAK,CACjB,UACA,IAAI,SAAe,YAAY;IAE7B,WAD2B,SAAS,uBAC9B,CAAC,CAAC,MAAM;GAChB,CAAC,CACH,CAAC;EACH;CACF;AACF"}
+{"version":3,"file":"smoother.cjs","names":["iterator"],"sources":["../../../../src/llm/stream/smoother.ts"],"mappings":";AAAA,MAAa,uBAAuB;AAIpC,MAAa,gCAAgC;AAE7C,MAAa,oBAAyC,IAAI,IAAI;CAC5D;CACA;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,MAAa,uBAAuB;AAUpC,MAAa,0BAA0B;AAQvC,SAAgB,mBAAmB,OAAwB;CACzD,IAAI,SAAS,QAAQ,CAAC,OAAO,SAAS,KAAK,GACzC,OAAA;CAEF,OAAO,KAAK,IAAI,GAAG,KAAK;AAC1B;AAEA,SAAgB,gBAAgB,QAA+B;CAC7D,OAAO,QAAQ,YAAY;AAC7B;AAWA,SAAgB,wBACd,MACA,SACQ;CACR,IAAI,WAAW,KAAK,QAClB,OAAO,KAAK;CAGd,MAAM,UAAU,KAAK,IACnB,KAAK,QACL,UAAA,EACF;CACA,KAAK,IAAI,WAAW,SAAS,WAAW,SAAS,YAC/C,IAAI,kBAAkB,IAAI,KAAK,SAAS,GACtC,OAAO,WAAW;CAItB,OAAO;AACT;AAQA,SAAgB,yBACd,oBACA,QACA,kBAAA,KACQ;CACR,IAAI,sBAAsB,GACxB,OAAA;CAEF,IAAI,UAAU,KAAK,mBAAmB,GACpC,OAAO;CAET,OAAO,KAAK,IAAA,GAEV,KAAK,KAAM,qBAAqB,SAAU,eAAe,CAC3D;AACF;AAOA,SAAgB,uBAAuB,EACrC,aACA,mBACA,OAKS;CACT,IAAI,eAAe,KAAK,qBAAqB,MAC3C,OAAO;CAET,OAAO,KAAK,IAAI,GAAG,eAAe,MAAM,kBAAkB;AAC5D;AAGA,eAAsB,mBACpB,OACA,QACe;CACf,IAAI,SAAS,KAAK,gBAAgB,MAAM,GACtC;CAEF,MAAM,IAAI,SAAe,YAAY;EACnC,MAAM,aAA0D,CAAC;EACjE,MAAM,gBAAsB;GAC1B,IAAI,WAAW,SACb,aAAa,WAAW,OAAO;GAEjC,QAAQ,oBAAoB,SAAS,OAAO;GAC5C,QAAQ;EACV;EACA,WAAW,UAAU,iBAAiB;GACpC,QAAQ,oBAAoB,SAAS,OAAO;GAC5C,QAAQ;EACV,GAAG,KAAK;EACR,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;EACzD,IAAI,gBAAgB,MAAM,GACxB,QAAQ;CAEZ,CAAC;AACH;AAmDA,gBAAuB,aAAoB,EACzC,QACA,SACA,QACA,iBAMwB;CACxB,IAAI,EAAE,UAAU,IAAI;EAIlB,WAAW,MAAM,QAAQ,QAAQ;GAC/B,IAAI,gBAAgB,MAAM,GAAG;IAC3B,gBAAgB;IAChB,MAAM,IAAI,MAAM,oBAAoB;GACtC;GACA,MAAM,KAAK,KAAK;IAAE,MAAM,KAAK;IAAM,SAAS;IAAM,QAAQ;GAAK,CAAC;EAClE;EACA;CACF;CAEA,MAAM,cAAyC,CAAC;CAChD,MAAM,gBAA+B;EAAE,MAAM;EAAO,QAAQ;CAAM;CAClE,IAAI,kBAAkB;CACtB,IAAI,qBAAqB;CACzB,IAAI,iBAAiB;CACrB,IAAI;CACJ,IAAI;CAEJ,MAAM,8BAAoC;EACxC,iBAAiB;EACjB,iBAAiB,KAAA;CACnB;CAEA,MAAM,+BAAqC;EACzC,iBAAiB;EACjB,iBAAiB,KAAA;CACnB;CAEA,MAAM,uBAAgC,kBAAkB,YAAY;CAEpE,MAAM,2BACJ,YAAY,SAAS;CAEvB,MAAM,0BACJ,mBAAmB,KAAA,OACnB,sBAAA;CAKF,MAAM,kBAAkB,YAA2B;EACjD,IACE,eAAe,KACf,cAAc,QACd,cAAc,UACd,gBAAgB,MAAM,GAEtB;EAEF,MAAM,IAAI,SAAe,YAAY;GACnC,MAAM,gBAAsB;IAC1B,QAAQ,oBAAoB,SAAS,OAAO;IAC5C,QAAQ;GACV;GACA,uBAA6B;IAC3B,QAAQ,oBAAoB,SAAS,OAAO;IAC5C,QAAQ;GACV;GACA,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;GACzD,IAAI,gBAAgB,MAAM,GACxB,QAAQ;EAEZ,CAAC;CACH;CAEA,MAAM,oBAAoB,YAA2B;EACnD,OACE,kBAAkB,KAClB,CAAC,kBACD,CAAC,gBAAgB,MAAM,GAEvB,MAAM,IAAI,SAAe,YAAY;GACnC,MAAM,gBAAsB;IAC1B,QAAQ,oBAAoB,SAAS,OAAO;IAC5C,QAAQ;GACV;GACA,MAAM,gBAAsB;IAC1B,QAAQ,oBAAoB,SAAS,OAAO;IAC5C,QAAQ;GACV;GACA,iBAAiB;GACjB,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;GACzD,IAAI,gBAAgB,MAAM,GACxB,QAAQ;EAEZ,CAAC;CAEL;CAEA,MAAM,gBAAqD;EACzD,IAAI,CAAC,eAAe,GAClB;EAEF,MAAM,aAAa,YAAY;EAC/B;EACA,IAAI,kBAAkB,OAAO,kBAAkB,KAAK,YAAY,QAAQ;GACtE,YAAY,OAAO,GAAG,eAAe;GACrC,kBAAkB;EACpB;EACA,OAAO;CACT;CAEA,MAAM,qBAA4B;EAChC,gBAAgB;EAChB,MAAM,IAAI,MAAM,oBAAoB;CACtC;CAEA,MAAM,UAAU,OAAO,SAA2C;EAChE,MAAM,kBAAkB;EACxB,IAAI,kBAAkB,gBAAgB,MAAM,GAC1C,aAAa;EAEf,MAAM,aAAa,KAAK,SAAS,KAAK,KAAK,SAAS;EACpD,YAAY,KAAK;GAAE;GAAM;EAAW,CAAC;EACrC,sBAAsB;EACtB,sBAAsB;CACxB;CAUA,MAAM,mBAAmB,OAAO,SAA2C;EACzE,IACE,CAAC,KAAK,UACN,KAAK,WAAW,QAChB,KAAK,KAAK,UAAA,MACV;GACA,MAAM,QAAQ,IAAI;GAClB;EACF;EAEA,MAAM,WAA6C,CAAC;EACpD,IAAI,SAAS;EACb,OAAO,SAAS,KAAK,KAAK,QAAQ;GAChC,MAAM,MACJ,SACA,wBACE,KAAK,KAAK,MAAM,MAAM,GACtB,6BACF;GACF,SAAS,KAAK;IAAE,OAAO;IAAQ;GAAI,CAAC;GACpC,SAAS;EACX;EAEA,KAAK,IAAI,IAAI,GAAG,IAAI,SAAS,QAAQ,KAAK;GACxC,MAAM,iBAAiB,MAAM;GAC7B,MAAM,gBAAgB,MAAM,SAAS,SAAS;GAC9C,MAAM,QAAQ;IACZ,MAAM,KAAK,KAAK,MAAM,SAAS,EAAE,CAAC,OAAO,SAAS,EAAE,CAAC,GAAG;IACxD,QAAQ;IACR,OAAO,UACL,KAAK,KAAK;KACR,MAAM,MAAM;KACZ,SAAS,kBAAkB,MAAM;KACjC,QAAQ,iBAAiB,MAAM;IACjC,CAAC;GACL,CAAC;EACH;CACF;CAEA,MAAM,YAAY,YAA2B;EAC3C,IAAI;GACF,WAAW,MAAM,QAAQ,QAAQ;IAC/B,IAAI,gBAAgB,MAAM,GACxB,aAAa;IAEf,MAAM,iBAAiB,IAAI;GAC7B;EACF,SAAS,OAAO;GACd,cAAc,SAAS;GACvB,cAAc,QAAQ;EACxB,UAAU;GACR,cAAc,OAAO;GACrB,sBAAsB;EACxB;CACF,EAAA,CAAG;CAEH,IAAI,iBAAiB;CACrB,IAAI;CACJ,IAAI;CACJ,IAAI;CACJ,IAAI,aAAa;CACjB,IAAI,gBAAgB;CACpB,IAAI;EACF,OAAO,eAAe;GACpB,IAAI,gBAAgB,MAAM,GACxB,aAAa;GAGf,IAAI,WAAW,MAAM;IACnB,MAAM,gBAAgB;IACtB,UAAU,QAAQ;IAClB,aAAa;GACf;GAEA,IAAI,WAAW,MAAM;IACnB,IAAI,cAAc,QAChB,MAAM,cAAc,yBAAS,IAAI,MAAA,yBAA6B;IAEhE,IAAI,cAAc,MAChB,gBAAgB;IAElB;GACF;GAEA,MAAM,EAAE,SAAS;GAEjB,IAAI,CAAC,KAAK,QAAQ;IAChB,uBAAuB;IACvB,UAAU,KAAA;IACV,MAAM,KAAK,KAAK;KAAE,MAAM,KAAK;KAAM,SAAS;KAAM,QAAQ;IAAK,CAAC;IAChE;GACF;GAEA,IAAI,KAAK,SAAS,IAAI;IACpB,qBAAqB,KAAK,IACxB,GACA,qBAAqB,QAAQ,UAC/B;IACA,uBAAuB;IACvB,UAAU,KAAA;IACV;GACF;GAKA,IAAI,cAAc,QAAQ,uBAAuB,MAC/C,sBAAsB,KAAK,IACzB,GACA,KAAK,MAAA,MAAiC,OAAO,CAC/C;GAEF,MAAM,aACJ,uBAAuB,OACnB,KAAK,IAAA,GAEL,KAAK,KAAK,qBAAqB,mBAAmB,CACpD,IACE,yBAAyB,oBAAoB,OAAO;GAC1D,IAAI,uBAAuB,QAAQ,sBAAsB,GACvD,uBAAuB;GAGzB,MAAM,mBACJ,uBAAuB;IACrB,aAAa,iBAAiB,UAAU;IACxC;IACA,KAAK,KAAK,IAAI;GAChB,CAAC,GACD,MACF;GACA,IAAI,gBAAgB,MAAM,GACxB,aAAa;GAEf,iBAAiB;GACjB,oBAAoB,KAAK,IAAI;GAE7B,IAAI,KAAK,WAAW,MAAM;IACxB,qBAAqB,KAAK,IACxB,GACA,qBAAqB,QAAQ,UAC/B;IACA,uBAAuB;IACvB,UAAU,KAAA;IACV,MAAM,KAAK,KAAK;KAAE,MAAM,KAAK;KAAM,SAAS;KAAM,QAAQ;IAAK,CAAC;IAChE;GACF;GAMA,IAAI,WAAW;GACf,OAAO,WAAW,YAAY;IAC5B,IAAI,gBAAgB,MAAM,GACxB,aAAa;IAEf,IAAI,WAAW,MAAM;KACnB,IAAI,CAAC,eAAe,GAClB;KAEF,UAAU,QAAQ;KAClB,aAAa;KACb,IAAI,WAAW,MACb;IAEJ;IAEA,MAAM,YAAY,QAAQ;IAC1B,IAAI,CAAC,UAAU,QAAQ;KACrB,uBAAuB;KACvB,UAAU,KAAA;KACV,MAAM,UAAU,KAAK;MACnB,MAAM,UAAU;MAChB,SAAS;MACT,QAAQ;KACV,CAAC;KACD;IACF;IACA,IAAI,UAAU,SAAS,IAAI;KACzB,qBAAqB,KAAK,IACxB,GACA,qBAAqB,QAAQ,UAC/B;KACA,uBAAuB;KACvB,UAAU,KAAA;KACV;IACF;IACA,IAAI,UAAU,WAAW,MACvB;IAIF,MAAM,cAAc,wBADE,UAAU,KAAK,MAAM,UAE7B,GACZ,aAAa,QACf;IACA,MAAM,WAAW,aAAa;IAC9B,MAAM,QAAQ,UAAU,KAAK,MAAM,YAAY,QAAQ;IACvD,MAAM,UAAU,eAAe;IAC/B,MAAM,SAAS,aAAa,UAAU,KAAK;IAE3C,qBAAqB,KAAK,IAAI,GAAG,qBAAqB,MAAM,MAAM;IAClE,uBAAuB;IACvB,IAAI,QACF,UAAU,KAAA;SAEV,aAAa;IAEf,YAAY,MAAM;IAClB,MAAM,UAAU,KAAK;KAAE,MAAM;KAAO;KAAS;IAAO,CAAC;GACvD;EACF;CACF,UAAU;EACR,iBAAiB;EACjB,IAAI,cAAc,MAChB,MAAM;OACD;GACL,gBAAgB;GAChB,uBAAuB;GAEvB,MAAM,UAAUA,OAAS,QAAQ,KAAK,QAAQ,KAAA,CAAkB;GAChE,IAAI,WAAW,MACb,QAAa,WACL,KAAA,SACA,KAAA,CACR;GAEF,MAAM,QAAQ,KAAK,CACjB,UACA,IAAI,SAAe,YAAY;IAE7B,WAD2B,SAAS,uBAC9B,CAAC,CAAC,MAAM;GAChB,CAAC,CACH,CAAC;EACH;CACF;AACF"}