@dreb/ai 2.61.0 → 2.61.2

package/dist/providers/openai-responses.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"openai-responses.d.ts","sourceRoot":"","sources":["../../src/providers/openai-responses.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,yCAAyC,CAAC;AAG7F,OAAO,KAAK,EAMX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAEb,MAAM,aAAa,CAAC;AAqCrB,MAAM,WAAW,sBAAuB,SAAQ,aAAa;IAC5D,eAAe,CAAC,EAAE,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;IAClE,gBAAgB,CAAC,EAAE,MAAM,GAAG,UAAU,GAAG,SAAS,GAAG,IAAI,CAAC;IAC1D,WAAW,CAAC,EAAE,6BAA6B,CAAC,cAAc,CAAC,CAAC;CAC5D;AAED;;GAEG;AACH,eAAO,MAAM,qBAAqB,EAAE,cAAc,CAAC,kBAAkB,EAAE,sBAAsB,CAoE5F,CAAC;AAEF,eAAO,MAAM,2BAA2B,EAAE,cAAc,CAAC,kBAAkB,EAAE,mBAAmB,CAiB/F,CAAC","sourcesContent":["import OpenAI from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { supportsXhigh } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses DREB_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.DREB_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\n/**\n * Get prompt cache retention based on cacheRetention and base URL.\n * Only applies to direct OpenAI API calls (api.openai.com).\n */\nfunction getPromptCacheRetention(baseUrl: string, cacheRetention: CacheRetention): \"24h\" | undefined {\n\tif (cacheRetention !== \"long\") {\n\t\treturn undefined;\n\t}\n\tif (baseUrl.includes(\"api.openai.com\")) {\n\t\treturn \"24h\";\n\t}\n\treturn undefined;\n}\n\n// OpenAI Responses-specific options\nexport interface OpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\n\treasoningSummary?: \"auto\" | \"detailed\" | \"concise\" | null;\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const streamOpenAIResponses: StreamFunction<\"openai-responses\", OpenAIResponsesOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: OpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t// Start async processing\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create OpenAI client\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider) || \"\";\n\t\t\tconst client = createClient(model, context, apiKey, options?.headers);\n\t\t\tlet params = buildParams(model, context, options);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst openaiStream = await client.responses.create(\n\t\t\t\tparams,\n\t\t\t\toptions?.signal ? { signal: options.signal } : undefined,\n\t\t\t);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tapplyServiceTierPricing,\n\t\t\t\tonWarning: options?.onWarning,\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) delete (block as { index?: number }).index;\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleOpenAIResponses: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\n\treturn streamOpenAIResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\tapiKey?: string,\n\toptionsHeaders?: Record<string, string>,\n) {\n\tif (!apiKey) {\n\t\tif (!process.env.OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"OpenAI API key is required. Set OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = buildCopilotDynamicHeaders({\n\t\t\tmessages: context.messages,\n\t\t\thasImages,\n\t\t});\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\t// Merge options headers last so they can override defaults\n\tif (optionsHeaders) {\n\t\tObject.assign(headers, optionsHeaders);\n\t}\n\n\treturn new OpenAI({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(model: Model<\"openai-responses\">, context: Context, options?: OpenAIResponsesOptions) {\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS);\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention);\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : options?.sessionId,\n\t\tprompt_cache_retention: getPromptCacheRetention(model.baseUrl, cacheRetention),\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = options?.maxTokens;\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertResponsesTools(context.tools);\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort || options?.reasoningSummary) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: options?.reasoningEffort || \"medium\",\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\") {\n\t\t\tparams.reasoning = { effort: \"none\" };\n\t\t}\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(serviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\t\treturn 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(usage: Usage, serviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined) {\n\tconst multiplier = getServiceTierCostMultiplier(serviceTier);\n\tif (multiplier === 1) return;\n\n\tusage.cost.input *= multiplier;\n\tusage.cost.output *= multiplier;\n\tusage.cost.cacheRead *= multiplier;\n\tusage.cost.cacheWrite *= multiplier;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n}\n"]}
+{"version":3,"file":"openai-responses.d.ts","sourceRoot":"","sources":["../../src/providers/openai-responses.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,yCAAyC,CAAC;AAE7F,OAAO,KAAK,EAMX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAEb,MAAM,aAAa,CAAC;AAqCrB,MAAM,WAAW,sBAAuB,SAAQ,aAAa;IAC5D,eAAe,CAAC,EAAE,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;IAC1E,gBAAgB,CAAC,EAAE,MAAM,GAAG,UAAU,GAAG,SAAS,GAAG,IAAI,CAAC;IAC1D,WAAW,CAAC,EAAE,6BAA6B,CAAC,cAAc,CAAC,CAAC;CAC5D;AAED;;GAEG;AACH,eAAO,MAAM,qBAAqB,EAAE,cAAc,CAAC,kBAAkB,EAAE,sBAAsB,CAoE5F,CAAC;AAEF,eAAO,MAAM,2BAA2B,EAAE,cAAc,CAAC,kBAAkB,EAAE,mBAAmB,CAiB/F,CAAC","sourcesContent":["import OpenAI from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, resolveReasoningEffort } from \"./simple-options.js\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses DREB_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.DREB_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\n/**\n * Get prompt cache retention based on cacheRetention and base URL.\n * Only applies to direct OpenAI API calls (api.openai.com).\n */\nfunction getPromptCacheRetention(baseUrl: string, cacheRetention: CacheRetention): \"24h\" | undefined {\n\tif (cacheRetention !== \"long\") {\n\t\treturn undefined;\n\t}\n\tif (baseUrl.includes(\"api.openai.com\")) {\n\t\treturn \"24h\";\n\t}\n\treturn undefined;\n}\n\n// OpenAI Responses-specific options\nexport interface OpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\treasoningSummary?: \"auto\" | \"detailed\" | \"concise\" | null;\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const streamOpenAIResponses: StreamFunction<\"openai-responses\", OpenAIResponsesOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: OpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t// Start async processing\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create OpenAI client\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider) || \"\";\n\t\t\tconst client = createClient(model, context, apiKey, options?.headers);\n\t\t\tlet params = buildParams(model, context, options);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst openaiStream = await client.responses.create(\n\t\t\t\tparams,\n\t\t\t\toptions?.signal ? { signal: options.signal } : undefined,\n\t\t\t);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tapplyServiceTierPricing,\n\t\t\t\tonWarning: options?.onWarning,\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) delete (block as { index?: number }).index;\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleOpenAIResponses: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = resolveReasoningEffort(model, options?.reasoning);\n\n\treturn streamOpenAIResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\tapiKey?: string,\n\toptionsHeaders?: Record<string, string>,\n) {\n\tif (!apiKey) {\n\t\tif (!process.env.OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"OpenAI API key is required. Set OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = buildCopilotDynamicHeaders({\n\t\t\tmessages: context.messages,\n\t\t\thasImages,\n\t\t});\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\t// Merge options headers last so they can override defaults\n\tif (optionsHeaders) {\n\t\tObject.assign(headers, optionsHeaders);\n\t}\n\n\treturn new OpenAI({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(model: Model<\"openai-responses\">, context: Context, options?: OpenAIResponsesOptions) {\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS);\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention);\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : options?.sessionId,\n\t\tprompt_cache_retention: getPromptCacheRetention(model.baseUrl, cacheRetention),\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = options?.maxTokens;\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertResponsesTools(context.tools);\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort || options?.reasoningSummary) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: options?.reasoningEffort || \"medium\",\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\") {\n\t\t\tparams.reasoning = { effort: \"none\" };\n\t\t}\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(serviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\t\treturn 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(usage: Usage, serviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined) {\n\tconst multiplier = getServiceTierCostMultiplier(serviceTier);\n\tif (multiplier === 1) return;\n\n\tusage.cost.input *= multiplier;\n\tusage.cost.output *= multiplier;\n\tusage.cost.cacheRead *= multiplier;\n\tusage.cost.cacheWrite *= multiplier;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n}\n"]}

package/dist/providers/openai-responses.js

@@ -1,10 +1,9 @@
 import OpenAI from "openai";
 import { getEnvApiKey } from "../env-api-keys.js";
-import { supportsXhigh } from "../models.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
 import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
-import { buildBaseOptions, clampReasoning } from "./simple-options.js";
+import { buildBaseOptions, resolveReasoningEffort } from "./simple-options.js";
 const OPENAI_TOOL_CALL_PROVIDERS = new Set(["openai", "openai-codex", "opencode"]);
 /**
  * Resolve cache retention preference.
@@ -98,7 +97,7 @@ export const streamSimpleOpenAIResponses = (model, context, options) => {
         throw new Error(`No API key for provider: ${model.provider}`);
     }
     const base = buildBaseOptions(model, options, apiKey);
-    const reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);
+    const reasoningEffort = resolveReasoningEffort(model, options?.reasoning);
     return streamOpenAIResponses(model, context, {
         ...base,
         reasoningEffort,

package/dist/providers/openai-responses.js.map

@@ -1 +1 @@
-{"version":3,"file":"openai-responses.js","sourceRoot":"","sources":["../../src/providers/openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAE5B,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAY7C,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,0BAA0B,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAChG,OAAO,EAAE,wBAAwB,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,8BAA8B,CAAC;AACvH,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAEvE,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC;AAEnF;;;GAGG;AACH,SAAS,qBAAqB,CAAC,cAA+B,EAAkB;IAC/E,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,OAAO,CAAC,GAAG,CAAC,oBAAoB,KAAK,MAAM,EAAE,CAAC;QACnF,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAAA,CACf;AAED;;;GAGG;AACH,SAAS,uBAAuB,CAAC,OAAe,EAAE,cAA8B,EAAqB;IACpG,IAAI,cAAc,KAAK,MAAM,EAAE,CAAC;QAC/B,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,IAAI,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;QACxC,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AASD;;GAEG;AACH,MAAM,CAAC,MAAM,qBAAqB,GAA+D,CAChG,KAAgC,EAChC,OAAgB,EAChB,OAAgC,EACF,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE,CAAC;QACZ,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,KAAK,CAAC,GAAU;YACrB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,MAAM;YAClB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,uBAAuB;YACvB,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACrE,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YACtE,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAClD,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAA2C,CAAC;YACtD,CAAC;YACD,MAAM,YAAY,GAAG,MAAM,MAAM,CAAC,SAAS,CAAC,MAAM,CACjD,MAAM,EACN,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,SAAS,CACxD,CAAC;YACF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAEhD,MAAM,sBAAsB,CAAC,YAAY,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE;gBACjE,WAAW,EAAE,OAAO,EAAE,WAAW;gBACjC,uBAAuB;gBACvB,SAAS,EAAE,OAAO,EAAE,SAAS;aAC7B,CAAC,CAAC;YAEH,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;YAC9C,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO;gBAAE,OAAQ,KAA4B,CAAC,KAAK,CAAC;YAC/E,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YACrF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IAAA,CACD,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AAAA,CACd,CAAC;AAEF,MAAM,CAAC,MAAM,2BAA2B,GAA4D,CACnG,KAAgC,EAChC,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,IAAI,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACtD,MAAM,eAAe,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;IAEvG,OAAO,qBAAqB,CAAC,KAAK,EAAE,OAAO,EAAE;QAC5C,GAAG,IAAI;QACP,eAAe;KACkB,CAAC,CAAC;AAAA,CACpC,CAAC;AAEF,SAAS,YAAY,CACpB,KAAgC,EAChC,OAAgB,EAChB,MAAe,EACf,cAAuC,EACtC;IACD,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,EAAE,CAAC;YACjC,MAAM,IAAI,KAAK,CACd,gGAAgG,CAChG,CAAC;QACH,CAAC;QACD,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC;IACrC,CAAC;IAED,MAAM,OAAO,GAAG,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IACrC,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,EAAE,CAAC;QACzC,MAAM,SAAS,GAAG,qBAAqB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QAC1D,MAAM,cAAc,GAAG,0BAA0B,CAAC;YACjD,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS;SACT,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,2DAA2D;IAC3D,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,OAAO,IAAI,MAAM,CAAC;QACjB,MAAM;QACN,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,uBAAuB,EAAE,IAAI;QAC7B,cAAc,EAAE,OAAO;KACvB,CAAC,CAAC;AAAA,CACH;AAED,SAAS,WAAW,CAAC,KAAgC,EAAE,OAAgB,EAAE,OAAgC,EAAE;IAC1G,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,0BAA0B,CAAC,CAAC;IAEtF,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACtE,MAAM,MAAM,GAAkC;QAC7C,KAAK,EAAE,KAAK,CAAC,EAAE;QACf,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,gBAAgB,EAAE,cAAc,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS;QAC5E,sBAAsB,EAAE,uBAAuB,CAAC,KAAK,CAAC,OAAO,EAAE,cAAc,CAAC;QAC9E,KAAK,EAAE,KAAK;KACZ,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,OAAO,EAAE,SAAS,CAAC;IAC/C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IACrD,CAAC;IAED,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,OAAO,EAAE,eAAe,IAAI,OAAO,EAAE,gBAAgB,EAAE,CAAC;YAC3D,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,OAAO,EAAE,eAAe,IAAI,QAAQ;gBAC5C,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,EAAE,CAAC;YAChD,MAAM,CAAC,SAAS,GAAG,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;QACvC,CAAC;IACF,CAAC;IAED,OAAO,MAAM,CAAC;AAAA,CACd;AAED,SAAS,4BAA4B,CAAC,WAAsE,EAAU;IACrH,QAAQ,WAAW,EAAE,CAAC;QACrB,KAAK,MAAM;YACV,OAAO,GAAG,CAAC;QACZ,KAAK,UAAU;YACd,OAAO,CAAC,CAAC;QACV;YACC,OAAO,CAAC,CAAC;IACX,CAAC;AAAA,CACD;AAED,SAAS,uBAAuB,CAAC,KAAY,EAAE,WAAsE,EAAE;IACtH,MAAM,UAAU,GAAG,4BAA4B,CAAC,WAAW,CAAC,CAAC;IAC7D,IAAI,UAAU,KAAK,CAAC;QAAE,OAAO;IAE7B,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,UAAU,CAAC;IAC/B,KAAK,CAAC,IAAI,CAAC,MAAM,IAAI,UAAU,CAAC;IAChC,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,UAAU,CAAC;IACnC,KAAK,CAAC,IAAI,CAAC,UAAU,IAAI,UAAU,CAAC;IACpC,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC;AAAA,CACvG","sourcesContent":["import OpenAI from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { supportsXhigh } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses DREB_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.DREB_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\n/**\n * Get prompt cache retention based on cacheRetention and base URL.\n * Only applies to direct OpenAI API calls (api.openai.com).\n */\nfunction getPromptCacheRetention(baseUrl: string, cacheRetention: CacheRetention): \"24h\" | undefined {\n\tif (cacheRetention !== \"long\") {\n\t\treturn undefined;\n\t}\n\tif (baseUrl.includes(\"api.openai.com\")) {\n\t\treturn \"24h\";\n\t}\n\treturn undefined;\n}\n\n// OpenAI Responses-specific options\nexport interface OpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\n\treasoningSummary?: \"auto\" | \"detailed\" | \"concise\" | null;\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const streamOpenAIResponses: StreamFunction<\"openai-responses\", OpenAIResponsesOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: OpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t// Start async processing\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create OpenAI client\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider) || \"\";\n\t\t\tconst client = createClient(model, context, apiKey, options?.headers);\n\t\t\tlet params = buildParams(model, context, options);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst openaiStream = await client.responses.create(\n\t\t\t\tparams,\n\t\t\t\toptions?.signal ? { signal: options.signal } : undefined,\n\t\t\t);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tapplyServiceTierPricing,\n\t\t\t\tonWarning: options?.onWarning,\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) delete (block as { index?: number }).index;\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleOpenAIResponses: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\n\treturn streamOpenAIResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\tapiKey?: string,\n\toptionsHeaders?: Record<string, string>,\n) {\n\tif (!apiKey) {\n\t\tif (!process.env.OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"OpenAI API key is required. Set OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = buildCopilotDynamicHeaders({\n\t\t\tmessages: context.messages,\n\t\t\thasImages,\n\t\t});\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\t// Merge options headers last so they can override defaults\n\tif (optionsHeaders) {\n\t\tObject.assign(headers, optionsHeaders);\n\t}\n\n\treturn new OpenAI({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(model: Model<\"openai-responses\">, context: Context, options?: OpenAIResponsesOptions) {\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS);\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention);\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : options?.sessionId,\n\t\tprompt_cache_retention: getPromptCacheRetention(model.baseUrl, cacheRetention),\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = options?.maxTokens;\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertResponsesTools(context.tools);\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort || options?.reasoningSummary) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: options?.reasoningEffort || \"medium\",\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\") {\n\t\t\tparams.reasoning = { effort: \"none\" };\n\t\t}\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(serviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\t\treturn 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(usage: Usage, serviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined) {\n\tconst multiplier = getServiceTierCostMultiplier(serviceTier);\n\tif (multiplier === 1) return;\n\n\tusage.cost.input *= multiplier;\n\tusage.cost.output *= multiplier;\n\tusage.cost.cacheRead *= multiplier;\n\tusage.cost.cacheWrite *= multiplier;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n}\n"]}
+{"version":3,"file":"openai-responses.js","sourceRoot":"","sources":["../../src/providers/openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAE5B,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAYlD,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,0BAA0B,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAChG,OAAO,EAAE,wBAAwB,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,8BAA8B,CAAC;AACvH,OAAO,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAE/E,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC;AAEnF;;;GAGG;AACH,SAAS,qBAAqB,CAAC,cAA+B,EAAkB;IAC/E,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,OAAO,CAAC,GAAG,CAAC,oBAAoB,KAAK,MAAM,EAAE,CAAC;QACnF,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAAA,CACf;AAED;;;GAGG;AACH,SAAS,uBAAuB,CAAC,OAAe,EAAE,cAA8B,EAAqB;IACpG,IAAI,cAAc,KAAK,MAAM,EAAE,CAAC;QAC/B,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,IAAI,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;QACxC,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AASD;;GAEG;AACH,MAAM,CAAC,MAAM,qBAAqB,GAA+D,CAChG,KAAgC,EAChC,OAAgB,EAChB,OAAgC,EACF,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE,CAAC;QACZ,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,KAAK,CAAC,GAAU;YACrB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,MAAM;YAClB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,uBAAuB;YACvB,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACrE,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YACtE,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAClD,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAA2C,CAAC;YACtD,CAAC;YACD,MAAM,YAAY,GAAG,MAAM,MAAM,CAAC,SAAS,CAAC,MAAM,CACjD,MAAM,EACN,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,SAAS,CACxD,CAAC;YACF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAEhD,MAAM,sBAAsB,CAAC,YAAY,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE;gBACjE,WAAW,EAAE,OAAO,EAAE,WAAW;gBACjC,uBAAuB;gBACvB,SAAS,EAAE,OAAO,EAAE,SAAS;aAC7B,CAAC,CAAC;YAEH,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;YAC9C,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO;gBAAE,OAAQ,KAA4B,CAAC,KAAK,CAAC;YAC/E,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YACrF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IAAA,CACD,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AAAA,CACd,CAAC;AAEF,MAAM,CAAC,MAAM,2BAA2B,GAA4D,CACnG,KAAgC,EAChC,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,IAAI,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACtD,MAAM,eAAe,GAAG,sBAAsB,CAAC,KAAK,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IAE1E,OAAO,qBAAqB,CAAC,KAAK,EAAE,OAAO,EAAE;QAC5C,GAAG,IAAI;QACP,eAAe;KACkB,CAAC,CAAC;AAAA,CACpC,CAAC;AAEF,SAAS,YAAY,CACpB,KAAgC,EAChC,OAAgB,EAChB,MAAe,EACf,cAAuC,EACtC;IACD,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,EAAE,CAAC;YACjC,MAAM,IAAI,KAAK,CACd,gGAAgG,CAChG,CAAC;QACH,CAAC;QACD,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC;IACrC,CAAC;IAED,MAAM,OAAO,GAAG,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IACrC,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,EAAE,CAAC;QACzC,MAAM,SAAS,GAAG,qBAAqB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QAC1D,MAAM,cAAc,GAAG,0BAA0B,CAAC;YACjD,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS;SACT,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,2DAA2D;IAC3D,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,OAAO,IAAI,MAAM,CAAC;QACjB,MAAM;QACN,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,uBAAuB,EAAE,IAAI;QAC7B,cAAc,EAAE,OAAO;KACvB,CAAC,CAAC;AAAA,CACH;AAED,SAAS,WAAW,CAAC,KAAgC,EAAE,OAAgB,EAAE,OAAgC,EAAE;IAC1G,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,0BAA0B,CAAC,CAAC;IAEtF,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACtE,MAAM,MAAM,GAAkC;QAC7C,KAAK,EAAE,KAAK,CAAC,EAAE;QACf,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,gBAAgB,EAAE,cAAc,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS;QAC5E,sBAAsB,EAAE,uBAAuB,CAAC,KAAK,CAAC,OAAO,EAAE,cAAc,CAAC;QAC9E,KAAK,EAAE,KAAK;KACZ,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,OAAO,EAAE,SAAS,CAAC;IAC/C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IACrD,CAAC;IAED,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,OAAO,EAAE,eAAe,IAAI,OAAO,EAAE,gBAAgB,EAAE,CAAC;YAC3D,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,OAAO,EAAE,eAAe,IAAI,QAAQ;gBAC5C,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,EAAE,CAAC;YAChD,MAAM,CAAC,SAAS,GAAG,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;QACvC,CAAC;IACF,CAAC;IAED,OAAO,MAAM,CAAC;AAAA,CACd;AAED,SAAS,4BAA4B,CAAC,WAAsE,EAAU;IACrH,QAAQ,WAAW,EAAE,CAAC;QACrB,KAAK,MAAM;YACV,OAAO,GAAG,CAAC;QACZ,KAAK,UAAU;YACd,OAAO,CAAC,CAAC;QACV;YACC,OAAO,CAAC,CAAC;IACX,CAAC;AAAA,CACD;AAED,SAAS,uBAAuB,CAAC,KAAY,EAAE,WAAsE,EAAE;IACtH,MAAM,UAAU,GAAG,4BAA4B,CAAC,WAAW,CAAC,CAAC;IAC7D,IAAI,UAAU,KAAK,CAAC;QAAE,OAAO;IAE7B,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,UAAU,CAAC;IAC/B,KAAK,CAAC,IAAI,CAAC,MAAM,IAAI,UAAU,CAAC;IAChC,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,UAAU,CAAC;IACnC,KAAK,CAAC,IAAI,CAAC,UAAU,IAAI,UAAU,CAAC;IACpC,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC;AAAA,CACvG","sourcesContent":["import OpenAI from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, resolveReasoningEffort } from \"./simple-options.js\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses DREB_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.DREB_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\n/**\n * Get prompt cache retention based on cacheRetention and base URL.\n * Only applies to direct OpenAI API calls (api.openai.com).\n */\nfunction getPromptCacheRetention(baseUrl: string, cacheRetention: CacheRetention): \"24h\" | undefined {\n\tif (cacheRetention !== \"long\") {\n\t\treturn undefined;\n\t}\n\tif (baseUrl.includes(\"api.openai.com\")) {\n\t\treturn \"24h\";\n\t}\n\treturn undefined;\n}\n\n// OpenAI Responses-specific options\nexport interface OpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\treasoningSummary?: \"auto\" | \"detailed\" | \"concise\" | null;\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const streamOpenAIResponses: StreamFunction<\"openai-responses\", OpenAIResponsesOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: OpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t// Start async processing\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create OpenAI client\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider) || \"\";\n\t\t\tconst client = createClient(model, context, apiKey, options?.headers);\n\t\t\tlet params = buildParams(model, context, options);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst openaiStream = await client.responses.create(\n\t\t\t\tparams,\n\t\t\t\toptions?.signal ? { signal: options.signal } : undefined,\n\t\t\t);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tapplyServiceTierPricing,\n\t\t\t\tonWarning: options?.onWarning,\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) delete (block as { index?: number }).index;\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleOpenAIResponses: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = resolveReasoningEffort(model, options?.reasoning);\n\n\treturn streamOpenAIResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\tapiKey?: string,\n\toptionsHeaders?: Record<string, string>,\n) {\n\tif (!apiKey) {\n\t\tif (!process.env.OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"OpenAI API key is required. Set OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = buildCopilotDynamicHeaders({\n\t\t\tmessages: context.messages,\n\t\t\thasImages,\n\t\t});\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\t// Merge options headers last so they can override defaults\n\tif (optionsHeaders) {\n\t\tObject.assign(headers, optionsHeaders);\n\t}\n\n\treturn new OpenAI({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(model: Model<\"openai-responses\">, context: Context, options?: OpenAIResponsesOptions) {\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS);\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention);\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : options?.sessionId,\n\t\tprompt_cache_retention: getPromptCacheRetention(model.baseUrl, cacheRetention),\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = options?.maxTokens;\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertResponsesTools(context.tools);\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort || options?.reasoningSummary) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: options?.reasoningEffort || \"medium\",\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\") {\n\t\t\tparams.reasoning = { effort: \"none\" };\n\t\t}\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(serviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\t\treturn 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(usage: Usage, serviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined) {\n\tconst multiplier = getServiceTierCostMultiplier(serviceTier);\n\tif (multiplier === 1) return;\n\n\tusage.cost.input *= multiplier;\n\tusage.cost.output *= multiplier;\n\tusage.cost.cacheRead *= multiplier;\n\tusage.cost.cacheWrite *= multiplier;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n}\n"]}

package/dist/providers/simple-options.d.ts

@@ -8,7 +8,8 @@ import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, T
  */
 export declare const DEFAULT_MAX_OUTPUT_TOKENS = 32000;
 export declare function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions;
-export declare function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, "xhigh"> | undefined;
+export declare function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, "xhigh" | "max"> | undefined;
+export declare function resolveReasoningEffort(model: Model<Api>, effort: ThinkingLevel | undefined): ThinkingLevel | undefined;
 export declare function adjustMaxTokensForThinking(baseMaxTokens: number, modelMaxTokens: number, reasoningLevel: ThinkingLevel, customBudgets?: ThinkingBudgets): {
     maxTokens: number;
     thinkingBudget: number;

package/dist/providers/simple-options.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"simple-options.d.ts","sourceRoot":"","sources":["../../src/providers/simple-options.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,mBAAmB,EAAE,aAAa,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAElH;;;;;;GAMG;AACH,eAAO,MAAM,yBAAyB,QAAQ,CAAC;AAE/C,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,CAAC,EAAE,mBAAmB,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,aAAa,CAajH;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,aAAa,GAAG,SAAS,GAAG,OAAO,CAAC,aAAa,EAAE,OAAO,CAAC,GAAG,SAAS,CAE7G;AAED,wBAAgB,0BAA0B,CACzC,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,EACtB,cAAc,EAAE,aAAa,EAC7B,aAAa,CAAC,EAAE,eAAe,GAC7B;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,MAAM,CAAA;CAAE,CAmB/C","sourcesContent":["import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from \"../types.js\";\n\n/**\n * Default cap on output tokens when no explicit maxTokens is requested. The\n * provider sends `Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS)` rather\n * than the full model ceiling. Consumers (e.g. the agent loop's length-retry\n * guard) must reference this constant to reason correctly about the real\n * budget a default request uses.\n */\nexport const DEFAULT_MAX_OUTPUT_TOKENS = 32000;\n\nexport function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tcacheRetention: options?.cacheRetention,\n\t\tsessionId: options?.sessionId,\n\t\theaders: options?.headers,\n\t\tonPayload: options?.onPayload,\n\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\tmetadata: options?.metadata,\n\t};\n}\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\"> | undefined {\n\treturn effort === \"xhigh\" ? \"high\" : effort;\n}\n\nexport function adjustMaxTokensForThinking(\n\tbaseMaxTokens: number,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tconst defaultBudgets: ThinkingBudgets = {\n\t\tminimal: 1024,\n\t\tlow: 2048,\n\t\tmedium: 8192,\n\t\thigh: 16384,\n\t};\n\tconst budgets = { ...defaultBudgets, ...customBudgets };\n\n\tconst minOutputTokens = 1024;\n\tconst level = clampReasoning(reasoningLevel)!;\n\tlet thinkingBudget = budgets[level]!;\n\tconst maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}
+{"version":3,"file":"simple-options.d.ts","sourceRoot":"","sources":["../../src/providers/simple-options.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,mBAAmB,EAAE,aAAa,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAElH;;;;;;GAMG;AACH,eAAO,MAAM,yBAAyB,QAAQ,CAAC;AAE/C,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,CAAC,EAAE,mBAAmB,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,aAAa,CAajH;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,aAAa,GAAG,SAAS,GAAG,OAAO,CAAC,aAAa,EAAE,OAAO,GAAG,KAAK,CAAC,GAAG,SAAS,CAErH;AAED,wBAAgB,sBAAsB,CACrC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,MAAM,EAAE,aAAa,GAAG,SAAS,GAC/B,aAAa,GAAG,SAAS,CAI3B;AAED,wBAAgB,0BAA0B,CACzC,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,EACtB,cAAc,EAAE,aAAa,EAC7B,aAAa,CAAC,EAAE,eAAe,GAC7B;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,MAAM,CAAA;CAAE,CAmB/C","sourcesContent":["import { supportsMax, supportsXhigh } from \"../models.js\";\nimport type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from \"../types.js\";\n\n/**\n * Default cap on output tokens when no explicit maxTokens is requested. The\n * provider sends `Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS)` rather\n * than the full model ceiling. Consumers (e.g. the agent loop's length-retry\n * guard) must reference this constant to reason correctly about the real\n * budget a default request uses.\n */\nexport const DEFAULT_MAX_OUTPUT_TOKENS = 32000;\n\nexport function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tcacheRetention: options?.cacheRetention,\n\t\tsessionId: options?.sessionId,\n\t\theaders: options?.headers,\n\t\tonPayload: options?.onPayload,\n\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\tmetadata: options?.metadata,\n\t};\n}\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\" | \"max\"> | undefined {\n\treturn effort === \"xhigh\" || effort === \"max\" ? \"high\" : effort;\n}\n\nexport function resolveReasoningEffort(\n\tmodel: Model<Api>,\n\teffort: ThinkingLevel | undefined,\n): ThinkingLevel | undefined {\n\tif (effort === \"max\" && !supportsMax(model)) return supportsXhigh(model) ? \"xhigh\" : \"high\";\n\tif (effort === \"xhigh\" && !supportsXhigh(model)) return \"high\";\n\treturn effort;\n}\n\nexport function adjustMaxTokensForThinking(\n\tbaseMaxTokens: number,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tconst defaultBudgets: ThinkingBudgets = {\n\t\tminimal: 1024,\n\t\tlow: 2048,\n\t\tmedium: 8192,\n\t\thigh: 16384,\n\t};\n\tconst budgets = { ...defaultBudgets, ...customBudgets };\n\n\tconst minOutputTokens = 1024;\n\tconst level = clampReasoning(reasoningLevel)!;\n\tlet thinkingBudget = budgets[level]!;\n\tconst maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}

package/dist/providers/simple-options.js

@@ -1,3 +1,4 @@
+import { supportsMax, supportsXhigh } from "../models.js";
 /**
  * Default cap on output tokens when no explicit maxTokens is requested. The
  * provider sends `Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS)` rather
@@ -21,7 +22,14 @@ export function buildBaseOptions(model, options, apiKey) {
     };
 }
 export function clampReasoning(effort) {
-    return effort === "xhigh" ? "high" : effort;
+    return effort === "xhigh" || effort === "max" ? "high" : effort;
+}
+export function resolveReasoningEffort(model, effort) {
+    if (effort === "max" && !supportsMax(model))
+        return supportsXhigh(model) ? "xhigh" : "high";
+    if (effort === "xhigh" && !supportsXhigh(model))
+        return "high";
+    return effort;
 }
 export function adjustMaxTokensForThinking(baseMaxTokens, modelMaxTokens, reasoningLevel, customBudgets) {
     const defaultBudgets = {

package/dist/providers/simple-options.js.map

@@ -1 +1 @@
-{"version":3,"file":"simple-options.js","sourceRoot":"","sources":["../../src/providers/simple-options.ts"],"names":[],"mappings":"AAEA;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,yBAAyB,GAAG,KAAK,CAAC;AAE/C,MAAM,UAAU,gBAAgB,CAAC,KAAiB,EAAE,OAA6B,EAAE,MAAe,EAAiB;IAClH,OAAO;QACN,WAAW,EAAE,OAAO,EAAE,WAAW;QACjC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,yBAAyB,CAAC;QACrF,MAAM,EAAE,OAAO,EAAE,MAAM;QACvB,MAAM,EAAE,MAAM,IAAI,OAAO,EAAE,MAAM;QACjC,cAAc,EAAE,OAAO,EAAE,cAAc;QACvC,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,OAAO,EAAE,OAAO,EAAE,OAAO;QACzB,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,eAAe,EAAE,OAAO,EAAE,eAAe;QACzC,QAAQ,EAAE,OAAO,EAAE,QAAQ;KAC3B,CAAC;AAAA,CACF;AAED,MAAM,UAAU,cAAc,CAAC,MAAiC,EAA+C;IAC9G,OAAO,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;AAAA,CAC5C;AAED,MAAM,UAAU,0BAA0B,CACzC,aAAqB,EACrB,cAAsB,EACtB,cAA6B,EAC7B,aAA+B,EACiB;IAChD,MAAM,cAAc,GAAoB;QACvC,OAAO,EAAE,IAAI;QACb,GAAG,EAAE,IAAI;QACT,MAAM,EAAE,IAAI;QACZ,IAAI,EAAE,KAAK;KACX,CAAC;IACF,MAAM,OAAO,GAAG,EAAE,GAAG,cAAc,EAAE,GAAG,aAAa,EAAE,CAAC;IAExD,MAAM,eAAe,GAAG,IAAI,CAAC;IAC7B,MAAM,KAAK,GAAG,cAAc,CAAC,cAAc,CAAE,CAAC;IAC9C,IAAI,cAAc,GAAG,OAAO,CAAC,KAAK,CAAE,CAAC;IACrC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,aAAa,GAAG,cAAc,EAAE,cAAc,CAAC,CAAC;IAE3E,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;QACjC,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,GAAG,eAAe,CAAC,CAAC;IAC3D,CAAC;IAED,OAAO,EAAE,SAAS,EAAE,cAAc,EAAE,CAAC;AAAA,CACrC","sourcesContent":["import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from \"../types.js\";\n\n/**\n * Default cap on output tokens when no explicit maxTokens is requested. The\n * provider sends `Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS)` rather\n * than the full model ceiling. Consumers (e.g. the agent loop's length-retry\n * guard) must reference this constant to reason correctly about the real\n * budget a default request uses.\n */\nexport const DEFAULT_MAX_OUTPUT_TOKENS = 32000;\n\nexport function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tcacheRetention: options?.cacheRetention,\n\t\tsessionId: options?.sessionId,\n\t\theaders: options?.headers,\n\t\tonPayload: options?.onPayload,\n\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\tmetadata: options?.metadata,\n\t};\n}\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\"> | undefined {\n\treturn effort === \"xhigh\" ? \"high\" : effort;\n}\n\nexport function adjustMaxTokensForThinking(\n\tbaseMaxTokens: number,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tconst defaultBudgets: ThinkingBudgets = {\n\t\tminimal: 1024,\n\t\tlow: 2048,\n\t\tmedium: 8192,\n\t\thigh: 16384,\n\t};\n\tconst budgets = { ...defaultBudgets, ...customBudgets };\n\n\tconst minOutputTokens = 1024;\n\tconst level = clampReasoning(reasoningLevel)!;\n\tlet thinkingBudget = budgets[level]!;\n\tconst maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}
+{"version":3,"file":"simple-options.js","sourceRoot":"","sources":["../../src/providers/simple-options.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAG1D;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,yBAAyB,GAAG,KAAK,CAAC;AAE/C,MAAM,UAAU,gBAAgB,CAAC,KAAiB,EAAE,OAA6B,EAAE,MAAe,EAAiB;IAClH,OAAO;QACN,WAAW,EAAE,OAAO,EAAE,WAAW;QACjC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,yBAAyB,CAAC;QACrF,MAAM,EAAE,OAAO,EAAE,MAAM;QACvB,MAAM,EAAE,MAAM,IAAI,OAAO,EAAE,MAAM;QACjC,cAAc,EAAE,OAAO,EAAE,cAAc;QACvC,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,OAAO,EAAE,OAAO,EAAE,OAAO;QACzB,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,eAAe,EAAE,OAAO,EAAE,eAAe;QACzC,QAAQ,EAAE,OAAO,EAAE,QAAQ;KAC3B,CAAC;AAAA,CACF;AAED,MAAM,UAAU,cAAc,CAAC,MAAiC,EAAuD;IACtH,OAAO,MAAM,KAAK,OAAO,IAAI,MAAM,KAAK,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;AAAA,CAChE;AAED,MAAM,UAAU,sBAAsB,CACrC,KAAiB,EACjB,MAAiC,EACL;IAC5B,IAAI,MAAM,KAAK,KAAK,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC;QAAE,OAAO,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC;IAC5F,IAAI,MAAM,KAAK,OAAO,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC;QAAE,OAAO,MAAM,CAAC;IAC/D,OAAO,MAAM,CAAC;AAAA,CACd;AAED,MAAM,UAAU,0BAA0B,CACzC,aAAqB,EACrB,cAAsB,EACtB,cAA6B,EAC7B,aAA+B,EACiB;IAChD,MAAM,cAAc,GAAoB;QACvC,OAAO,EAAE,IAAI;QACb,GAAG,EAAE,IAAI;QACT,MAAM,EAAE,IAAI;QACZ,IAAI,EAAE,KAAK;KACX,CAAC;IACF,MAAM,OAAO,GAAG,EAAE,GAAG,cAAc,EAAE,GAAG,aAAa,EAAE,CAAC;IAExD,MAAM,eAAe,GAAG,IAAI,CAAC;IAC7B,MAAM,KAAK,GAAG,cAAc,CAAC,cAAc,CAAE,CAAC;IAC9C,IAAI,cAAc,GAAG,OAAO,CAAC,KAAK,CAAE,CAAC;IACrC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,aAAa,GAAG,cAAc,EAAE,cAAc,CAAC,CAAC;IAE3E,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;QACjC,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,GAAG,eAAe,CAAC,CAAC;IAC3D,CAAC;IAED,OAAO,EAAE,SAAS,EAAE,cAAc,EAAE,CAAC;AAAA,CACrC","sourcesContent":["import { supportsMax, supportsXhigh } from \"../models.js\";\nimport type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from \"../types.js\";\n\n/**\n * Default cap on output tokens when no explicit maxTokens is requested. The\n * provider sends `Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS)` rather\n * than the full model ceiling. Consumers (e.g. the agent loop's length-retry\n * guard) must reference this constant to reason correctly about the real\n * budget a default request uses.\n */\nexport const DEFAULT_MAX_OUTPUT_TOKENS = 32000;\n\nexport function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, DEFAULT_MAX_OUTPUT_TOKENS),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tcacheRetention: options?.cacheRetention,\n\t\tsessionId: options?.sessionId,\n\t\theaders: options?.headers,\n\t\tonPayload: options?.onPayload,\n\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\tmetadata: options?.metadata,\n\t};\n}\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\" | \"max\"> | undefined {\n\treturn effort === \"xhigh\" || effort === \"max\" ? \"high\" : effort;\n}\n\nexport function resolveReasoningEffort(\n\tmodel: Model<Api>,\n\teffort: ThinkingLevel | undefined,\n): ThinkingLevel | undefined {\n\tif (effort === \"max\" && !supportsMax(model)) return supportsXhigh(model) ? \"xhigh\" : \"high\";\n\tif (effort === \"xhigh\" && !supportsXhigh(model)) return \"high\";\n\treturn effort;\n}\n\nexport function adjustMaxTokensForThinking(\n\tbaseMaxTokens: number,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tconst defaultBudgets: ThinkingBudgets = {\n\t\tminimal: 1024,\n\t\tlow: 2048,\n\t\tmedium: 8192,\n\t\thigh: 16384,\n\t};\n\tconst budgets = { ...defaultBudgets, ...customBudgets };\n\n\tconst minOutputTokens = 1024;\n\tconst level = clampReasoning(reasoningLevel)!;\n\tlet thinkingBudget = budgets[level]!;\n\tconst maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}

package/dist/types.d.ts

@@ -4,7 +4,7 @@ export type KnownApi = "openai-completions" | "mistral-conversations" | "openai-
 export type Api = KnownApi | (string & {});
 export type KnownProvider = "amazon-bedrock" | "anthropic" | "google" | "google-gemini-cli" | "google-antigravity" | "google-vertex" | "openai" | "azure-openai-responses" | "openai-codex" | "github-copilot" | "xai" | "groq" | "cerebras" | "openrouter" | "vercel-ai-gateway" | "mistral" | "minimax" | "minimax-cn" | "huggingface" | "opencode" | "opencode-go" | "kimi-coding" | "kimi-coding-oauth";
 export type Provider = KnownProvider | string;
-export type ThinkingLevel = "minimal" | "low" | "medium" | "high" | "xhigh";
+export type ThinkingLevel = "minimal" | "low" | "medium" | "high" | "xhigh" | "max";
 /** Token budgets for each thinking level (token-based providers only) */
 export interface ThinkingBudgets {
     minimal?: number;

package/dist/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AAE3E,YAAY,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AAE3E,MAAM,MAAM,QAAQ,GACjB,oBAAoB,GACpB,uBAAuB,GACvB,kBAAkB,GAClB,wBAAwB,GACxB,wBAAwB,GACxB,oBAAoB,GACpB,yBAAyB,GACzB,sBAAsB,GACtB,mBAAmB,GACnB,eAAe,CAAC;AAEnB,MAAM,MAAM,GAAG,GAAG,QAAQ,GAAG,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC;AAE3C,MAAM,MAAM,aAAa,GACtB,gBAAgB,GAChB,WAAW,GACX,QAAQ,GACR,mBAAmB,GACnB,oBAAoB,GACpB,eAAe,GACf,QAAQ,GACR,wBAAwB,GACxB,cAAc,GACd,gBAAgB,GAChB,KAAK,GACL,MAAM,GACN,UAAU,GACV,YAAY,GACZ,mBAAmB,GACnB,SAAS,GACT,SAAS,GACT,YAAY,GACZ,aAAa,GACb,UAAU,GACV,aAAa,GACb,aAAa,GACb,mBAAmB,CAAC;AACvB,MAAM,MAAM,QAAQ,GAAG,aAAa,GAAG,MAAM,CAAC;AAE9C,MAAM,MAAM,aAAa,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAE5E,yEAAyE;AACzE,MAAM,WAAW,eAAe;IAC/B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAGD,MAAM,MAAM,cAAc,GAAG,MAAM,GAAG,OAAO,GAAG,MAAM,CAAC;AAEvD,MAAM,MAAM,SAAS,GAAG,KAAK,GAAG,WAAW,GAAG,MAAM,CAAC;AAErD,sFAAsF;AACtF,MAAM,MAAM,QAAQ,GAAG,SAAS,GAAG,QAAQ,CAAC;AAE5C,MAAM,WAAW,aAAa;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;;;OAGG;IACH,SAAS,CAAC,EAAE,SAAS,CAAC;IACtB;;;OAGG;IACH,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC;;;;OAIG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;OAGG;IACH,SAAS,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,KAAK,OAAO,GAAG,SAAS,GAAG,OAAO,CAAC,OAAO,GAAG,SAAS,CAAC,CAAC;IACxG;;;;OAIG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC;;;;OAIG;IACH,SAAS,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;CACpD;AAED,MAAM,MAAM,qBAAqB,GAAG,aAAa,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAG5E,MAAM,WAAW,mBAAoB,SAAQ,aAAa;IACzD,SAAS,CAAC,EAAE,aAAa,CAAC;IAC1B,4EAA4E;IAC5E,eAAe,CAAC,EAAE,eAAe,CAAC;IAClC;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,YAAY,GAAG,SAAS,CAAC;CAC3C;AAUD,MAAM,MAAM,cAAc,CAAC,IAAI,SAAS,GAAG,GAAG,GAAG,EAAE,QAAQ,SAAS,aAAa,GAAG,aAAa,IAAI,CACpG,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,QAAQ,KACd,2BAA2B,CAAC;AAEjC,MAAM,WAAW,eAAe;IAC/B,CAAC,EAAE,CAAC,CAAC;IACL,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,CAAC,EAAE,YAAY,GAAG,cAAc,CAAC;CACtC;AAED,MAAM,WAAW,WAAW;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,eAAe;IAC/B,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B;;gDAE4C;IAC5C,QAAQ,CAAC,EAAE,OAAO,CAAC;CACnB;AAED,MAAM,WAAW,YAAY;IAC5B,IAAI,EAAE,OAAO,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,QAAQ;IACxB,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC/B,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,KAAK;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE;QACL,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;QACnB,KAAK,EAAE,MAAM,CAAC;KACd,CAAC;CACF;AAED,MAAM,MAAM,UAAU,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,WAAW;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,GAAG,CAAC,WAAW,GAAG,YAAY,CAAC,EAAE,CAAC;IACjD,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,gBAAgB;IAChC,IAAI,EAAE,WAAW,CAAC;IAClB,OAAO,EAAE,CAAC,WAAW,GAAG,eAAe,GAAG,QAAQ,CAAC,EAAE,CAAC;IACtD,GAAG,EAAE,GAAG,CAAC;IACT,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,KAAK,CAAC;IACb,UAAU,EAAE,UAAU,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,iBAAiB,CAAC,QAAQ,GAAG,GAAG;IAChD,IAAI,EAAE,YAAY,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,CAAC,WAAW,GAAG,YAAY,CAAC,EAAE,CAAC;IACxC,OAAO,CAAC,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,OAAO,GAAG,WAAW,GAAG,gBAAgB,GAAG,iBAAiB,CAAC;AAEzE,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAEjD,MAAM,WAAW,IAAI,CAAC,WAAW,SAAS,OAAO,GAAG,OAAO;IAC1D,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,WAAW,CAAC;CACxB;AAED,MAAM,WAAW,OAAO;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,OAAO,EAAE,CAAC;IACpB,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;CACf;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,qBAAqB,GAC9B;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC5C;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACvE;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC3E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC3E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,QAAQ,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC7F;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS,CAAC,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACvG;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,SAAS,GAAG,OAAO,CAAC,CAAC;IAAC,KAAK,EAAE,gBAAgB,CAAA;CAAE,CAAC;AAEhG;;;GAGG;AACH,MAAM,WAAW,uBAAuB;IACvC,wFAAwF;IACxF,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,yGAAyG;IACzG,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,yFAAyF;IACzF,uBAAuB,CAAC,EAAE,OAAO,CAAC;IAClC,wGAAwG;IACxG,kBAAkB,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC;IAC5D,qIAAqI;IACrI,wBAAwB,CAAC,EAAE,OAAO,CAAC;IACnC,0EAA0E;IAC1E,cAAc,CAAC,EAAE,uBAAuB,GAAG,YAAY,CAAC;IACxD,sFAAsF;IACtF,sBAAsB,CAAC,EAAE,OAAO,CAAC;IACjC,2HAA2H;IAC3H,gCAAgC,CAAC,EAAE,OAAO,CAAC;IAC3C,4HAA4H;IAC5H,sBAAsB,CAAC,EAAE,OAAO,CAAC;IACjC,8eAA8e;IAC9e,cAAc,CAAC,EAAE,QAAQ,GAAG,YAAY,GAAG,KAAK,GAAG,MAAM,GAAG,oBAAoB,GAAG,MAAM,CAAC;IAC1F,4FAA4F;IAC5F,iBAAiB,CAAC,EAAE,iBAAiB,CAAC;IACtC,iGAAiG;IACjG,oBAAoB,CAAC,EAAE,oBAAoB,CAAC;IAC5C,2FAA2F;IAC3F,kBAAkB,CAAC,EAAE,OAAO,CAAC;CAC7B;AAED,wDAAwD;AACxD,MAAM,WAAW,qBAAqB;CAErC;AAED;;;;GAIG;AACH,MAAM,WAAW,iBAAiB;IACjC,0GAA0G;IAC1G,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,8EAA8E;IAC9E,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED;;;;GAIG;AACH,MAAM,WAAW,oBAAoB;IACpC,mGAAmG;IACnG,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,8EAA8E;IAC9E,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAGD,MAAM,WAAW,KAAK,CAAC,IAAI,SAAS,GAAG;IACtC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,IAAI,CAAC;IACV,QAAQ,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,OAAO,CAAC;IACnB,KAAK,EAAE,CAAC,MAAM,GAAG,OAAO,CAAC,EAAE,CAAC;IAC5B,IAAI,EAAE;QACL,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;KACnB,CAAC;IACF,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC;;;;;OAKG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,8FAA8F;IAC9F,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,kGAAkG;IAClG,MAAM,CAAC,EAAE,IAAI,SAAS,oBAAoB,GACvC,uBAAuB,GACvB,IAAI,SAAS,kBAAkB,GAC9B,qBAAqB,GACrB,KAAK,CAAC;CACV","sourcesContent":["import type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type KnownApi =\n\t| \"openai-completions\"\n\t| \"mistral-conversations\"\n\t| \"openai-responses\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\n\nexport type Api = KnownApi | (string & {});\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"vercel-ai-gateway\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"minimax-cn\"\n\t| \"huggingface\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"kimi-coding-oauth\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\n\n/** Token budgets for each thinking level (token-based providers only) */\nexport interface ThinkingBudgets {\n\tminimal?: number;\n\tlow?: number;\n\tmedium?: number;\n\thigh?: number;\n}\n\n// Base options all providers share\nexport type CacheRetention = \"none\" | \"short\" | \"long\";\n\nexport type Transport = \"sse\" | \"websocket\" | \"auto\";\n\n/** Explicit credential transport for providers that support multiple auth schemes. */\nexport type AuthMode = \"api-key\" | \"bearer\";\n\nexport interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Preferred transport for providers that support multiple transports.\n\t * Providers that do not support this option ignore it.\n\t */\n\ttransport?: Transport;\n\t/**\n\t * Prompt cache retention preference. Providers map this to their supported values.\n\t * Default: \"short\".\n\t */\n\tcacheRetention?: CacheRetention;\n\t/**\n\t * Optional session identifier for providers that support session-based caching.\n\t * Providers can use this to enable prompt caching, request routing, or other\n\t * session-aware features. Ignored by providers that don't support it.\n\t */\n\tsessionId?: string;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: Model<Api>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; can override default headers.\n\t * Not supported by all providers (e.g., AWS Bedrock uses SDK auth).\n\t */\n\theaders?: Record<string, string>;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.\n\t */\n\tmetadata?: Record<string, unknown>;\n\t/**\n\t * Optional callback for non-fatal warnings during streaming (e.g. malformed JSON chunks).\n\t * Called by providers when errors are caught but don't terminate the stream.\n\t * The session layer wires this to surface warnings in the conversation.\n\t */\n\tonWarning?: (code: string, message: string) => void;\n}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n\t/**\n\t * Controls how thinking content is returned by providers that support it.\n\t * \"summarized\": return summarized thinking text. \"omitted\": return empty thinking\n\t * with only an encrypted signature (lower latency, nothing to display).\n\t * Only honored by adaptive-thinking Anthropic/Bedrock Claude models; ignored elsewhere.\n\t * When unset, the provider default applies (Opus 4.7+ default to \"omitted\").\n\t */\n\tthinkingDisplay?: \"summarized\" | \"omitted\";\n}\n\n// Generic StreamFunction with typed options.\n//\n// Contract:\n// - Must return an AssistantMessageEventStream.\n// - Once invoked, request/model/runtime failures should be encoded in the\n//   returned stream, not thrown.\n// - Error termination must produce an AssistantMessage with stopReason\n//   \"error\" or \"aborted\" and errorMessage, emitted via the stream protocol.\nexport type StreamFunction<TApi extends Api = Api, TOptions extends StreamOptions = StreamOptions> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: TOptions,\n) => AssistantMessageEventStream;\n\nexport interface TextSignatureV1 {\n\tv: 1;\n\tid: string;\n\tphase?: \"commentary\" | \"final_answer\";\n}\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n\t/** When true, the thinking content was redacted by safety filters. The opaque\n\t *  encrypted payload is stored in `thinkingSignature` so it can be passed back\n\t *  to the API for multi-turn continuity. */\n\tredacted?: boolean;\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tresponseId?: string; // Provider-specific response/message identifier when the upstream API exposes one\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n\tdurationMs?: number; // Time from stream start to completion in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"@sinclair/typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\n/**\n * Event protocol for AssistantMessageEventStream.\n *\n * Streams should emit `start` before partial updates, then terminate with either:\n * - `done` carrying the final successful AssistantMessage, or\n * - `error` carrying the final AssistantMessage with stopReason \"error\" or \"aborted\"\n *   and errorMessage.\n */\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for OpenAI-compatible completions APIs.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompletionsCompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Optional mapping from dreb reasoning levels to provider/model-specific `reasoning_effort` values. */\n\treasoningEffortMap?: Partial<Record<ThinkingLevel, string>>;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Format for reasoning/thinking parameter. \"openai\" uses reasoning_effort, \"openrouter\" uses reasoning: { effort }, \"zai\" uses top-level enable_thinking: boolean, \"qwen\" uses top-level enable_thinking: boolean, \"qwen-chat-template\" uses chat_template_kwargs.enable_thinking plus a top-level mapped reasoning_effort (Qwen3.8+ models default-map dreb levels onto the native low/medium/xhigh tiers), and \"kimi\" uses nested thinking: { type, effort? } plus prompt_cache_key. Default: \"openai\". */\n\tthinkingFormat?: \"openai\" | \"openrouter\" | \"zai\" | \"qwen\" | \"qwen-chat-template\" | \"kimi\";\n\t/** OpenRouter-specific routing preferences. Only used when baseUrl points to OpenRouter. */\n\topenRouterRouting?: OpenRouterRouting;\n\t/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */\n\tvercelGatewayRouting?: VercelGatewayRouting;\n\t/** Whether the provider supports the `strict` field in tool definitions. Default: true. */\n\tsupportsStrictMode?: boolean;\n}\n\n/** Compatibility settings for OpenAI Responses APIs. */\nexport interface OpenAIResponsesCompat {\n\t// Reserved for future use\n}\n\n/**\n * OpenRouter provider routing preferences.\n * Controls which upstream providers OpenRouter routes requests to.\n * @see https://openrouter.ai/docs/provider-routing\n */\nexport interface OpenRouterRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"amazon-bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n/**\n * Vercel AI Gateway routing preferences.\n * Controls which upstream providers the gateway routes requests to.\n * @see https://vercel.com/docs/ai-gateway/models-and-providers/provider-options\n */\nexport interface VercelGatewayRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n// Model interface for the unified model system\nexport interface Model<TApi extends Api> {\n\tid: string;\n\tname: string;\n\tapi: TApi;\n\tprovider: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/**\n\t * Model identifier sent to the API when it differs from the registry id.\n\t * Used by context-tiered models such as Kimi K3, where dreb exposes a\n\t * single user-facing model (`k3`) but sends the cheaper 256k variant\n\t * (`k3-256k`) on the wire until the session context outgrows it.\n\t */\n\twireModelId?: string;\n\t/** Explicit auth transport. When unset, the provider's default and token heuristics apply. */\n\tauthMode?: AuthMode;\n\t/** Compatibility overrides for OpenAI-compatible APIs. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\"\n\t\t? OpenAICompletionsCompat\n\t\t: TApi extends \"openai-responses\"\n\t\t\t? OpenAIResponsesCompat\n\t\t\t: never;\n}\n"]}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AAE3E,YAAY,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AAE3E,MAAM,MAAM,QAAQ,GACjB,oBAAoB,GACpB,uBAAuB,GACvB,kBAAkB,GAClB,wBAAwB,GACxB,wBAAwB,GACxB,oBAAoB,GACpB,yBAAyB,GACzB,sBAAsB,GACtB,mBAAmB,GACnB,eAAe,CAAC;AAEnB,MAAM,MAAM,GAAG,GAAG,QAAQ,GAAG,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC;AAE3C,MAAM,MAAM,aAAa,GACtB,gBAAgB,GAChB,WAAW,GACX,QAAQ,GACR,mBAAmB,GACnB,oBAAoB,GACpB,eAAe,GACf,QAAQ,GACR,wBAAwB,GACxB,cAAc,GACd,gBAAgB,GAChB,KAAK,GACL,MAAM,GACN,UAAU,GACV,YAAY,GACZ,mBAAmB,GACnB,SAAS,GACT,SAAS,GACT,YAAY,GACZ,aAAa,GACb,UAAU,GACV,aAAa,GACb,aAAa,GACb,mBAAmB,CAAC;AACvB,MAAM,MAAM,QAAQ,GAAG,aAAa,GAAG,MAAM,CAAC;AAE9C,MAAM,MAAM,aAAa,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;AAEpF,yEAAyE;AACzE,MAAM,WAAW,eAAe;IAC/B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAGD,MAAM,MAAM,cAAc,GAAG,MAAM,GAAG,OAAO,GAAG,MAAM,CAAC;AAEvD,MAAM,MAAM,SAAS,GAAG,KAAK,GAAG,WAAW,GAAG,MAAM,CAAC;AAErD,sFAAsF;AACtF,MAAM,MAAM,QAAQ,GAAG,SAAS,GAAG,QAAQ,CAAC;AAE5C,MAAM,WAAW,aAAa;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;;;OAGG;IACH,SAAS,CAAC,EAAE,SAAS,CAAC;IACtB;;;OAGG;IACH,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC;;;;OAIG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;OAGG;IACH,SAAS,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,KAAK,OAAO,GAAG,SAAS,GAAG,OAAO,CAAC,OAAO,GAAG,SAAS,CAAC,CAAC;IACxG;;;;OAIG;IACH,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACnC;;;;OAIG;IACH,SAAS,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;CACpD;AAED,MAAM,MAAM,qBAAqB,GAAG,aAAa,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAG5E,MAAM,WAAW,mBAAoB,SAAQ,aAAa;IACzD,SAAS,CAAC,EAAE,aAAa,CAAC;IAC1B,4EAA4E;IAC5E,eAAe,CAAC,EAAE,eAAe,CAAC;IAClC;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,YAAY,GAAG,SAAS,CAAC;CAC3C;AAUD,MAAM,MAAM,cAAc,CAAC,IAAI,SAAS,GAAG,GAAG,GAAG,EAAE,QAAQ,SAAS,aAAa,GAAG,aAAa,IAAI,CACpG,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,QAAQ,KACd,2BAA2B,CAAC;AAEjC,MAAM,WAAW,eAAe;IAC/B,CAAC,EAAE,CAAC,CAAC;IACL,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,CAAC,EAAE,YAAY,GAAG,cAAc,CAAC;CACtC;AAED,MAAM,WAAW,WAAW;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,eAAe;IAC/B,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B;;gDAE4C;IAC5C,QAAQ,CAAC,EAAE,OAAO,CAAC;CACnB;AAED,MAAM,WAAW,YAAY;IAC5B,IAAI,EAAE,OAAO,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,QAAQ;IACxB,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC/B,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,KAAK;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE;QACL,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;QACnB,KAAK,EAAE,MAAM,CAAC;KACd,CAAC;CACF;AAED,MAAM,MAAM,UAAU,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,WAAW;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,GAAG,CAAC,WAAW,GAAG,YAAY,CAAC,EAAE,CAAC;IACjD,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,gBAAgB;IAChC,IAAI,EAAE,WAAW,CAAC;IAClB,OAAO,EAAE,CAAC,WAAW,GAAG,eAAe,GAAG,QAAQ,CAAC,EAAE,CAAC;IACtD,GAAG,EAAE,GAAG,CAAC;IACT,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,KAAK,CAAC;IACb,UAAU,EAAE,UAAU,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,iBAAiB,CAAC,QAAQ,GAAG,GAAG;IAChD,IAAI,EAAE,YAAY,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,CAAC,WAAW,GAAG,YAAY,CAAC,EAAE,CAAC;IACxC,OAAO,CAAC,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,OAAO,GAAG,WAAW,GAAG,gBAAgB,GAAG,iBAAiB,CAAC;AAEzE,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAEjD,MAAM,WAAW,IAAI,CAAC,WAAW,SAAS,OAAO,GAAG,OAAO;IAC1D,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,WAAW,CAAC;CACxB;AAED,MAAM,WAAW,OAAO;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,OAAO,EAAE,CAAC;IACpB,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;CACf;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,qBAAqB,GAC9B;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC5C;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACvE;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC3E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC3E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,QAAQ,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC7F;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS,CAAC,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACvG;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,SAAS,GAAG,OAAO,CAAC,CAAC;IAAC,KAAK,EAAE,gBAAgB,CAAA;CAAE,CAAC;AAEhG;;;GAGG;AACH,MAAM,WAAW,uBAAuB;IACvC,wFAAwF;IACxF,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,yGAAyG;IACzG,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,yFAAyF;IACzF,uBAAuB,CAAC,EAAE,OAAO,CAAC;IAClC,wGAAwG;IACxG,kBAAkB,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC;IAC5D,qIAAqI;IACrI,wBAAwB,CAAC,EAAE,OAAO,CAAC;IACnC,0EAA0E;IAC1E,cAAc,CAAC,EAAE,uBAAuB,GAAG,YAAY,CAAC;IACxD,sFAAsF;IACtF,sBAAsB,CAAC,EAAE,OAAO,CAAC;IACjC,2HAA2H;IAC3H,gCAAgC,CAAC,EAAE,OAAO,CAAC;IAC3C,4HAA4H;IAC5H,sBAAsB,CAAC,EAAE,OAAO,CAAC;IACjC,8eAA8e;IAC9e,cAAc,CAAC,EAAE,QAAQ,GAAG,YAAY,GAAG,KAAK,GAAG,MAAM,GAAG,oBAAoB,GAAG,MAAM,CAAC;IAC1F,4FAA4F;IAC5F,iBAAiB,CAAC,EAAE,iBAAiB,CAAC;IACtC,iGAAiG;IACjG,oBAAoB,CAAC,EAAE,oBAAoB,CAAC;IAC5C,2FAA2F;IAC3F,kBAAkB,CAAC,EAAE,OAAO,CAAC;CAC7B;AAED,wDAAwD;AACxD,MAAM,WAAW,qBAAqB;CAErC;AAED;;;;GAIG;AACH,MAAM,WAAW,iBAAiB;IACjC,0GAA0G;IAC1G,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,8EAA8E;IAC9E,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED;;;;GAIG;AACH,MAAM,WAAW,oBAAoB;IACpC,mGAAmG;IACnG,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,8EAA8E;IAC9E,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAGD,MAAM,WAAW,KAAK,CAAC,IAAI,SAAS,GAAG;IACtC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,IAAI,CAAC;IACV,QAAQ,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,OAAO,CAAC;IACnB,KAAK,EAAE,CAAC,MAAM,GAAG,OAAO,CAAC,EAAE,CAAC;IAC5B,IAAI,EAAE;QACL,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;KACnB,CAAC;IACF,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC;;;;;OAKG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,8FAA8F;IAC9F,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,kGAAkG;IAClG,MAAM,CAAC,EAAE,IAAI,SAAS,oBAAoB,GACvC,uBAAuB,GACvB,IAAI,SAAS,kBAAkB,GAC9B,qBAAqB,GACrB,KAAK,CAAC;CACV","sourcesContent":["import type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type KnownApi =\n\t| \"openai-completions\"\n\t| \"mistral-conversations\"\n\t| \"openai-responses\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\n\nexport type Api = KnownApi | (string & {});\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"vercel-ai-gateway\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"minimax-cn\"\n\t| \"huggingface\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"kimi-coding-oauth\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\n/** Token budgets for each thinking level (token-based providers only) */\nexport interface ThinkingBudgets {\n\tminimal?: number;\n\tlow?: number;\n\tmedium?: number;\n\thigh?: number;\n}\n\n// Base options all providers share\nexport type CacheRetention = \"none\" | \"short\" | \"long\";\n\nexport type Transport = \"sse\" | \"websocket\" | \"auto\";\n\n/** Explicit credential transport for providers that support multiple auth schemes. */\nexport type AuthMode = \"api-key\" | \"bearer\";\n\nexport interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Preferred transport for providers that support multiple transports.\n\t * Providers that do not support this option ignore it.\n\t */\n\ttransport?: Transport;\n\t/**\n\t * Prompt cache retention preference. Providers map this to their supported values.\n\t * Default: \"short\".\n\t */\n\tcacheRetention?: CacheRetention;\n\t/**\n\t * Optional session identifier for providers that support session-based caching.\n\t * Providers can use this to enable prompt caching, request routing, or other\n\t * session-aware features. Ignored by providers that don't support it.\n\t */\n\tsessionId?: string;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: Model<Api>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; can override default headers.\n\t * Not supported by all providers (e.g., AWS Bedrock uses SDK auth).\n\t */\n\theaders?: Record<string, string>;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.\n\t */\n\tmetadata?: Record<string, unknown>;\n\t/**\n\t * Optional callback for non-fatal warnings during streaming (e.g. malformed JSON chunks).\n\t * Called by providers when errors are caught but don't terminate the stream.\n\t * The session layer wires this to surface warnings in the conversation.\n\t */\n\tonWarning?: (code: string, message: string) => void;\n}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n\t/**\n\t * Controls how thinking content is returned by providers that support it.\n\t * \"summarized\": return summarized thinking text. \"omitted\": return empty thinking\n\t * with only an encrypted signature (lower latency, nothing to display).\n\t * Only honored by adaptive-thinking Anthropic/Bedrock Claude models; ignored elsewhere.\n\t * When unset, the provider default applies (Opus 4.7+ default to \"omitted\").\n\t */\n\tthinkingDisplay?: \"summarized\" | \"omitted\";\n}\n\n// Generic StreamFunction with typed options.\n//\n// Contract:\n// - Must return an AssistantMessageEventStream.\n// - Once invoked, request/model/runtime failures should be encoded in the\n//   returned stream, not thrown.\n// - Error termination must produce an AssistantMessage with stopReason\n//   \"error\" or \"aborted\" and errorMessage, emitted via the stream protocol.\nexport type StreamFunction<TApi extends Api = Api, TOptions extends StreamOptions = StreamOptions> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: TOptions,\n) => AssistantMessageEventStream;\n\nexport interface TextSignatureV1 {\n\tv: 1;\n\tid: string;\n\tphase?: \"commentary\" | \"final_answer\";\n}\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n\t/** When true, the thinking content was redacted by safety filters. The opaque\n\t *  encrypted payload is stored in `thinkingSignature` so it can be passed back\n\t *  to the API for multi-turn continuity. */\n\tredacted?: boolean;\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tresponseId?: string; // Provider-specific response/message identifier when the upstream API exposes one\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n\tdurationMs?: number; // Time from stream start to completion in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"@sinclair/typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\n/**\n * Event protocol for AssistantMessageEventStream.\n *\n * Streams should emit `start` before partial updates, then terminate with either:\n * - `done` carrying the final successful AssistantMessage, or\n * - `error` carrying the final AssistantMessage with stopReason \"error\" or \"aborted\"\n *   and errorMessage.\n */\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for OpenAI-compatible completions APIs.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompletionsCompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Optional mapping from dreb reasoning levels to provider/model-specific `reasoning_effort` values. */\n\treasoningEffortMap?: Partial<Record<ThinkingLevel, string>>;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Format for reasoning/thinking parameter. \"openai\" uses reasoning_effort, \"openrouter\" uses reasoning: { effort }, \"zai\" uses top-level enable_thinking: boolean, \"qwen\" uses top-level enable_thinking: boolean, \"qwen-chat-template\" uses chat_template_kwargs.enable_thinking plus a top-level mapped reasoning_effort (Qwen3.8+ models default-map dreb levels onto the native low/medium/xhigh tiers), and \"kimi\" uses nested thinking: { type, effort? } plus prompt_cache_key. Default: \"openai\". */\n\tthinkingFormat?: \"openai\" | \"openrouter\" | \"zai\" | \"qwen\" | \"qwen-chat-template\" | \"kimi\";\n\t/** OpenRouter-specific routing preferences. Only used when baseUrl points to OpenRouter. */\n\topenRouterRouting?: OpenRouterRouting;\n\t/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */\n\tvercelGatewayRouting?: VercelGatewayRouting;\n\t/** Whether the provider supports the `strict` field in tool definitions. Default: true. */\n\tsupportsStrictMode?: boolean;\n}\n\n/** Compatibility settings for OpenAI Responses APIs. */\nexport interface OpenAIResponsesCompat {\n\t// Reserved for future use\n}\n\n/**\n * OpenRouter provider routing preferences.\n * Controls which upstream providers OpenRouter routes requests to.\n * @see https://openrouter.ai/docs/provider-routing\n */\nexport interface OpenRouterRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"amazon-bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n/**\n * Vercel AI Gateway routing preferences.\n * Controls which upstream providers the gateway routes requests to.\n * @see https://vercel.com/docs/ai-gateway/models-and-providers/provider-options\n */\nexport interface VercelGatewayRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n// Model interface for the unified model system\nexport interface Model<TApi extends Api> {\n\tid: string;\n\tname: string;\n\tapi: TApi;\n\tprovider: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/**\n\t * Model identifier sent to the API when it differs from the registry id.\n\t * Used by context-tiered models such as Kimi K3, where dreb exposes a\n\t * single user-facing model (`k3`) but sends the cheaper 256k variant\n\t * (`k3-256k`) on the wire until the session context outgrows it.\n\t */\n\twireModelId?: string;\n\t/** Explicit auth transport. When unset, the provider's default and token heuristics apply. */\n\tauthMode?: AuthMode;\n\t/** Compatibility overrides for OpenAI-compatible APIs. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\"\n\t\t? OpenAICompletionsCompat\n\t\t: TApi extends \"openai-responses\"\n\t\t\t? OpenAIResponsesCompat\n\t\t\t: never;\n}\n"]}

package/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"","sourcesContent":["import type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type KnownApi =\n\t| \"openai-completions\"\n\t| \"mistral-conversations\"\n\t| \"openai-responses\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\n\nexport type Api = KnownApi | (string & {});\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"vercel-ai-gateway\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"minimax-cn\"\n\t| \"huggingface\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"kimi-coding-oauth\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\n\n/** Token budgets for each thinking level (token-based providers only) */\nexport interface ThinkingBudgets {\n\tminimal?: number;\n\tlow?: number;\n\tmedium?: number;\n\thigh?: number;\n}\n\n// Base options all providers share\nexport type CacheRetention = \"none\" | \"short\" | \"long\";\n\nexport type Transport = \"sse\" | \"websocket\" | \"auto\";\n\n/** Explicit credential transport for providers that support multiple auth schemes. */\nexport type AuthMode = \"api-key\" | \"bearer\";\n\nexport interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Preferred transport for providers that support multiple transports.\n\t * Providers that do not support this option ignore it.\n\t */\n\ttransport?: Transport;\n\t/**\n\t * Prompt cache retention preference. Providers map this to their supported values.\n\t * Default: \"short\".\n\t */\n\tcacheRetention?: CacheRetention;\n\t/**\n\t * Optional session identifier for providers that support session-based caching.\n\t * Providers can use this to enable prompt caching, request routing, or other\n\t * session-aware features. Ignored by providers that don't support it.\n\t */\n\tsessionId?: string;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: Model<Api>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; can override default headers.\n\t * Not supported by all providers (e.g., AWS Bedrock uses SDK auth).\n\t */\n\theaders?: Record<string, string>;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.\n\t */\n\tmetadata?: Record<string, unknown>;\n\t/**\n\t * Optional callback for non-fatal warnings during streaming (e.g. malformed JSON chunks).\n\t * Called by providers when errors are caught but don't terminate the stream.\n\t * The session layer wires this to surface warnings in the conversation.\n\t */\n\tonWarning?: (code: string, message: string) => void;\n}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n\t/**\n\t * Controls how thinking content is returned by providers that support it.\n\t * \"summarized\": return summarized thinking text. \"omitted\": return empty thinking\n\t * with only an encrypted signature (lower latency, nothing to display).\n\t * Only honored by adaptive-thinking Anthropic/Bedrock Claude models; ignored elsewhere.\n\t * When unset, the provider default applies (Opus 4.7+ default to \"omitted\").\n\t */\n\tthinkingDisplay?: \"summarized\" | \"omitted\";\n}\n\n// Generic StreamFunction with typed options.\n//\n// Contract:\n// - Must return an AssistantMessageEventStream.\n// - Once invoked, request/model/runtime failures should be encoded in the\n//   returned stream, not thrown.\n// - Error termination must produce an AssistantMessage with stopReason\n//   \"error\" or \"aborted\" and errorMessage, emitted via the stream protocol.\nexport type StreamFunction<TApi extends Api = Api, TOptions extends StreamOptions = StreamOptions> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: TOptions,\n) => AssistantMessageEventStream;\n\nexport interface TextSignatureV1 {\n\tv: 1;\n\tid: string;\n\tphase?: \"commentary\" | \"final_answer\";\n}\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n\t/** When true, the thinking content was redacted by safety filters. The opaque\n\t *  encrypted payload is stored in `thinkingSignature` so it can be passed back\n\t *  to the API for multi-turn continuity. */\n\tredacted?: boolean;\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tresponseId?: string; // Provider-specific response/message identifier when the upstream API exposes one\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n\tdurationMs?: number; // Time from stream start to completion in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"@sinclair/typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\n/**\n * Event protocol for AssistantMessageEventStream.\n *\n * Streams should emit `start` before partial updates, then terminate with either:\n * - `done` carrying the final successful AssistantMessage, or\n * - `error` carrying the final AssistantMessage with stopReason \"error\" or \"aborted\"\n *   and errorMessage.\n */\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for OpenAI-compatible completions APIs.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompletionsCompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Optional mapping from dreb reasoning levels to provider/model-specific `reasoning_effort` values. */\n\treasoningEffortMap?: Partial<Record<ThinkingLevel, string>>;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Format for reasoning/thinking parameter. \"openai\" uses reasoning_effort, \"openrouter\" uses reasoning: { effort }, \"zai\" uses top-level enable_thinking: boolean, \"qwen\" uses top-level enable_thinking: boolean, \"qwen-chat-template\" uses chat_template_kwargs.enable_thinking plus a top-level mapped reasoning_effort (Qwen3.8+ models default-map dreb levels onto the native low/medium/xhigh tiers), and \"kimi\" uses nested thinking: { type, effort? } plus prompt_cache_key. Default: \"openai\". */\n\tthinkingFormat?: \"openai\" | \"openrouter\" | \"zai\" | \"qwen\" | \"qwen-chat-template\" | \"kimi\";\n\t/** OpenRouter-specific routing preferences. Only used when baseUrl points to OpenRouter. */\n\topenRouterRouting?: OpenRouterRouting;\n\t/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */\n\tvercelGatewayRouting?: VercelGatewayRouting;\n\t/** Whether the provider supports the `strict` field in tool definitions. Default: true. */\n\tsupportsStrictMode?: boolean;\n}\n\n/** Compatibility settings for OpenAI Responses APIs. */\nexport interface OpenAIResponsesCompat {\n\t// Reserved for future use\n}\n\n/**\n * OpenRouter provider routing preferences.\n * Controls which upstream providers OpenRouter routes requests to.\n * @see https://openrouter.ai/docs/provider-routing\n */\nexport interface OpenRouterRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"amazon-bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n/**\n * Vercel AI Gateway routing preferences.\n * Controls which upstream providers the gateway routes requests to.\n * @see https://vercel.com/docs/ai-gateway/models-and-providers/provider-options\n */\nexport interface VercelGatewayRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n// Model interface for the unified model system\nexport interface Model<TApi extends Api> {\n\tid: string;\n\tname: string;\n\tapi: TApi;\n\tprovider: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/**\n\t * Model identifier sent to the API when it differs from the registry id.\n\t * Used by context-tiered models such as Kimi K3, where dreb exposes a\n\t * single user-facing model (`k3`) but sends the cheaper 256k variant\n\t * (`k3-256k`) on the wire until the session context outgrows it.\n\t */\n\twireModelId?: string;\n\t/** Explicit auth transport. When unset, the provider's default and token heuristics apply. */\n\tauthMode?: AuthMode;\n\t/** Compatibility overrides for OpenAI-compatible APIs. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\"\n\t\t? OpenAICompletionsCompat\n\t\t: TApi extends \"openai-responses\"\n\t\t\t? OpenAIResponsesCompat\n\t\t\t: never;\n}\n"]}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"","sourcesContent":["import type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type KnownApi =\n\t| \"openai-completions\"\n\t| \"mistral-conversations\"\n\t| \"openai-responses\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\n\nexport type Api = KnownApi | (string & {});\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"vercel-ai-gateway\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"minimax-cn\"\n\t| \"huggingface\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"kimi-coding-oauth\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\n\n/** Token budgets for each thinking level (token-based providers only) */\nexport interface ThinkingBudgets {\n\tminimal?: number;\n\tlow?: number;\n\tmedium?: number;\n\thigh?: number;\n}\n\n// Base options all providers share\nexport type CacheRetention = \"none\" | \"short\" | \"long\";\n\nexport type Transport = \"sse\" | \"websocket\" | \"auto\";\n\n/** Explicit credential transport for providers that support multiple auth schemes. */\nexport type AuthMode = \"api-key\" | \"bearer\";\n\nexport interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Preferred transport for providers that support multiple transports.\n\t * Providers that do not support this option ignore it.\n\t */\n\ttransport?: Transport;\n\t/**\n\t * Prompt cache retention preference. Providers map this to their supported values.\n\t * Default: \"short\".\n\t */\n\tcacheRetention?: CacheRetention;\n\t/**\n\t * Optional session identifier for providers that support session-based caching.\n\t * Providers can use this to enable prompt caching, request routing, or other\n\t * session-aware features. Ignored by providers that don't support it.\n\t */\n\tsessionId?: string;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: Model<Api>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; can override default headers.\n\t * Not supported by all providers (e.g., AWS Bedrock uses SDK auth).\n\t */\n\theaders?: Record<string, string>;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.\n\t */\n\tmetadata?: Record<string, unknown>;\n\t/**\n\t * Optional callback for non-fatal warnings during streaming (e.g. malformed JSON chunks).\n\t * Called by providers when errors are caught but don't terminate the stream.\n\t * The session layer wires this to surface warnings in the conversation.\n\t */\n\tonWarning?: (code: string, message: string) => void;\n}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n\t/**\n\t * Controls how thinking content is returned by providers that support it.\n\t * \"summarized\": return summarized thinking text. \"omitted\": return empty thinking\n\t * with only an encrypted signature (lower latency, nothing to display).\n\t * Only honored by adaptive-thinking Anthropic/Bedrock Claude models; ignored elsewhere.\n\t * When unset, the provider default applies (Opus 4.7+ default to \"omitted\").\n\t */\n\tthinkingDisplay?: \"summarized\" | \"omitted\";\n}\n\n// Generic StreamFunction with typed options.\n//\n// Contract:\n// - Must return an AssistantMessageEventStream.\n// - Once invoked, request/model/runtime failures should be encoded in the\n//   returned stream, not thrown.\n// - Error termination must produce an AssistantMessage with stopReason\n//   \"error\" or \"aborted\" and errorMessage, emitted via the stream protocol.\nexport type StreamFunction<TApi extends Api = Api, TOptions extends StreamOptions = StreamOptions> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: TOptions,\n) => AssistantMessageEventStream;\n\nexport interface TextSignatureV1 {\n\tv: 1;\n\tid: string;\n\tphase?: \"commentary\" | \"final_answer\";\n}\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n\t/** When true, the thinking content was redacted by safety filters. The opaque\n\t *  encrypted payload is stored in `thinkingSignature` so it can be passed back\n\t *  to the API for multi-turn continuity. */\n\tredacted?: boolean;\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tresponseId?: string; // Provider-specific response/message identifier when the upstream API exposes one\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n\tdurationMs?: number; // Time from stream start to completion in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"@sinclair/typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\n/**\n * Event protocol for AssistantMessageEventStream.\n *\n * Streams should emit `start` before partial updates, then terminate with either:\n * - `done` carrying the final successful AssistantMessage, or\n * - `error` carrying the final AssistantMessage with stopReason \"error\" or \"aborted\"\n *   and errorMessage.\n */\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for OpenAI-compatible completions APIs.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompletionsCompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Optional mapping from dreb reasoning levels to provider/model-specific `reasoning_effort` values. */\n\treasoningEffortMap?: Partial<Record<ThinkingLevel, string>>;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Format for reasoning/thinking parameter. \"openai\" uses reasoning_effort, \"openrouter\" uses reasoning: { effort }, \"zai\" uses top-level enable_thinking: boolean, \"qwen\" uses top-level enable_thinking: boolean, \"qwen-chat-template\" uses chat_template_kwargs.enable_thinking plus a top-level mapped reasoning_effort (Qwen3.8+ models default-map dreb levels onto the native low/medium/xhigh tiers), and \"kimi\" uses nested thinking: { type, effort? } plus prompt_cache_key. Default: \"openai\". */\n\tthinkingFormat?: \"openai\" | \"openrouter\" | \"zai\" | \"qwen\" | \"qwen-chat-template\" | \"kimi\";\n\t/** OpenRouter-specific routing preferences. Only used when baseUrl points to OpenRouter. */\n\topenRouterRouting?: OpenRouterRouting;\n\t/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */\n\tvercelGatewayRouting?: VercelGatewayRouting;\n\t/** Whether the provider supports the `strict` field in tool definitions. Default: true. */\n\tsupportsStrictMode?: boolean;\n}\n\n/** Compatibility settings for OpenAI Responses APIs. */\nexport interface OpenAIResponsesCompat {\n\t// Reserved for future use\n}\n\n/**\n * OpenRouter provider routing preferences.\n * Controls which upstream providers OpenRouter routes requests to.\n * @see https://openrouter.ai/docs/provider-routing\n */\nexport interface OpenRouterRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"amazon-bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n/**\n * Vercel AI Gateway routing preferences.\n * Controls which upstream providers the gateway routes requests to.\n * @see https://vercel.com/docs/ai-gateway/models-and-providers/provider-options\n */\nexport interface VercelGatewayRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n// Model interface for the unified model system\nexport interface Model<TApi extends Api> {\n\tid: string;\n\tname: string;\n\tapi: TApi;\n\tprovider: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/**\n\t * Model identifier sent to the API when it differs from the registry id.\n\t * Used by context-tiered models such as Kimi K3, where dreb exposes a\n\t * single user-facing model (`k3`) but sends the cheaper 256k variant\n\t * (`k3-256k`) on the wire until the session context outgrows it.\n\t */\n\twireModelId?: string;\n\t/** Explicit auth transport. When unset, the provider's default and token heuristics apply. */\n\tauthMode?: AuthMode;\n\t/** Compatibility overrides for OpenAI-compatible APIs. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\"\n\t\t? OpenAICompletionsCompat\n\t\t: TApi extends \"openai-responses\"\n\t\t\t? OpenAIResponsesCompat\n\t\t\t: never;\n}\n"]}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@dreb/ai",
-	"version": "2.61.0",
+	"version": "2.61.2",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -83,7 +83,7 @@
 		"ajv": "^8.17.1",
 		"ajv-formats": "^3.0.1",
 		"chalk": "^5.6.2",
-		"openai": "6.26.0",
+		"openai": "6.49.0",
 		"partial-json": "^0.1.7",
 		"proxy-agent": "^6.5.0",
 		"undici": "^7.19.1",