gsd-pi 2.22.0 → 2.24.0

package/packages/pi-ai/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;;;GAGG;AACH,SAAS,sBAAsB,CAAC,SAAiB,EAAE,MAAc;IAChE,MAAM,IAAI,GAAG,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAG,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/E,IAAI,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,KAAK,CAAC;IACnE,OAAO,MAAM,CAAC;AACf,CAAC;AAED,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC;AAEnF;;;GAGG;AACH,SAAS,qBAAqB,CAAC,cAA+B;IAC7D,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,OAAO,CAAC,GAAG,CAAC,kBAAkB,KAAK,MAAM,EAAE,CAAC;QACjF,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED;;;GAGG;AACH,SAAS,uBAAuB,CAAC,OAAe,EAAE,cAA8B;IAC/E,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;AAClB,CAAC;AASD;;GAEG;AACH,MAAM,CAAC,MAAM,qBAAqB,GAA+D,CAChG,KAAgC,EAChC,OAAgB,EAChB,OAAgC,EACF,EAAE;IAChC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE;QACX,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;aACvB,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;IACF,CAAC,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AACf,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,2BAA2B,GAA4D,CACnG,KAAgC,EAChC,OAAgB,EAChB,OAA6B,EACC,EAAE;IAChC,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;AACrC,CAAC,CAAC;AAEF,SAAS,YAAY,CACpB,KAAgC,EAChC,OAAgB,EAChB,MAAe,EACf,cAAuC;IAEvC,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;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,KAAgC,EAAE,OAAgB,EAAE,OAAgC;IACxG,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,MAAM,GAAG,sBAAsB,CAAC,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,eAAe,IAAI,QAAQ,CAAmC,CAAC;YAC1H,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,MAAM,IAAI,QAAQ;gBAC1B,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,CAAC;YACP,IAAI,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;gBACpC,mGAAmG;gBACnG,QAAQ,CAAC,IAAI,CAAC;oBACb,IAAI,EAAE,WAAW;oBACjB,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,YAAY;4BAClB,IAAI,EAAE,uBAAuB;yBAC7B;qBACD;iBACD,CAAC,CAAC;YACJ,CAAC;QACF,CAAC;IACF,CAAC;IAED,OAAO,MAAM,CAAC;AACf,CAAC;AAED,SAAS,4BAA4B,CAAC,WAAsE;IAC3G,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;AACF,CAAC;AAED,SAAS,uBAAuB,CAAC,KAAY,EAAE,WAAsE;IACpH,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;AACxG,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\n/**\n * Clamp reasoning effort for models that don't support all levels.\n * gpt-5.x models don't support \"minimal\" — map to \"low\".\n */\nfunction clampReasoningForModel(modelName: string, effort: string): string {\n\tconst name = modelName.includes(\"/\") ? modelName.split(\"/\").pop()! : modelName;\n\tif (name.startsWith(\"gpt-5\") && effort === \"minimal\") return \"low\";\n\treturn effort;\n}\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 PI_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.PI_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});\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\tconst effort = clampReasoningForModel(model.name, options?.reasoningEffort || \"medium\") as typeof options.reasoningEffort;\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort || \"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 {\n\t\t\tif (model.name.startsWith(\"gpt-5\")) {\n\t\t\t\t// Jesus Christ, see https://community.openai.com/t/need-reasoning-false-option-for-gpt-5/1351588/7\n\t\t\t\tmessages.push({\n\t\t\t\t\trole: \"developer\",\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"input_text\",\n\t\t\t\t\t\t\ttext: \"# Juice: 0 !important\",\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t});\n\t\t\t}\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":"AAIA,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,IAAI,qBAAgD,CAAC;AACrD,KAAK,UAAU,uBAAuB;IACrC,IAAI,CAAC,qBAAqB,EAAE,CAAC;QAC5B,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,CAAC;QACnC,qBAAqB,GAAG,GAAG,CAAC,OAAO,CAAC;IACrC,CAAC;IACD,OAAO,qBAAqB,CAAC;AAC9B,CAAC;AAED;;;GAGG;AACH,SAAS,sBAAsB,CAAC,SAAiB,EAAE,MAAc;IAChE,MAAM,IAAI,GAAG,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAG,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/E,IAAI,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,KAAK,CAAC;IACnE,OAAO,MAAM,CAAC;AACf,CAAC;AAED,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC;AAEnF;;;GAGG;AACH,SAAS,qBAAqB,CAAC,cAA+B;IAC7D,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,OAAO,CAAC,GAAG,CAAC,kBAAkB,KAAK,MAAM,EAAE,CAAC;QACjF,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED;;;GAGG;AACH,SAAS,uBAAuB,CAAC,OAAe,EAAE,cAA8B;IAC/E,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;AAClB,CAAC;AASD;;GAEG;AACH,MAAM,CAAC,MAAM,qBAAqB,GAA+D,CAChG,KAAgC,EAChC,OAAgB,EAChB,OAAgC,EACF,EAAE;IAChC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE;QACX,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,MAAM,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAC5E,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;aACvB,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;IACF,CAAC,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AACf,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,2BAA2B,GAA4D,CACnG,KAAgC,EAChC,OAAgB,EAChB,OAA6B,EACC,EAAE;IAChC,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;AACrC,CAAC,CAAC;AAEF,KAAK,UAAU,YAAY,CAC1B,KAAgC,EAChC,OAAgB,EAChB,MAAe,EACf,cAAuC;IAEvC,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,MAAM,WAAW,GAAG,MAAM,uBAAuB,EAAE,CAAC;IACpD,OAAO,IAAI,WAAW,CAAC;QACtB,MAAM;QACN,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,uBAAuB,EAAE,IAAI;QAC7B,cAAc,EAAE,OAAO;KACvB,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,KAAgC,EAAE,OAAgB,EAAE,OAAgC;IACxG,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,MAAM,GAAG,sBAAsB,CAAC,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,eAAe,IAAI,QAAQ,CAAmC,CAAC;YAC1H,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,MAAM,IAAI,QAAQ;gBAC1B,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,CAAC;YACP,IAAI,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;gBACpC,mGAAmG;gBACnG,QAAQ,CAAC,IAAI,CAAC;oBACb,IAAI,EAAE,WAAW;oBACjB,OAAO,EAAE;wBACR;4BACC,IAAI,EAAE,YAAY;4BAClB,IAAI,EAAE,uBAAuB;yBAC7B;qBACD;iBACD,CAAC,CAAC;YACJ,CAAC;QACF,CAAC;IACF,CAAC;IAED,OAAO,MAAM,CAAC;AACf,CAAC;AAED,SAAS,4BAA4B,CAAC,WAAsE;IAC3G,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;AACF,CAAC;AAED,SAAS,uBAAuB,CAAC,KAAY,EAAE,WAAsE;IACpH,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;AACxG,CAAC","sourcesContent":["// Lazy-loaded: OpenAI SDK is imported on first use, not at startup.\n// This avoids penalizing users who don't use OpenAI models.\nimport type 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\nlet _OpenAIResponsesClass: typeof OpenAI | undefined;\nasync function getOpenAIResponsesClass(): Promise<typeof OpenAI> {\n\tif (!_OpenAIResponsesClass) {\n\t\tconst mod = await import(\"openai\");\n\t\t_OpenAIResponsesClass = mod.default;\n\t}\n\treturn _OpenAIResponsesClass;\n}\n\n/**\n * Clamp reasoning effort for models that don't support all levels.\n * gpt-5.x models don't support \"minimal\" — map to \"low\".\n */\nfunction clampReasoningForModel(modelName: string, effort: string): string {\n\tconst name = modelName.includes(\"/\") ? modelName.split(\"/\").pop()! : modelName;\n\tif (name.startsWith(\"gpt-5\") && effort === \"minimal\") return \"low\";\n\treturn effort;\n}\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 PI_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.PI_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 = await 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});\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\nasync function 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\tconst OpenAIClass = await getOpenAIResponsesClass();\n\treturn new OpenAIClass({\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\tconst effort = clampReasoningForModel(model.name, options?.reasoningEffort || \"medium\") as typeof options.reasoningEffort;\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort || \"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 {\n\t\t\tif (model.name.startsWith(\"gpt-5\")) {\n\t\t\t\t// Jesus Christ, see https://community.openai.com/t/need-reasoning-false-option-for-gpt-5/1351588/7\n\t\t\t\tmessages.push({\n\t\t\t\t\trole: \"developer\",\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"input_text\",\n\t\t\t\t\t\t\ttext: \"# Juice: 0 !important\",\n\t\t\t\t\t\t},\n\t\t\t\t\t],\n\t\t\t\t});\n\t\t\t}\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/packages/pi-ai/src/providers/anthropic.ts

@@ -1,4 +1,6 @@
-import Anthropic from "@anthropic-ai/sdk";
+// Lazy-loaded: Anthropic SDK (~500ms) is imported on first use, not at startup.
+// This avoids penalizing users who don't use Anthropic models.
+import type Anthropic from "@anthropic-ai/sdk";
 import type {
 	ContentBlockParam,
 	MessageCreateParamsStreaming,
@@ -32,6 +34,15 @@ import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copi
 import { adjustMaxTokensForThinking, buildBaseOptions } from "./simple-options.js";
 import { transformMessages } from "./transform-messages.js";
 
+let _AnthropicClass: typeof Anthropic | undefined;
+async function getAnthropicClass(): Promise<typeof Anthropic> {
+	if (!_AnthropicClass) {
+		const mod = await import("@anthropic-ai/sdk");
+		_AnthropicClass = mod.default;
+	}
+	return _AnthropicClass;
+}
+
 /**
  * Resolve cache retention preference.
  * Defaults to "short" and uses PI_CACHE_RETENTION for backward compatibility.
@@ -265,7 +276,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
 				});
 			}
 
-			const { client, isOAuthToken } = createClient(
+			const { client, isOAuthToken } = await createClient(
 				model,
 				apiKey,
 				options?.interleavedThinking ?? true,
@@ -455,7 +466,8 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
 			if (model.provider === "alibaba-coding-plan") {
 				output.errorMessage = `[alibaba-coding-plan] ${output.errorMessage}`;
 			}
-			if (error instanceof Anthropic.APIError && error.headers) {
+			const AnthropicSdk = _AnthropicClass;
+			if (AnthropicSdk && error instanceof AnthropicSdk.APIError && error.headers) {
 				const retryAfterMs = extractRetryAfterMs(error.headers, error.message);
 				if (retryAfterMs !== undefined) {
 					output.retryAfterMs = retryAfterMs;
@@ -548,13 +560,14 @@ function isOAuthToken(apiKey: string): boolean {
 	return apiKey.includes("sk-ant-oat");
 }
 
-function createClient(
+async function createClient(
 	model: Model<"anthropic-messages">,
 	apiKey: string,
 	interleavedThinking: boolean,
 	optionsHeaders?: Record<string, string>,
 	dynamicHeaders?: Record<string, string>,
-): { client: Anthropic; isOAuthToken: boolean } {
+): Promise<{ client: Anthropic; isOAuthToken: boolean }> {
+	const AnthropicClass = await getAnthropicClass();
 	// Adaptive thinking models (Opus 4.6, Sonnet 4.6) have interleaved thinking built-in.
 	// The beta header is deprecated on Opus 4.6 and redundant on Sonnet 4.6, so skip it.
 	const needsInterleavedBeta = interleavedThinking && !supportsAdaptiveThinking(model.id);
@@ -566,7 +579,7 @@ function createClient(
 			betaFeatures.push("interleaved-thinking-2025-05-14");
 		}
 
-		const client = new Anthropic({
+		const client = new AnthropicClass({
 			apiKey: null,
 			authToken: apiKey,
 			baseURL: model.baseUrl,
@@ -595,7 +608,7 @@ function createClient(
 
 	// OAuth: Bearer auth, Claude Code identity headers
 	if (isOAuthToken(apiKey)) {
-		const client = new Anthropic({
+		const client = new AnthropicClass({
 			apiKey: null,
 			authToken: apiKey,
 			baseURL: model.baseUrl,
@@ -619,7 +632,7 @@ function createClient(
 	// API key auth
 	// Alibaba Coding Plan uses Bearer token auth instead of x-api-key
 	const isAlibabaProvider = model.provider === "alibaba-coding-plan";
-	const client = new Anthropic({
+	const client = new AnthropicClass({
 		apiKey: isAlibabaProvider ? null : apiKey,
 		authToken: isAlibabaProvider ? apiKey : undefined,
 		baseURL: model.baseUrl,

package/packages/pi-ai/src/providers/azure-openai-responses.ts

@@ -1,4 +1,6 @@
-import { AzureOpenAI } from "openai";
+// Lazy-loaded: OpenAI SDK (AzureOpenAI) is imported on first use, not at startup.
+// This avoids penalizing users who don't use Azure OpenAI models.
+import type { AzureOpenAI } from "openai";
 import type { ResponseCreateParamsStreaming } from "openai/resources/responses/responses.js";
 import { getEnvApiKey } from "../env-api-keys.js";
 import { supportsXhigh } from "../models.js";
@@ -15,6 +17,15 @@ import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
 import { buildBaseOptions, clampReasoning } from "./simple-options.js";
 
+let _AzureOpenAIClass: typeof AzureOpenAI | undefined;
+async function getAzureOpenAIClass(): Promise<typeof AzureOpenAI> {
+	if (!_AzureOpenAIClass) {
+		const mod = await import("openai");
+		_AzureOpenAIClass = mod.AzureOpenAI;
+	}
+	return _AzureOpenAIClass;
+}
+
 /**
  * Clamp reasoning effort for models that don't support all levels.
  * gpt-5.x models don't support "minimal" — map to "low".
@@ -94,7 +105,7 @@ export const streamAzureOpenAIResponses: StreamFunction<"azure-openai-responses"
 		try {
 			// Create Azure OpenAI client
 			const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
-			const client = createClient(model, apiKey, options);
+			const client = await createClient(model, apiKey, options);
 			let params = buildParams(model, context, options, deploymentName);
 			const nextParams = await options?.onPayload?.(params, model);
 			if (nextParams !== undefined) {
@@ -188,7 +199,7 @@ function resolveAzureConfig(
 	};
 }
 
-function createClient(model: Model<"azure-openai-responses">, apiKey: string, options?: AzureOpenAIResponsesOptions) {
+async function createClient(model: Model<"azure-openai-responses">, apiKey: string, options?: AzureOpenAIResponsesOptions) {
 	if (!apiKey) {
 		if (!process.env.AZURE_OPENAI_API_KEY) {
 			throw new Error(
@@ -205,8 +216,9 @@ function createClient(model: Model<"azure-openai-responses">, apiKey: string, op
 	}
 
 	const { baseUrl, apiVersion } = resolveAzureConfig(model, options);
+	const AzureOpenAIClass = await getAzureOpenAIClass();
 
-	return new AzureOpenAI({
+	return new AzureOpenAIClass({
 		apiKey,
 		apiVersion,
 		dangerouslyAllowBrowser: true,

package/packages/pi-ai/src/providers/google-vertex.ts

@@ -1,9 +1,10 @@
-import {
-	type GenerateContentConfig,
-	type GenerateContentParameters,
-	GoogleGenAI,
-	type ThinkingConfig,
-	ThinkingLevel,
+// Lazy-loaded: Google GenAI SDK is imported on first use, not at startup.
+// This avoids penalizing users who don't use Google Vertex models.
+import type { GoogleGenAI } from "@google/genai";
+import type {
+	GenerateContentConfig,
+	GenerateContentParameters,
+	ThinkingConfig,
 } from "@google/genai";
 import { calculateCost } from "../models.js";
 import type {
@@ -33,6 +34,15 @@ import {
 } from "./google-shared.js";
 import { buildBaseOptions, clampReasoning } from "./simple-options.js";
 
+let _GoogleVertexClass: typeof GoogleGenAI | undefined;
+async function getGoogleVertexClass(): Promise<typeof GoogleGenAI> {
+	if (!_GoogleVertexClass) {
+		const mod = await import("@google/genai");
+		_GoogleVertexClass = mod.GoogleGenAI;
+	}
+	return _GoogleVertexClass;
+}
+
 export interface GoogleVertexOptions extends StreamOptions {
 	toolChoice?: "auto" | "none" | "any";
 	thinking?: {
@@ -46,12 +56,14 @@ export interface GoogleVertexOptions extends StreamOptions {
 
 const API_VERSION = "v1";
 
-const THINKING_LEVEL_MAP: Record<GoogleThinkingLevel, ThinkingLevel> = {
-	THINKING_LEVEL_UNSPECIFIED: ThinkingLevel.THINKING_LEVEL_UNSPECIFIED,
-	MINIMAL: ThinkingLevel.MINIMAL,
-	LOW: ThinkingLevel.LOW,
-	MEDIUM: ThinkingLevel.MEDIUM,
-	HIGH: ThinkingLevel.HIGH,
+// ThinkingLevel is a string enum where each value equals its key name.
+// Using string literals avoids importing the SDK at module load time.
+const THINKING_LEVEL_MAP: Record<GoogleThinkingLevel, string> = {
+	THINKING_LEVEL_UNSPECIFIED: "THINKING_LEVEL_UNSPECIFIED",
+	MINIMAL: "MINIMAL",
+	LOW: "LOW",
+	MEDIUM: "MEDIUM",
+	HIGH: "HIGH",
 };
 
 // Counter for generating unique tool call IDs
@@ -86,7 +98,7 @@ export const streamGoogleVertex: StreamFunction<"google-vertex", GoogleVertexOpt
 		try {
 			const project = resolveProject(options);
 			const location = resolveLocation(options);
-			const client = createClient(model, project, location, options?.headers);
+			const client = await createClient(model, project, location, options?.headers);
 			let params = buildParams(model, context, options);
 			const nextParams = await options?.onPayload?.(params, model);
 			if (nextParams !== undefined) {
@@ -318,12 +330,12 @@ export const streamSimpleGoogleVertex: StreamFunction<"google-vertex", SimpleStr
 	} satisfies GoogleVertexOptions);
 };
 
-function createClient(
+async function createClient(
 	model: Model<"google-vertex">,
 	project: string,
 	location: string,
 	optionsHeaders?: Record<string, string>,
-): GoogleGenAI {
+): Promise<GoogleGenAI> {
 	const httpOptions: { headers?: Record<string, string> } = {};
 
 	if (model.headers || optionsHeaders) {
@@ -331,8 +343,9 @@ function createClient(
 	}
 
 	const hasHttpOptions = Object.values(httpOptions).some(Boolean);
+	const GoogleGenAIClass = await getGoogleVertexClass();
 
-	return new GoogleGenAI({
+	return new GoogleGenAIClass({
 		vertexai: true,
 		project,
 		location,
@@ -393,7 +406,9 @@ function buildParams(
 	if (options.thinking?.enabled && model.reasoning) {
 		const thinkingConfig: ThinkingConfig = { includeThoughts: true };
 		if (options.thinking.level !== undefined) {
-			thinkingConfig.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level];
+			// Cast safe: string values match ThinkingLevel enum values exactly
+			// eslint-disable-next-line @typescript-eslint/no-explicit-any
+			thinkingConfig.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level] as any;
 		} else if (options.thinking.budgetTokens !== undefined) {
 			thinkingConfig.thinkingBudget = options.thinking.budgetTokens;
 		}

package/packages/pi-ai/src/providers/openai-completions.ts

@@ -1,4 +1,6 @@
-import OpenAI from "openai";
+// Lazy-loaded: OpenAI SDK is imported on first use, not at startup.
+// This avoids penalizing users who don't use OpenAI models.
+import type OpenAI from "openai";
 import type {
 	ChatCompletionAssistantMessageParam,
 	ChatCompletionChunk,
@@ -33,6 +35,15 @@ import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copi
 import { buildBaseOptions, clampReasoning } from "./simple-options.js";
 import { transformMessages } from "./transform-messages.js";
 
+let _OpenAICompletionsClass: typeof OpenAI | undefined;
+async function getOpenAICompletionsClass(): Promise<typeof OpenAI> {
+	if (!_OpenAICompletionsClass) {
+		const mod = await import("openai");
+		_OpenAICompletionsClass = mod.default;
+	}
+	return _OpenAICompletionsClass;
+}
+
 /**
  * Check if conversation messages contain tool calls or tool results.
  * This is needed because Anthropic (via proxy) requires the tools param
@@ -85,7 +96,7 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions", OpenA
 
 		try {
 			const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
-			const client = createClient(model, context, apiKey, options?.headers);
+			const client = await createClient(model, context, apiKey, options?.headers);
 			let params = buildParams(model, context, options);
 			const nextParams = await options?.onPayload?.(params, model);
 			if (nextParams !== undefined) {
@@ -327,7 +338,7 @@ export const streamSimpleOpenAICompletions: StreamFunction<"openai-completions",
 	} satisfies OpenAICompletionsOptions);
 };
 
-function createClient(
+async function createClient(
 	model: Model<"openai-completions">,
 	context: Context,
 	apiKey?: string,
@@ -358,8 +369,9 @@ function createClient(
 	}
 
 	const isZai = model.provider === "zai" || model.baseUrl.includes("api.z.ai");
+	const OpenAIClass = await getOpenAICompletionsClass();
 
-	return new OpenAI({
+	return new OpenAIClass({
 		apiKey,
 		baseURL: model.baseUrl,
 		dangerouslyAllowBrowser: true,

package/packages/pi-ai/src/providers/openai-responses.ts

@@ -1,4 +1,6 @@
-import OpenAI from "openai";
+// Lazy-loaded: OpenAI SDK is imported on first use, not at startup.
+// This avoids penalizing users who don't use OpenAI models.
+import type OpenAI from "openai";
 import type { ResponseCreateParamsStreaming } from "openai/resources/responses/responses.js";
 import { getEnvApiKey } from "../env-api-keys.js";
 import { supportsXhigh } from "../models.js";
@@ -18,6 +20,15 @@ import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copi
 import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
 import { buildBaseOptions, clampReasoning } from "./simple-options.js";
 
+let _OpenAIResponsesClass: typeof OpenAI | undefined;
+async function getOpenAIResponsesClass(): Promise<typeof OpenAI> {
+	if (!_OpenAIResponsesClass) {
+		const mod = await import("openai");
+		_OpenAIResponsesClass = mod.default;
+	}
+	return _OpenAIResponsesClass;
+}
+
 /**
  * Clamp reasoning effort for models that don't support all levels.
  * gpt-5.x models don't support "minimal" — map to "low".
@@ -98,7 +109,7 @@ export const streamOpenAIResponses: StreamFunction<"openai-responses", OpenAIRes
 		try {
 			// Create OpenAI client
 			const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
-			const client = createClient(model, context, apiKey, options?.headers);
+			const client = await createClient(model, context, apiKey, options?.headers);
 			let params = buildParams(model, context, options);
 			const nextParams = await options?.onPayload?.(params, model);
 			if (nextParams !== undefined) {
@@ -156,7 +167,7 @@ export const streamSimpleOpenAIResponses: StreamFunction<"openai-responses", Sim
 	} satisfies OpenAIResponsesOptions);
 };
 
-function createClient(
+async function createClient(
 	model: Model<"openai-responses">,
 	context: Context,
 	apiKey?: string,
@@ -186,7 +197,8 @@ function createClient(
 		Object.assign(headers, optionsHeaders);
 	}
 
-	return new OpenAI({
+	const OpenAIClass = await getOpenAIResponsesClass();
+	return new OpenAIClass({
 		apiKey,
 		baseURL: model.baseUrl,
 		dangerouslyAllowBrowser: true,

package/packages/pi-coding-agent/dist/core/agent-session.js

@@ -2041,7 +2041,7 @@ export class AgentSession {
         if (message.retryAfterMs !== undefined) {
             const cap = settings.maxDelayMs > 0 ? settings.maxDelayMs : Infinity;
             if (message.retryAfterMs > cap) {
-                // Server wants us to wait longer than our max — give up immediately
+                // Server wants us to wait longer than maxDelayMs — give up to let auto-mode handle recovery
                 this._emit({
                     type: "auto_retry_end",
                     success: false,