@bastani/pi-ai 0.9.14 → 0.9.15

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

@@ -1 +1 @@
-{"version":3,"file":"openai-responses.d.ts","sourceRoot":"","sources":["../../src/api/openai-responses.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,yCAAyC,CAAC;AAE7F,OAAO,KAAK,EASX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAEb,MAAM,aAAa,CAAC;AA2ErB,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;IAC5D,UAAU,CAAC,EAAE,6BAA6B,CAAC,aAAa,CAAC,CAAC;CAC1D;AAED;;GAEG;AACH,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,kBAAkB,EAAE,sBAAsB,CA6F7E,CAAC;AAEF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,kBAAkB,EAAE,mBAAmB,CAehF,CAAC","sourcesContent":["import OpenAI from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tOpenAIResponsesCompat,\n\tProviderEnv,\n\tProviderHeaders,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.ts\";\nimport { splitDeferredTools } from \"../utils/deferred-tools.ts\";\nimport { formatProviderError, normalizeProviderError } from \"../utils/error-body.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord } from \"../utils/headers.ts\";\nimport { getProviderEnvValue } from \"../utils/provider-env.ts\";\nimport { retryProviderRequest } from \"../utils/provider-retry.ts\";\nimport { createGrammarToolInputProperties } from \"./constrained-sampling.ts\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.ts\";\nimport { clampOpenAIPromptCacheKey } from \"./openai-prompt-cache.ts\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.ts\";\nimport { buildBaseOptions } from \"./simple-options.ts\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n// OpenAI Responses rejects max_output_tokens below 16: https://github.com/earendil-works/pi/issues/6265\nconst OPENAI_RESPONSES_MIN_OUTPUT_TOKENS = 16;\n\nfunction hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {\n\tif (!headers) return false;\n\tconst expected = name.toLowerCase();\n\tfor (const [key, value] of Object.entries(headers)) {\n\t\tif (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;\n\t}\n\treturn false;\n}\n\nfunction getClientApiKey(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): string {\n\tif (apiKey) return apiKey;\n\tif (hasHeader(headers, \"authorization\") || hasHeader(headers, \"cf-aig-authorization\")) return \"unused\";\n\tthrow new Error(`No API key for provider: ${provider}`);\n}\n\nfunction detectSessionAffinityFormat(model: Pick<Model<\"openai-responses\">, \"provider\" | \"baseUrl\">) {\n\treturn model.provider === \"openrouter\" || model.baseUrl.includes(\"openrouter.ai\") ? \"openrouter\" : \"openai\";\n}\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses PI_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction getCompat(model: Model<\"openai-responses\">): Required<OpenAIResponsesCompat> {\n\treturn {\n\t\tsupportsDeveloperRole: model.compat?.supportsDeveloperRole ?? true,\n\t\tsessionAffinityFormat: model.compat?.sessionAffinityFormat ?? detectSessionAffinityFormat(model),\n\t\tsupportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,\n\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? false,\n\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\tsupportsAdditionalTools: model.compat?.supportsAdditionalTools ?? false,\n\t\tsupportsToolSearch: model.compat?.supportsToolSearch ?? false,\n\t\tsupportsExplicitPromptCacheMode: model.compat?.supportsExplicitPromptCacheMode ?? false,\n\t};\n}\n\nfunction getPromptCacheRetention(\n\tcompat: Required<OpenAIResponsesCompat>,\n\tcacheRetention: CacheRetention,\n): \"24h\" | undefined {\n\treturn cacheRetention === \"long\" && compat.supportsLongCacheRetention ? \"24h\" : undefined;\n}\n\nfunction formatOpenAIResponsesError(error: unknown): string {\n\treturn formatProviderError(normalizeProviderError(error), \"OpenAI API error\");\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\ttoolChoice?: ResponseCreateParamsStreaming[\"tool_choice\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const stream: 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: \"pending\",\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 = getClientApiKey(model.provider, options?.apiKey, options?.headers);\n\t\t\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\t\t\tconst cacheSessionId = cacheRetention === \"none\" ? undefined : options?.sessionId;\n\t\t\tconst compat = getCompat(model);\n\t\t\tconst grammarToolInputProperties = createGrammarToolInputProperties(\n\t\t\t\tcontext.tools,\n\t\t\t\tcompat.supportsOpenAIGrammarTools,\n\t\t\t);\n\t\t\tconst client = createClient(model, context, apiKey, options?.headers, options?.fetch, cacheSessionId);\n\t\t\tlet params = buildParams(model, context, options, compat, grammarToolInputProperties);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst requestOptions = {\n\t\t\t\t...(options?.signal ? { signal: options.signal } : {}),\n\t\t\t\t...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),\n\t\t\t\tmaxRetries: 0,\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await retryProviderRequest(\n\t\t\t\t() => client.responses.create(params, requestOptions).withResponse(),\n\t\t\t\t{\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t},\n\t\t\t);\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tgrammarToolInputProperties,\n\t\t\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"pending\") {\n\t\t\t\tthrow new Error(\"OpenAI Responses stream ended without a stop reason\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage || \"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// Streaming scratch buffers are only used during parsing; never persist them.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t\tdelete (block as { customInput?: unknown }).customInput;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = formatOpenAIResponsesError(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimple: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tgetClientApiKey(model.provider, options?.apiKey, options?.headers);\n\n\tconst base = buildBaseOptions(model, context, options, options?.apiKey);\n\tconst clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;\n\tconst reasoningEffort = clampedReasoning === \"off\" ? undefined : clampedReasoning;\n\n\treturn stream(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\tapiKey: string,\n\toptionsHeaders?: ProviderHeaders,\n\tfetch?: typeof globalThis.fetch,\n\tsessionId?: string,\n) {\n\tconst compat = getCompat(model);\n\tconst headers: ProviderHeaders = { ...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\tif (sessionId) {\n\t\tif (compat.sessionAffinityFormat === \"openrouter\") {\n\t\t\theaders[\"x-session-id\"] = sessionId;\n\t\t} else {\n\t\t\tif (compat.sessionAffinityFormat === \"openai\") {\n\t\t\t\theaders.session_id = sessionId;\n\t\t\t}\n\t\t\theaders[\"x-client-request-id\"] = sessionId;\n\t\t}\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\tfetch,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions: OpenAIResponsesOptions | undefined,\n\tcompat: Required<OpenAIResponsesCompat> = getCompat(model),\n\tgrammarToolInputProperties: ReadonlyMap<string, string> = createGrammarToolInputProperties(\n\t\tcontext.tools,\n\t\tcompat.supportsOpenAIGrammarTools,\n\t),\n) {\n\tconst deferredToolsMode = compat.supportsAdditionalTools\n\t\t? \"additional-tools\"\n\t\t: compat.supportsToolSearch\n\t\t\t? \"tool-search\"\n\t\t\t: undefined;\n\tconst toolPlacement = splitDeferredTools(context, deferredToolsMode !== undefined);\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS, {\n\t\tgrammarToolInputProperties,\n\t\tdeferredTools: toolPlacement.deferred,\n\t\tdeferredToolsMode,\n\t\ttoolOptions: {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t},\n\t});\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\tconst disableImplicitPromptCache = cacheRetention === \"none\" && compat.supportsExplicitPromptCacheMode;\n\tconst params: ResponseCreateParamsStreaming & { prompt_cache_options?: { mode: \"explicit\" } } = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),\n\t\tprompt_cache_retention: getPromptCacheRetention(compat, cacheRetention),\n\t\tprompt_cache_options: disableImplicitPromptCache ? { mode: \"explicit\" } : undefined,\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = Math.max(options.maxTokens, OPENAI_RESPONSES_MIN_OUTPUT_TOKENS);\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (toolPlacement.immediate.length > 0) {\n\t\tparams.tools = convertResponsesTools(toolPlacement.immediate, {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t});\n\t}\n\n\tif (options?.toolChoice !== undefined) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort || options?.reasoningSummary) {\n\t\t\tconst effort = options?.reasoningEffort\n\t\t\t\t? (model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort)\n\t\t\t\t: \"medium\";\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\" && model.thinkingLevelMap?.off !== null) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: (model.thinkingLevelMap?.off ?? \"none\") as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t};\n\t\t}\n\t\tif (model.provider === \"xai\") params.include = [\"reasoning.encrypted_content\"];\n\t}\n\n\t// Last so custom keys override the named request fields.\n\tif (options?.samplingParams) {\n\t\tObject.assign(params, options.samplingParams);\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\t\treturn model.id === \"gpt-5.5\" ? 2.5 : 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(\n\tusage: Usage,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n) {\n\tconst multiplier = getServiceTierCostMultiplier(model, 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/api/openai-responses.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,6BAA6B,EAAE,MAAM,yCAAyC,CAAC;AAE7F,OAAO,KAAK,EASX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAEb,MAAM,aAAa,CAAC;AAiFrB,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;IAC5D,UAAU,CAAC,EAAE,6BAA6B,CAAC,aAAa,CAAC,CAAC;CAC1D;AAED;;GAEG;AACH,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,kBAAkB,EAAE,sBAAsB,CA+G7E,CAAC;AAEF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,kBAAkB,EAAE,mBAAmB,CAkBhF,CAAC","sourcesContent":["import OpenAI from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tOpenAIResponsesCompat,\n\tProviderEnv,\n\tProviderHeaders,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.ts\";\nimport { splitDeferredTools } from \"../utils/deferred-tools.ts\";\nimport { formatProviderError, normalizeProviderError } from \"../utils/error-body.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord } from \"../utils/headers.ts\";\nimport { getPiUserAgent } from \"../utils/pi-user-agent.ts\";\nimport { getProviderEnvValue } from \"../utils/provider-env.ts\";\nimport { retryProviderRequest } from \"../utils/provider-retry.ts\";\nimport { createStreamDeadline, withStreamDeadline } from \"../utils/stream-deadline.ts\";\nimport { createGrammarToolInputProperties } from \"./constrained-sampling.ts\";\nimport {\n\tbuildCopilotDynamicHeaders,\n\thasCopilotVisionInput,\n\tpreserveCopilotIntegrationHeader,\n} from \"./github-copilot-headers.ts\";\nimport { clampOpenAIPromptCacheKey } from \"./openai-prompt-cache.ts\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.ts\";\nimport { buildBaseOptions } from \"./simple-options.ts\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n// OpenAI Responses rejects max_output_tokens below 16: https://github.com/earendil-works/pi/issues/6265\nconst OPENAI_RESPONSES_MIN_OUTPUT_TOKENS = 16;\n\nfunction hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {\n\tif (!headers) return false;\n\tconst expected = name.toLowerCase();\n\tfor (const [key, value] of Object.entries(headers)) {\n\t\tif (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;\n\t}\n\treturn false;\n}\n\nfunction getClientApiKey(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): string {\n\tif (apiKey) return apiKey;\n\tif (hasHeader(headers, \"authorization\") || hasHeader(headers, \"cf-aig-authorization\")) return \"unused\";\n\tthrow new Error(`No API key for provider: ${provider}`);\n}\n\nfunction detectSessionAffinityFormat(model: Pick<Model<\"openai-responses\">, \"provider\" | \"baseUrl\">) {\n\treturn model.provider === \"openrouter\" || model.baseUrl.includes(\"openrouter.ai\") ? \"openrouter\" : \"openai\";\n}\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses PI_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction getCompat(model: Model<\"openai-responses\">): Required<OpenAIResponsesCompat> {\n\treturn {\n\t\tsupportsDeveloperRole: model.compat?.supportsDeveloperRole ?? true,\n\t\tsessionAffinityFormat: model.compat?.sessionAffinityFormat ?? detectSessionAffinityFormat(model),\n\t\tsupportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,\n\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? false,\n\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\tsupportsAdditionalTools: model.compat?.supportsAdditionalTools ?? false,\n\t\tsupportsToolSearch: model.compat?.supportsToolSearch ?? false,\n\t\tsupportsExplicitPromptCacheMode: model.compat?.supportsExplicitPromptCacheMode ?? false,\n\t};\n}\n\nfunction getPromptCacheRetention(\n\tcompat: Required<OpenAIResponsesCompat>,\n\tcacheRetention: CacheRetention,\n): \"24h\" | undefined {\n\treturn cacheRetention === \"long\" && compat.supportsLongCacheRetention ? \"24h\" : undefined;\n}\n\nfunction formatOpenAIResponsesError(error: unknown): string {\n\treturn formatProviderError(normalizeProviderError(error), \"OpenAI API error\");\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\ttoolChoice?: ResponseCreateParamsStreaming[\"tool_choice\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const stream: 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: \"pending\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\tconst streamDeadline = createStreamDeadline(options?.streamDeadlineMs, options?.signal);\n\n\t\ttry {\n\t\t\t// Create OpenAI client\n\t\t\tconst apiKey = getClientApiKey(model.provider, options?.apiKey, options?.headers);\n\t\t\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\t\t\tconst cacheSessionId = cacheRetention === \"none\" ? undefined : options?.sessionId;\n\t\t\tconst compat = getCompat(model);\n\t\t\tconst grammarToolInputProperties = createGrammarToolInputProperties(\n\t\t\t\tcontext.tools,\n\t\t\t\tcompat.supportsOpenAIGrammarTools,\n\t\t\t);\n\t\t\tconst client = createClient(\n\t\t\t\tmodel,\n\t\t\t\tcontext,\n\t\t\t\tapiKey,\n\t\t\t\toptions?.headers,\n\t\t\t\toptions?.fetch,\n\t\t\t\tcacheSessionId,\n\t\t\t\toptions?.apiKey,\n\t\t\t);\n\t\t\tlet params = buildParams(model, context, options, compat, grammarToolInputProperties);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst requestOptions = {\n\t\t\t\t...(streamDeadline.signal ? { signal: streamDeadline.signal } : {}),\n\t\t\t\t...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),\n\t\t\t\tmaxRetries: 0,\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await retryProviderRequest(\n\t\t\t\t() => client.responses.create(params, requestOptions).withResponse(),\n\t\t\t\t{\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\t\t\t\tsignal: streamDeadline.signal,\n\t\t\t\t},\n\t\t\t);\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(\n\t\t\t\twithStreamDeadline(openaiStream, streamDeadline.deadlineMs, streamDeadline.abort),\n\t\t\t\toutput,\n\t\t\t\tstream,\n\t\t\t\tmodel,\n\t\t\t\t{\n\t\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\t\tgrammarToolInputProperties,\n\t\t\t\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\t\t\t},\n\t\t\t);\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"pending\") {\n\t\t\t\tthrow new Error(\"OpenAI Responses stream ended without a stop reason\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage || \"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// Streaming scratch buffers are only used during parsing; never persist them.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t\tdelete (block as { customInput?: unknown }).customInput;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = formatOpenAIResponsesError(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t} finally {\n\t\t\tstreamDeadline.cleanup();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimple: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tgetClientApiKey(model.provider, options?.apiKey, options?.headers);\n\n\tconst base = {\n\t\t...buildBaseOptions(model, context, options, options?.apiKey),\n\t\ttoolChoice: options?.toolChoice,\n\t} satisfies OpenAIResponsesOptions;\n\tconst clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;\n\tconst reasoningEffort = clampedReasoning === \"off\" ? undefined : clampedReasoning;\n\n\treturn stream(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\tapiKey: string,\n\toptionsHeaders?: ProviderHeaders,\n\tfetch?: typeof globalThis.fetch,\n\tsessionId?: string,\n\tcopilotApiKey?: string,\n) {\n\tconst compat = getCompat(model);\n\tconst headers: ProviderHeaders = { \"User-Agent\": getPiUserAgent(), ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = preserveCopilotIntegrationHeader(\n\t\t\tmodel.headers,\n\t\t\tbuildCopilotDynamicHeaders({\n\t\t\t\tmessages: context.messages,\n\t\t\t\thasImages,\n\t\t\t\tapiKey: copilotApiKey,\n\t\t\t}),\n\t\t);\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\tif (sessionId) {\n\t\tif (compat.sessionAffinityFormat === \"openrouter\") {\n\t\t\theaders[\"x-session-id\"] = sessionId;\n\t\t} else {\n\t\t\tif (compat.sessionAffinityFormat === \"openai\") {\n\t\t\t\theaders.session_id = sessionId;\n\t\t\t}\n\t\t\theaders[\"x-client-request-id\"] = sessionId;\n\t\t}\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\tfetch,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions: OpenAIResponsesOptions | undefined,\n\tcompat: Required<OpenAIResponsesCompat> = getCompat(model),\n\tgrammarToolInputProperties: ReadonlyMap<string, string> = createGrammarToolInputProperties(\n\t\tcontext.tools,\n\t\tcompat.supportsOpenAIGrammarTools,\n\t),\n) {\n\tconst deferredToolsMode = compat.supportsAdditionalTools\n\t\t? \"additional-tools\"\n\t\t: compat.supportsToolSearch\n\t\t\t? \"tool-search\"\n\t\t\t: undefined;\n\tconst toolPlacement = splitDeferredTools(context, deferredToolsMode !== undefined);\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS, {\n\t\tgrammarToolInputProperties,\n\t\tdeferredTools: toolPlacement.deferred,\n\t\tdeferredToolsMode,\n\t\ttoolOptions: {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t},\n\t});\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\tconst disableImplicitPromptCache = cacheRetention === \"none\" && compat.supportsExplicitPromptCacheMode;\n\tconst params: ResponseCreateParamsStreaming & { prompt_cache_options?: { mode: \"explicit\" } } = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),\n\t\tprompt_cache_retention: getPromptCacheRetention(compat, cacheRetention),\n\t\tprompt_cache_options: disableImplicitPromptCache ? { mode: \"explicit\" } : undefined,\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = Math.max(options.maxTokens, OPENAI_RESPONSES_MIN_OUTPUT_TOKENS);\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (toolPlacement.immediate.length > 0) {\n\t\tparams.tools = convertResponsesTools(toolPlacement.immediate, {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t});\n\t}\n\n\tif (options?.toolChoice !== undefined) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort || options?.reasoningSummary) {\n\t\t\tconst effort = options?.reasoningEffort\n\t\t\t\t? (model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort)\n\t\t\t\t: \"medium\";\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\" && model.thinkingLevelMap?.off !== null) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: (model.thinkingLevelMap?.off ?? \"none\") as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t};\n\t\t}\n\t\tif (model.provider === \"xai\") params.include = [\"reasoning.encrypted_content\"];\n\t}\n\n\t// Last so custom keys override the named request fields.\n\tif (options?.samplingParams) {\n\t\tObject.assign(params, options.samplingParams);\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\t\treturn model.id === \"gpt-5.5\" ? 2.5 : 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(\n\tusage: Usage,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n) {\n\tconst multiplier = getServiceTierCostMultiplier(model, 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/api/openai-responses.js

@@ -4,10 +4,12 @@ import { splitDeferredTools } from "../utils/deferred-tools.js";
 import { formatProviderError, normalizeProviderError } from "../utils/error-body.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { headersToRecord } from "../utils/headers.js";
+import { getPiUserAgent } from "../utils/pi-user-agent.js";
 import { getProviderEnvValue } from "../utils/provider-env.js";
 import { retryProviderRequest } from "../utils/provider-retry.js";
+import { createStreamDeadline, withStreamDeadline } from "../utils/stream-deadline.js";
 import { createGrammarToolInputProperties } from "./constrained-sampling.js";
-import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
+import { buildCopilotDynamicHeaders, hasCopilotVisionInput, preserveCopilotIntegrationHeader, } from "./github-copilot-headers.js";
 import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.js";
 import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
 import { buildBaseOptions } from "./simple-options.js";
@@ -89,6 +91,7 @@ export const stream = (model, context, options) => {
             stopReason: "pending",
             timestamp: Date.now(),
         };
+        const streamDeadline = createStreamDeadline(options?.streamDeadlineMs, options?.signal);
         try {
             // Create OpenAI client
             const apiKey = getClientApiKey(model.provider, options?.apiKey, options?.headers);
@@ -96,25 +99,25 @@ export const stream = (model, context, options) => {
             const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
             const compat = getCompat(model);
             const grammarToolInputProperties = createGrammarToolInputProperties(context.tools, compat.supportsOpenAIGrammarTools);
-            const client = createClient(model, context, apiKey, options?.headers, options?.fetch, cacheSessionId);
+            const client = createClient(model, context, apiKey, options?.headers, options?.fetch, cacheSessionId, options?.apiKey);
             let params = buildParams(model, context, options, compat, grammarToolInputProperties);
             const nextParams = await options?.onPayload?.(params, model);
             if (nextParams !== undefined) {
                 params = nextParams;
             }
             const requestOptions = {
-                ...(options?.signal ? { signal: options.signal } : {}),
+                ...(streamDeadline.signal ? { signal: streamDeadline.signal } : {}),
                 ...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),
                 maxRetries: 0,
             };
             const { data: openaiStream, response } = await retryProviderRequest(() => client.responses.create(params, requestOptions).withResponse(), {
                 maxRetries: options?.maxRetries,
                 maxRetryDelayMs: options?.maxRetryDelayMs,
-                signal: options?.signal,
+                signal: streamDeadline.signal,
             });
             await options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);
             stream.push({ type: "start", partial: output });
-            await processResponsesStream(openaiStream, output, stream, model, {
+            await processResponsesStream(withStreamDeadline(openaiStream, streamDeadline.deadlineMs, streamDeadline.abort), output, stream, model, {
                 serviceTier: options?.serviceTier,
                 grammarToolInputProperties,
                 applyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),
@@ -143,12 +146,18 @@ export const stream = (model, context, options) => {
             stream.push({ type: "error", reason: output.stopReason, error: output });
             stream.end();
         }
+        finally {
+            streamDeadline.cleanup();
+        }
     })();
     return stream;
 };
 export const streamSimple = (model, context, options) => {
     getClientApiKey(model.provider, options?.apiKey, options?.headers);
-    const base = buildBaseOptions(model, context, options, options?.apiKey);
+    const base = {
+        ...buildBaseOptions(model, context, options, options?.apiKey),
+        toolChoice: options?.toolChoice,
+    };
     const clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;
     const reasoningEffort = clampedReasoning === "off" ? undefined : clampedReasoning;
     return stream(model, context, {
@@ -156,15 +165,16 @@ export const streamSimple = (model, context, options) => {
         reasoningEffort,
     });
 };
-function createClient(model, context, apiKey, optionsHeaders, fetch, sessionId) {
+function createClient(model, context, apiKey, optionsHeaders, fetch, sessionId, copilotApiKey) {
     const compat = getCompat(model);
-    const headers = { ...model.headers };
+    const headers = { "User-Agent": getPiUserAgent(), ...model.headers };
     if (model.provider === "github-copilot") {
         const hasImages = hasCopilotVisionInput(context.messages);
-        const copilotHeaders = buildCopilotDynamicHeaders({
+        const copilotHeaders = preserveCopilotIntegrationHeader(model.headers, buildCopilotDynamicHeaders({
             messages: context.messages,
             hasImages,
-        });
+            apiKey: copilotApiKey,
+        }));
         Object.assign(headers, copilotHeaders);
     }
     if (sessionId) {

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

@@ -1 +1 @@
-{"version":3,"file":"openai-responses.js","sourceRoot":"","sources":["../../src/api/openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAE5B,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAelD,OAAO,EAAE,kBAAkB,EAAE,MAAM,4BAA4B,CAAC;AAChE,OAAO,EAAE,mBAAmB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrF,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAC/D,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,gCAAgC,EAAE,MAAM,2BAA2B,CAAC;AAC7E,OAAO,EAAE,0BAA0B,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AAChG,OAAO,EAAE,yBAAyB,EAAE,MAAM,0BAA0B,CAAC;AACrE,OAAO,EAAE,wBAAwB,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,8BAA8B,CAAC;AACvH,OAAO,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAC;AAEvD,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC;AACnF,wGAAwG;AACxG,MAAM,kCAAkC,GAAG,EAAE,CAAC;AAE9C,SAAS,SAAS,CAAC,OAAoC,EAAE,IAAY,EAAW;IAC/E,IAAI,CAAC,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3B,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;IACpC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACpD,IAAI,GAAG,CAAC,WAAW,EAAE,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,IAAI,CAAC;IAC9F,CAAC;IACD,OAAO,KAAK,CAAC;AAAA,CACb;AAED,SAAS,eAAe,CAAC,QAAgB,EAAE,MAA0B,EAAE,OAAoC,EAAU;IACpH,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1B,IAAI,SAAS,CAAC,OAAO,EAAE,eAAe,CAAC,IAAI,SAAS,CAAC,OAAO,EAAE,sBAAsB,CAAC;QAAE,OAAO,QAAQ,CAAC;IACvG,MAAM,IAAI,KAAK,CAAC,4BAA4B,QAAQ,EAAE,CAAC,CAAC;AAAA,CACxD;AAED,SAAS,2BAA2B,CAAC,KAA8D,EAAE;IACpG,OAAO,KAAK,CAAC,QAAQ,KAAK,YAAY,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC5G;AAED;;;GAGG;AACH,SAAS,qBAAqB,CAAC,cAA+B,EAAE,GAAiB,EAAkB;IAClG,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAAA,CACf;AAED,SAAS,SAAS,CAAC,KAAgC,EAAmC;IACrF,OAAO;QACN,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,IAAI;QAClE,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,2BAA2B,CAAC,KAAK,CAAC;QAChG,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,IAAI;QAC5E,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK;QAC7D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK;QAC7E,uBAAuB,EAAE,KAAK,CAAC,MAAM,EAAE,uBAAuB,IAAI,KAAK;QACvE,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK;QAC7D,+BAA+B,EAAE,KAAK,CAAC,MAAM,EAAE,+BAA+B,IAAI,KAAK;KACvF,CAAC;AAAA,CACF;AAED,SAAS,uBAAuB,CAC/B,MAAuC,EACvC,cAA8B,EACV;IACpB,OAAO,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,0BAA0B,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAC1F;AAED,SAAS,0BAA0B,CAAC,KAAc,EAAU;IAC3D,OAAO,mBAAmB,CAAC,sBAAsB,CAAC,KAAK,CAAC,EAAE,kBAAkB,CAAC,CAAC;AAAA,CAC9E;AAUD;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAA+D,CACjF,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,SAAS;YACrB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,uBAAuB;YACvB,MAAM,MAAM,GAAG,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAClF,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;YACpF,MAAM,cAAc,GAAG,cAAc,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC;YAClF,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;YAChC,MAAM,0BAA0B,GAAG,gCAAgC,CAClE,OAAO,CAAC,KAAK,EACb,MAAM,CAAC,0BAA0B,CACjC,CAAC;YACF,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,cAAc,CAAC,CAAC;YACtG,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,0BAA0B,CAAC,CAAC;YACtF,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAA2C,CAAC;YACtD,CAAC;YACD,MAAM,cAAc,GAAG;gBACtB,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACtD,GAAG,CAAC,OAAO,EAAE,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3E,UAAU,EAAE,CAAC;aACb,CAAC;YACF,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,oBAAoB,CAClE,GAAG,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,YAAY,EAAE,EACpE;gBACC,UAAU,EAAE,OAAO,EAAE,UAAU;gBAC/B,eAAe,EAAE,OAAO,EAAE,eAAe;gBACzC,MAAM,EAAE,OAAO,EAAE,MAAM;aACvB,CACD,CAAC;YACF,MAAM,OAAO,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,eAAe,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;YAC5G,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAEhD,MAAM,sBAAsB,CAAC,YAAY,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE;gBACjE,WAAW,EAAE,OAAO,EAAE,WAAW;gBACjC,0BAA0B;gBAC1B,uBAAuB,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,EAAE,CAAC,uBAAuB,CAAC,KAAK,EAAE,WAAW,EAAE,KAAK,CAAC;aACnG,CAAC,CAAC;YAEH,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;gBACrC,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;YACxE,CAAC;YACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,YAAY,IAAI,2BAA2B,CAAC,CAAC;YACrE,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpC,OAAQ,KAA4B,CAAC,KAAK,CAAC;gBAC3C,8EAA8E;gBAC9E,OAAQ,KAAkC,CAAC,WAAW,CAAC;gBACvD,OAAQ,KAAmC,CAAC,WAAW,CAAC;YACzD,CAAC;YACD,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,0BAA0B,CAAC,KAAK,CAAC,CAAC;YACxD,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,YAAY,GAA4D,CACpF,KAAgC,EAChC,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAEnE,MAAM,IAAI,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACxE,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvG,MAAM,eAAe,GAAG,gBAAgB,KAAK,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IAElF,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE;QAC7B,GAAG,IAAI;QACP,eAAe;KACkB,CAAC,CAAC;AAAA,CACpC,CAAC;AAEF,SAAS,YAAY,CACpB,KAAgC,EAChC,OAAgB,EAChB,MAAc,EACd,cAAgC,EAChC,KAA+B,EAC/B,SAAkB,EACjB;IACD,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;IAChC,MAAM,OAAO,GAAoB,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IACtD,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,IAAI,SAAS,EAAE,CAAC;QACf,IAAI,MAAM,CAAC,qBAAqB,KAAK,YAAY,EAAE,CAAC;YACnD,OAAO,CAAC,cAAc,CAAC,GAAG,SAAS,CAAC;QACrC,CAAC;aAAM,CAAC;YACP,IAAI,MAAM,CAAC,qBAAqB,KAAK,QAAQ,EAAE,CAAC;gBAC/C,OAAO,CAAC,UAAU,GAAG,SAAS,CAAC;YAChC,CAAC;YACD,OAAO,CAAC,qBAAqB,CAAC,GAAG,SAAS,CAAC;QAC5C,CAAC;IACF,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,KAAK;QACL,cAAc,EAAE,OAAO;KACvB,CAAC,CAAC;AAAA,CACH;AAED,SAAS,WAAW,CACnB,KAAgC,EAChC,OAAgB,EAChB,OAA2C,EAC3C,MAAM,GAAoC,SAAS,CAAC,KAAK,CAAC,EAC1D,0BAA0B,GAAgC,gCAAgC,CACzF,OAAO,CAAC,KAAK,EACb,MAAM,CAAC,0BAA0B,CACjC,EACA;IACD,MAAM,iBAAiB,GAAG,MAAM,CAAC,uBAAuB;QACvD,CAAC,CAAC,kBAAkB;QACpB,CAAC,CAAC,MAAM,CAAC,kBAAkB;YAC1B,CAAC,CAAC,aAAa;YACf,CAAC,CAAC,SAAS,CAAC;IACd,MAAM,aAAa,GAAG,kBAAkB,CAAC,OAAO,EAAE,iBAAiB,KAAK,SAAS,CAAC,CAAC;IACnF,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,0BAA0B,EAAE;QACrF,0BAA0B;QAC1B,aAAa,EAAE,aAAa,CAAC,QAAQ;QACrC,iBAAiB;QACjB,WAAW,EAAE;YACZ,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;YAC7C,0BAA0B,EAAE,MAAM,CAAC,0BAA0B;SAC7D;KACD,CAAC,CAAC;IAEH,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;IACpF,MAAM,0BAA0B,GAAG,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,+BAA+B,CAAC;IACvG,MAAM,MAAM,GAAoF;QAC/F,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,yBAAyB,CAAC,OAAO,EAAE,SAAS,CAAC;QACvG,sBAAsB,EAAE,uBAAuB,CAAC,MAAM,EAAE,cAAc,CAAC;QACvE,oBAAoB,EAAE,0BAA0B,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,SAAS;QACnF,KAAK,EAAE,KAAK;KACZ,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,kCAAkC,CAAC,CAAC;IAC5F,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,aAAa,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxC,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,aAAa,CAAC,SAAS,EAAE;YAC7D,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;YAC7C,0BAA0B,EAAE,MAAM,CAAC,0BAA0B;SAC7D,CAAC,CAAC;IACJ,CAAC;IAED,IAAI,OAAO,EAAE,UAAU,KAAK,SAAS,EAAE,CAAC;QACvC,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,UAAU,CAAC;IACzC,CAAC;IAED,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,OAAO,EAAE,eAAe,IAAI,OAAO,EAAE,gBAAgB,EAAE,CAAC;YAC3D,MAAM,MAAM,GAAG,OAAO,EAAE,eAAe;gBACtC,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,eAAe,CAAC;gBAChF,CAAC,CAAC,QAAQ,CAAC;YACZ,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,MAAwD;gBAChE,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,IAAI,KAAK,CAAC,gBAAgB,EAAE,GAAG,KAAK,IAAI,EAAE,CAAC;YACxF,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,CAAC,KAAK,CAAC,gBAAgB,EAAE,GAAG,IAAI,MAAM,CAAmD;aACjG,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,QAAQ,KAAK,KAAK;YAAE,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;IAChF,CAAC;IAED,yDAAyD;IACzD,IAAI,OAAO,EAAE,cAAc,EAAE,CAAC;QAC7B,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,cAAc,CAAC,CAAC;IAC/C,CAAC;IAED,OAAO,MAAM,CAAC;AAAA,CACd;AAED,SAAS,4BAA4B,CACpC,KAA4C,EAC5C,WAAsE,EAC7D;IACT,QAAQ,WAAW,EAAE,CAAC;QACrB,KAAK,MAAM;YACV,OAAO,GAAG,CAAC;QACZ,KAAK,UAAU;YACd,OAAO,KAAK,CAAC,EAAE,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACzC;YACC,OAAO,CAAC,CAAC;IACX,CAAC;AAAA,CACD;AAED,SAAS,uBAAuB,CAC/B,KAAY,EACZ,WAAsE,EACtE,KAA4C,EAC3C;IACD,MAAM,UAAU,GAAG,4BAA4B,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;IACpE,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 { clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tOpenAIResponsesCompat,\n\tProviderEnv,\n\tProviderHeaders,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.ts\";\nimport { splitDeferredTools } from \"../utils/deferred-tools.ts\";\nimport { formatProviderError, normalizeProviderError } from \"../utils/error-body.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord } from \"../utils/headers.ts\";\nimport { getProviderEnvValue } from \"../utils/provider-env.ts\";\nimport { retryProviderRequest } from \"../utils/provider-retry.ts\";\nimport { createGrammarToolInputProperties } from \"./constrained-sampling.ts\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.ts\";\nimport { clampOpenAIPromptCacheKey } from \"./openai-prompt-cache.ts\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.ts\";\nimport { buildBaseOptions } from \"./simple-options.ts\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n// OpenAI Responses rejects max_output_tokens below 16: https://github.com/earendil-works/pi/issues/6265\nconst OPENAI_RESPONSES_MIN_OUTPUT_TOKENS = 16;\n\nfunction hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {\n\tif (!headers) return false;\n\tconst expected = name.toLowerCase();\n\tfor (const [key, value] of Object.entries(headers)) {\n\t\tif (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;\n\t}\n\treturn false;\n}\n\nfunction getClientApiKey(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): string {\n\tif (apiKey) return apiKey;\n\tif (hasHeader(headers, \"authorization\") || hasHeader(headers, \"cf-aig-authorization\")) return \"unused\";\n\tthrow new Error(`No API key for provider: ${provider}`);\n}\n\nfunction detectSessionAffinityFormat(model: Pick<Model<\"openai-responses\">, \"provider\" | \"baseUrl\">) {\n\treturn model.provider === \"openrouter\" || model.baseUrl.includes(\"openrouter.ai\") ? \"openrouter\" : \"openai\";\n}\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses PI_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction getCompat(model: Model<\"openai-responses\">): Required<OpenAIResponsesCompat> {\n\treturn {\n\t\tsupportsDeveloperRole: model.compat?.supportsDeveloperRole ?? true,\n\t\tsessionAffinityFormat: model.compat?.sessionAffinityFormat ?? detectSessionAffinityFormat(model),\n\t\tsupportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,\n\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? false,\n\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\tsupportsAdditionalTools: model.compat?.supportsAdditionalTools ?? false,\n\t\tsupportsToolSearch: model.compat?.supportsToolSearch ?? false,\n\t\tsupportsExplicitPromptCacheMode: model.compat?.supportsExplicitPromptCacheMode ?? false,\n\t};\n}\n\nfunction getPromptCacheRetention(\n\tcompat: Required<OpenAIResponsesCompat>,\n\tcacheRetention: CacheRetention,\n): \"24h\" | undefined {\n\treturn cacheRetention === \"long\" && compat.supportsLongCacheRetention ? \"24h\" : undefined;\n}\n\nfunction formatOpenAIResponsesError(error: unknown): string {\n\treturn formatProviderError(normalizeProviderError(error), \"OpenAI API error\");\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\ttoolChoice?: ResponseCreateParamsStreaming[\"tool_choice\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const stream: 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: \"pending\",\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 = getClientApiKey(model.provider, options?.apiKey, options?.headers);\n\t\t\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\t\t\tconst cacheSessionId = cacheRetention === \"none\" ? undefined : options?.sessionId;\n\t\t\tconst compat = getCompat(model);\n\t\t\tconst grammarToolInputProperties = createGrammarToolInputProperties(\n\t\t\t\tcontext.tools,\n\t\t\t\tcompat.supportsOpenAIGrammarTools,\n\t\t\t);\n\t\t\tconst client = createClient(model, context, apiKey, options?.headers, options?.fetch, cacheSessionId);\n\t\t\tlet params = buildParams(model, context, options, compat, grammarToolInputProperties);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst requestOptions = {\n\t\t\t\t...(options?.signal ? { signal: options.signal } : {}),\n\t\t\t\t...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),\n\t\t\t\tmaxRetries: 0,\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await retryProviderRequest(\n\t\t\t\t() => client.responses.create(params, requestOptions).withResponse(),\n\t\t\t\t{\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t},\n\t\t\t);\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(openaiStream, output, stream, model, {\n\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\tgrammarToolInputProperties,\n\t\t\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\t\t});\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"pending\") {\n\t\t\t\tthrow new Error(\"OpenAI Responses stream ended without a stop reason\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage || \"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// Streaming scratch buffers are only used during parsing; never persist them.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t\tdelete (block as { customInput?: unknown }).customInput;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = formatOpenAIResponsesError(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimple: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tgetClientApiKey(model.provider, options?.apiKey, options?.headers);\n\n\tconst base = buildBaseOptions(model, context, options, options?.apiKey);\n\tconst clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;\n\tconst reasoningEffort = clampedReasoning === \"off\" ? undefined : clampedReasoning;\n\n\treturn stream(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\tapiKey: string,\n\toptionsHeaders?: ProviderHeaders,\n\tfetch?: typeof globalThis.fetch,\n\tsessionId?: string,\n) {\n\tconst compat = getCompat(model);\n\tconst headers: ProviderHeaders = { ...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\tif (sessionId) {\n\t\tif (compat.sessionAffinityFormat === \"openrouter\") {\n\t\t\theaders[\"x-session-id\"] = sessionId;\n\t\t} else {\n\t\t\tif (compat.sessionAffinityFormat === \"openai\") {\n\t\t\t\theaders.session_id = sessionId;\n\t\t\t}\n\t\t\theaders[\"x-client-request-id\"] = sessionId;\n\t\t}\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\tfetch,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions: OpenAIResponsesOptions | undefined,\n\tcompat: Required<OpenAIResponsesCompat> = getCompat(model),\n\tgrammarToolInputProperties: ReadonlyMap<string, string> = createGrammarToolInputProperties(\n\t\tcontext.tools,\n\t\tcompat.supportsOpenAIGrammarTools,\n\t),\n) {\n\tconst deferredToolsMode = compat.supportsAdditionalTools\n\t\t? \"additional-tools\"\n\t\t: compat.supportsToolSearch\n\t\t\t? \"tool-search\"\n\t\t\t: undefined;\n\tconst toolPlacement = splitDeferredTools(context, deferredToolsMode !== undefined);\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS, {\n\t\tgrammarToolInputProperties,\n\t\tdeferredTools: toolPlacement.deferred,\n\t\tdeferredToolsMode,\n\t\ttoolOptions: {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t},\n\t});\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\tconst disableImplicitPromptCache = cacheRetention === \"none\" && compat.supportsExplicitPromptCacheMode;\n\tconst params: ResponseCreateParamsStreaming & { prompt_cache_options?: { mode: \"explicit\" } } = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),\n\t\tprompt_cache_retention: getPromptCacheRetention(compat, cacheRetention),\n\t\tprompt_cache_options: disableImplicitPromptCache ? { mode: \"explicit\" } : undefined,\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = Math.max(options.maxTokens, OPENAI_RESPONSES_MIN_OUTPUT_TOKENS);\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (toolPlacement.immediate.length > 0) {\n\t\tparams.tools = convertResponsesTools(toolPlacement.immediate, {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t});\n\t}\n\n\tif (options?.toolChoice !== undefined) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort || options?.reasoningSummary) {\n\t\t\tconst effort = options?.reasoningEffort\n\t\t\t\t? (model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort)\n\t\t\t\t: \"medium\";\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\" && model.thinkingLevelMap?.off !== null) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: (model.thinkingLevelMap?.off ?? \"none\") as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t};\n\t\t}\n\t\tif (model.provider === \"xai\") params.include = [\"reasoning.encrypted_content\"];\n\t}\n\n\t// Last so custom keys override the named request fields.\n\tif (options?.samplingParams) {\n\t\tObject.assign(params, options.samplingParams);\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\t\treturn model.id === \"gpt-5.5\" ? 2.5 : 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(\n\tusage: Usage,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n) {\n\tconst multiplier = getServiceTierCostMultiplier(model, 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/api/openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,QAAQ,CAAC;AAE5B,OAAO,EAAE,kBAAkB,EAAE,MAAM,cAAc,CAAC;AAelD,OAAO,EAAE,kBAAkB,EAAE,MAAM,4BAA4B,CAAC;AAChE,OAAO,EAAE,mBAAmB,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AACrF,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3D,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAC;AAC/D,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,oBAAoB,EAAE,kBAAkB,EAAE,MAAM,6BAA6B,CAAC;AACvF,OAAO,EAAE,gCAAgC,EAAE,MAAM,2BAA2B,CAAC;AAC7E,OAAO,EACN,0BAA0B,EAC1B,qBAAqB,EACrB,gCAAgC,GAChC,MAAM,6BAA6B,CAAC;AACrC,OAAO,EAAE,yBAAyB,EAAE,MAAM,0BAA0B,CAAC;AACrE,OAAO,EAAE,wBAAwB,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,8BAA8B,CAAC;AACvH,OAAO,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAC;AAEvD,MAAM,0BAA0B,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC;AACnF,wGAAwG;AACxG,MAAM,kCAAkC,GAAG,EAAE,CAAC;AAE9C,SAAS,SAAS,CAAC,OAAoC,EAAE,IAAY,EAAW;IAC/E,IAAI,CAAC,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3B,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;IACpC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACpD,IAAI,GAAG,CAAC,WAAW,EAAE,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,IAAI,CAAC;IAC9F,CAAC;IACD,OAAO,KAAK,CAAC;AAAA,CACb;AAED,SAAS,eAAe,CAAC,QAAgB,EAAE,MAA0B,EAAE,OAAoC,EAAU;IACpH,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1B,IAAI,SAAS,CAAC,OAAO,EAAE,eAAe,CAAC,IAAI,SAAS,CAAC,OAAO,EAAE,sBAAsB,CAAC;QAAE,OAAO,QAAQ,CAAC;IACvG,MAAM,IAAI,KAAK,CAAC,4BAA4B,QAAQ,EAAE,CAAC,CAAC;AAAA,CACxD;AAED,SAAS,2BAA2B,CAAC,KAA8D,EAAE;IACpG,OAAO,KAAK,CAAC,QAAQ,KAAK,YAAY,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC5G;AAED;;;GAGG;AACH,SAAS,qBAAqB,CAAC,cAA+B,EAAE,GAAiB,EAAkB;IAClG,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAAA,CACf;AAED,SAAS,SAAS,CAAC,KAAgC,EAAmC;IACrF,OAAO;QACN,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,IAAI;QAClE,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,2BAA2B,CAAC,KAAK,CAAC;QAChG,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,IAAI;QAC5E,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK;QAC7D,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,KAAK;QAC7E,uBAAuB,EAAE,KAAK,CAAC,MAAM,EAAE,uBAAuB,IAAI,KAAK;QACvE,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,KAAK;QAC7D,+BAA+B,EAAE,KAAK,CAAC,MAAM,EAAE,+BAA+B,IAAI,KAAK;KACvF,CAAC;AAAA,CACF;AAED,SAAS,uBAAuB,CAC/B,MAAuC,EACvC,cAA8B,EACV;IACpB,OAAO,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,0BAA0B,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAC1F;AAED,SAAS,0BAA0B,CAAC,KAAc,EAAU;IAC3D,OAAO,mBAAmB,CAAC,sBAAsB,CAAC,KAAK,CAAC,EAAE,kBAAkB,CAAC,CAAC;AAAA,CAC9E;AAUD;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAA+D,CACjF,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,SAAS;YACrB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,MAAM,cAAc,GAAG,oBAAoB,CAAC,OAAO,EAAE,gBAAgB,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAExF,IAAI,CAAC;YACJ,uBAAuB;YACvB,MAAM,MAAM,GAAG,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAClF,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;YACpF,MAAM,cAAc,GAAG,cAAc,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC;YAClF,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;YAChC,MAAM,0BAA0B,GAAG,gCAAgC,CAClE,OAAO,CAAC,KAAK,EACb,MAAM,CAAC,0BAA0B,CACjC,CAAC;YACF,MAAM,MAAM,GAAG,YAAY,CAC1B,KAAK,EACL,OAAO,EACP,MAAM,EACN,OAAO,EAAE,OAAO,EAChB,OAAO,EAAE,KAAK,EACd,cAAc,EACd,OAAO,EAAE,MAAM,CACf,CAAC;YACF,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,0BAA0B,CAAC,CAAC;YACtF,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAA2C,CAAC;YACtD,CAAC;YACD,MAAM,cAAc,GAAG;gBACtB,GAAG,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,cAAc,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACnE,GAAG,CAAC,OAAO,EAAE,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3E,UAAU,EAAE,CAAC;aACb,CAAC;YACF,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,oBAAoB,CAClE,GAAG,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,YAAY,EAAE,EACpE;gBACC,UAAU,EAAE,OAAO,EAAE,UAAU;gBAC/B,eAAe,EAAE,OAAO,EAAE,eAAe;gBACzC,MAAM,EAAE,cAAc,CAAC,MAAM;aAC7B,CACD,CAAC;YACF,MAAM,OAAO,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,eAAe,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC;YAC5G,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAEhD,MAAM,sBAAsB,CAC3B,kBAAkB,CAAC,YAAY,EAAE,cAAc,CAAC,UAAU,EAAE,cAAc,CAAC,KAAK,CAAC,EACjF,MAAM,EACN,MAAM,EACN,KAAK,EACL;gBACC,WAAW,EAAE,OAAO,EAAE,WAAW;gBACjC,0BAA0B;gBAC1B,uBAAuB,EAAE,CAAC,KAAK,EAAE,WAAW,EAAE,EAAE,CAAC,uBAAuB,CAAC,KAAK,EAAE,WAAW,EAAE,KAAK,CAAC;aACnG,CACD,CAAC;YAEF,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;gBACrC,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;YACxE,CAAC;YACD,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,YAAY,IAAI,2BAA2B,CAAC,CAAC;YACrE,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpC,OAAQ,KAA4B,CAAC,KAAK,CAAC;gBAC3C,8EAA8E;gBAC9E,OAAQ,KAAkC,CAAC,WAAW,CAAC;gBACvD,OAAQ,KAAmC,CAAC,WAAW,CAAC;YACzD,CAAC;YACD,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,0BAA0B,CAAC,KAAK,CAAC,CAAC;YACxD,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;gBAAS,CAAC;YACV,cAAc,CAAC,OAAO,EAAE,CAAC;QAC1B,CAAC;IAAA,CACD,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AAAA,CACd,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAA4D,CACpF,KAAgC,EAChC,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAEnE,MAAM,IAAI,GAAG;QACZ,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC;QAC7D,UAAU,EAAE,OAAO,EAAE,UAAU;KACE,CAAC;IACnC,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvG,MAAM,eAAe,GAAG,gBAAgB,KAAK,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IAElF,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE;QAC7B,GAAG,IAAI;QACP,eAAe;KACkB,CAAC,CAAC;AAAA,CACpC,CAAC;AAEF,SAAS,YAAY,CACpB,KAAgC,EAChC,OAAgB,EAChB,MAAc,EACd,cAAgC,EAChC,KAA+B,EAC/B,SAAkB,EAClB,aAAsB,EACrB;IACD,MAAM,MAAM,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;IAChC,MAAM,OAAO,GAAoB,EAAE,YAAY,EAAE,cAAc,EAAE,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IACtF,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,EAAE,CAAC;QACzC,MAAM,SAAS,GAAG,qBAAqB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QAC1D,MAAM,cAAc,GAAG,gCAAgC,CACtD,KAAK,CAAC,OAAO,EACb,0BAA0B,CAAC;YAC1B,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS;YACT,MAAM,EAAE,aAAa;SACrB,CAAC,CACF,CAAC;QACF,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IACxC,CAAC;IAED,IAAI,SAAS,EAAE,CAAC;QACf,IAAI,MAAM,CAAC,qBAAqB,KAAK,YAAY,EAAE,CAAC;YACnD,OAAO,CAAC,cAAc,CAAC,GAAG,SAAS,CAAC;QACrC,CAAC;aAAM,CAAC;YACP,IAAI,MAAM,CAAC,qBAAqB,KAAK,QAAQ,EAAE,CAAC;gBAC/C,OAAO,CAAC,UAAU,GAAG,SAAS,CAAC;YAChC,CAAC;YACD,OAAO,CAAC,qBAAqB,CAAC,GAAG,SAAS,CAAC;QAC5C,CAAC;IACF,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,KAAK;QACL,cAAc,EAAE,OAAO;KACvB,CAAC,CAAC;AAAA,CACH;AAED,SAAS,WAAW,CACnB,KAAgC,EAChC,OAAgB,EAChB,OAA2C,EAC3C,MAAM,GAAoC,SAAS,CAAC,KAAK,CAAC,EAC1D,0BAA0B,GAAgC,gCAAgC,CACzF,OAAO,CAAC,KAAK,EACb,MAAM,CAAC,0BAA0B,CACjC,EACA;IACD,MAAM,iBAAiB,GAAG,MAAM,CAAC,uBAAuB;QACvD,CAAC,CAAC,kBAAkB;QACpB,CAAC,CAAC,MAAM,CAAC,kBAAkB;YAC1B,CAAC,CAAC,aAAa;YACf,CAAC,CAAC,SAAS,CAAC;IACd,MAAM,aAAa,GAAG,kBAAkB,CAAC,OAAO,EAAE,iBAAiB,KAAK,SAAS,CAAC,CAAC;IACnF,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,0BAA0B,EAAE;QACrF,0BAA0B;QAC1B,aAAa,EAAE,aAAa,CAAC,QAAQ;QACrC,iBAAiB;QACjB,WAAW,EAAE;YACZ,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;YAC7C,0BAA0B,EAAE,MAAM,CAAC,0BAA0B;SAC7D;KACD,CAAC,CAAC;IAEH,MAAM,cAAc,GAAG,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;IACpF,MAAM,0BAA0B,GAAG,cAAc,KAAK,MAAM,IAAI,MAAM,CAAC,+BAA+B,CAAC;IACvG,MAAM,MAAM,GAAoF;QAC/F,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,yBAAyB,CAAC,OAAO,EAAE,SAAS,CAAC;QACvG,sBAAsB,EAAE,uBAAuB,CAAC,MAAM,EAAE,cAAc,CAAC;QACvE,oBAAoB,EAAE,0BAA0B,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,SAAS;QACnF,KAAK,EAAE,KAAK;KACZ,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,kCAAkC,CAAC,CAAC;IAC5F,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,aAAa,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxC,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,aAAa,CAAC,SAAS,EAAE;YAC7D,kBAAkB,EAAE,MAAM,CAAC,kBAAkB;YAC7C,0BAA0B,EAAE,MAAM,CAAC,0BAA0B;SAC7D,CAAC,CAAC;IACJ,CAAC;IAED,IAAI,OAAO,EAAE,UAAU,KAAK,SAAS,EAAE,CAAC;QACvC,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,UAAU,CAAC;IACzC,CAAC;IAED,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,OAAO,EAAE,eAAe,IAAI,OAAO,EAAE,gBAAgB,EAAE,CAAC;YAC3D,MAAM,MAAM,GAAG,OAAO,EAAE,eAAe;gBACtC,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC,OAAO,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,eAAe,CAAC;gBAChF,CAAC,CAAC,QAAQ,CAAC;YACZ,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,MAAwD;gBAChE,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,KAAK,CAAC,QAAQ,KAAK,gBAAgB,IAAI,KAAK,CAAC,gBAAgB,EAAE,GAAG,KAAK,IAAI,EAAE,CAAC;YACxF,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,CAAC,KAAK,CAAC,gBAAgB,EAAE,GAAG,IAAI,MAAM,CAAmD;aACjG,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,QAAQ,KAAK,KAAK;YAAE,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;IAChF,CAAC;IAED,yDAAyD;IACzD,IAAI,OAAO,EAAE,cAAc,EAAE,CAAC;QAC7B,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,cAAc,CAAC,CAAC;IAC/C,CAAC;IAED,OAAO,MAAM,CAAC;AAAA,CACd;AAED,SAAS,4BAA4B,CACpC,KAA4C,EAC5C,WAAsE,EAC7D;IACT,QAAQ,WAAW,EAAE,CAAC;QACrB,KAAK,MAAM;YACV,OAAO,GAAG,CAAC;QACZ,KAAK,UAAU;YACd,OAAO,KAAK,CAAC,EAAE,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACzC;YACC,OAAO,CAAC,CAAC;IACX,CAAC;AAAA,CACD;AAED,SAAS,uBAAuB,CAC/B,KAAY,EACZ,WAAsE,EACtE,KAA4C,EAC3C;IACD,MAAM,UAAU,GAAG,4BAA4B,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;IACpE,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 { clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tOpenAIResponsesCompat,\n\tProviderEnv,\n\tProviderHeaders,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tUsage,\n} from \"../types.ts\";\nimport { splitDeferredTools } from \"../utils/deferred-tools.ts\";\nimport { formatProviderError, normalizeProviderError } from \"../utils/error-body.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord } from \"../utils/headers.ts\";\nimport { getPiUserAgent } from \"../utils/pi-user-agent.ts\";\nimport { getProviderEnvValue } from \"../utils/provider-env.ts\";\nimport { retryProviderRequest } from \"../utils/provider-retry.ts\";\nimport { createStreamDeadline, withStreamDeadline } from \"../utils/stream-deadline.ts\";\nimport { createGrammarToolInputProperties } from \"./constrained-sampling.ts\";\nimport {\n\tbuildCopilotDynamicHeaders,\n\thasCopilotVisionInput,\n\tpreserveCopilotIntegrationHeader,\n} from \"./github-copilot-headers.ts\";\nimport { clampOpenAIPromptCacheKey } from \"./openai-prompt-cache.ts\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.ts\";\nimport { buildBaseOptions } from \"./simple-options.ts\";\n\nconst OPENAI_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\"]);\n// OpenAI Responses rejects max_output_tokens below 16: https://github.com/earendil-works/pi/issues/6265\nconst OPENAI_RESPONSES_MIN_OUTPUT_TOKENS = 16;\n\nfunction hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {\n\tif (!headers) return false;\n\tconst expected = name.toLowerCase();\n\tfor (const [key, value] of Object.entries(headers)) {\n\t\tif (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;\n\t}\n\treturn false;\n}\n\nfunction getClientApiKey(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): string {\n\tif (apiKey) return apiKey;\n\tif (hasHeader(headers, \"authorization\") || hasHeader(headers, \"cf-aig-authorization\")) return \"unused\";\n\tthrow new Error(`No API key for provider: ${provider}`);\n}\n\nfunction detectSessionAffinityFormat(model: Pick<Model<\"openai-responses\">, \"provider\" | \"baseUrl\">) {\n\treturn model.provider === \"openrouter\" || model.baseUrl.includes(\"openrouter.ai\") ? \"openrouter\" : \"openai\";\n}\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses PI_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction getCompat(model: Model<\"openai-responses\">): Required<OpenAIResponsesCompat> {\n\treturn {\n\t\tsupportsDeveloperRole: model.compat?.supportsDeveloperRole ?? true,\n\t\tsessionAffinityFormat: model.compat?.sessionAffinityFormat ?? detectSessionAffinityFormat(model),\n\t\tsupportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,\n\t\tsupportsStrictMode: model.compat?.supportsStrictMode ?? false,\n\t\tsupportsOpenAIGrammarTools: model.compat?.supportsOpenAIGrammarTools ?? false,\n\t\tsupportsAdditionalTools: model.compat?.supportsAdditionalTools ?? false,\n\t\tsupportsToolSearch: model.compat?.supportsToolSearch ?? false,\n\t\tsupportsExplicitPromptCacheMode: model.compat?.supportsExplicitPromptCacheMode ?? false,\n\t};\n}\n\nfunction getPromptCacheRetention(\n\tcompat: Required<OpenAIResponsesCompat>,\n\tcacheRetention: CacheRetention,\n): \"24h\" | undefined {\n\treturn cacheRetention === \"long\" && compat.supportsLongCacheRetention ? \"24h\" : undefined;\n}\n\nfunction formatOpenAIResponsesError(error: unknown): string {\n\treturn formatProviderError(normalizeProviderError(error), \"OpenAI API error\");\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\ttoolChoice?: ResponseCreateParamsStreaming[\"tool_choice\"];\n}\n\n/**\n * Generate function for OpenAI Responses API\n */\nexport const stream: 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: \"pending\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\tconst streamDeadline = createStreamDeadline(options?.streamDeadlineMs, options?.signal);\n\n\t\ttry {\n\t\t\t// Create OpenAI client\n\t\t\tconst apiKey = getClientApiKey(model.provider, options?.apiKey, options?.headers);\n\t\t\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\t\t\tconst cacheSessionId = cacheRetention === \"none\" ? undefined : options?.sessionId;\n\t\t\tconst compat = getCompat(model);\n\t\t\tconst grammarToolInputProperties = createGrammarToolInputProperties(\n\t\t\t\tcontext.tools,\n\t\t\t\tcompat.supportsOpenAIGrammarTools,\n\t\t\t);\n\t\t\tconst client = createClient(\n\t\t\t\tmodel,\n\t\t\t\tcontext,\n\t\t\t\tapiKey,\n\t\t\t\toptions?.headers,\n\t\t\t\toptions?.fetch,\n\t\t\t\tcacheSessionId,\n\t\t\t\toptions?.apiKey,\n\t\t\t);\n\t\t\tlet params = buildParams(model, context, options, compat, grammarToolInputProperties);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as ResponseCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst requestOptions = {\n\t\t\t\t...(streamDeadline.signal ? { signal: streamDeadline.signal } : {}),\n\t\t\t\t...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),\n\t\t\t\tmaxRetries: 0,\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await retryProviderRequest(\n\t\t\t\t() => client.responses.create(params, requestOptions).withResponse(),\n\t\t\t\t{\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t\tmaxRetryDelayMs: options?.maxRetryDelayMs,\n\t\t\t\t\tsignal: streamDeadline.signal,\n\t\t\t\t},\n\t\t\t);\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\tawait processResponsesStream(\n\t\t\t\twithStreamDeadline(openaiStream, streamDeadline.deadlineMs, streamDeadline.abort),\n\t\t\t\toutput,\n\t\t\t\tstream,\n\t\t\t\tmodel,\n\t\t\t\t{\n\t\t\t\t\tserviceTier: options?.serviceTier,\n\t\t\t\t\tgrammarToolInputProperties,\n\t\t\t\t\tapplyServiceTierPricing: (usage, serviceTier) => applyServiceTierPricing(usage, serviceTier, model),\n\t\t\t\t},\n\t\t\t);\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"pending\") {\n\t\t\t\tthrow new Error(\"OpenAI Responses stream ended without a stop reason\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage || \"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// Streaming scratch buffers are only used during parsing; never persist them.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t\tdelete (block as { customInput?: unknown }).customInput;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = formatOpenAIResponsesError(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t} finally {\n\t\t\tstreamDeadline.cleanup();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimple: StreamFunction<\"openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tgetClientApiKey(model.provider, options?.apiKey, options?.headers);\n\n\tconst base = {\n\t\t...buildBaseOptions(model, context, options, options?.apiKey),\n\t\ttoolChoice: options?.toolChoice,\n\t} satisfies OpenAIResponsesOptions;\n\tconst clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;\n\tconst reasoningEffort = clampedReasoning === \"off\" ? undefined : clampedReasoning;\n\n\treturn stream(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies OpenAIResponsesOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\tapiKey: string,\n\toptionsHeaders?: ProviderHeaders,\n\tfetch?: typeof globalThis.fetch,\n\tsessionId?: string,\n\tcopilotApiKey?: string,\n) {\n\tconst compat = getCompat(model);\n\tconst headers: ProviderHeaders = { \"User-Agent\": getPiUserAgent(), ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = preserveCopilotIntegrationHeader(\n\t\t\tmodel.headers,\n\t\t\tbuildCopilotDynamicHeaders({\n\t\t\t\tmessages: context.messages,\n\t\t\t\thasImages,\n\t\t\t\tapiKey: copilotApiKey,\n\t\t\t}),\n\t\t);\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\tif (sessionId) {\n\t\tif (compat.sessionAffinityFormat === \"openrouter\") {\n\t\t\theaders[\"x-session-id\"] = sessionId;\n\t\t} else {\n\t\t\tif (compat.sessionAffinityFormat === \"openai\") {\n\t\t\t\theaders.session_id = sessionId;\n\t\t\t}\n\t\t\theaders[\"x-client-request-id\"] = sessionId;\n\t\t}\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\tfetch,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"openai-responses\">,\n\tcontext: Context,\n\toptions: OpenAIResponsesOptions | undefined,\n\tcompat: Required<OpenAIResponsesCompat> = getCompat(model),\n\tgrammarToolInputProperties: ReadonlyMap<string, string> = createGrammarToolInputProperties(\n\t\tcontext.tools,\n\t\tcompat.supportsOpenAIGrammarTools,\n\t),\n) {\n\tconst deferredToolsMode = compat.supportsAdditionalTools\n\t\t? \"additional-tools\"\n\t\t: compat.supportsToolSearch\n\t\t\t? \"tool-search\"\n\t\t\t: undefined;\n\tconst toolPlacement = splitDeferredTools(context, deferredToolsMode !== undefined);\n\tconst messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS, {\n\t\tgrammarToolInputProperties,\n\t\tdeferredTools: toolPlacement.deferred,\n\t\tdeferredToolsMode,\n\t\ttoolOptions: {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t},\n\t});\n\n\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);\n\tconst disableImplicitPromptCache = cacheRetention === \"none\" && compat.supportsExplicitPromptCacheMode;\n\tconst params: ResponseCreateParamsStreaming & { prompt_cache_options?: { mode: \"explicit\" } } = {\n\t\tmodel: model.id,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: cacheRetention === \"none\" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),\n\t\tprompt_cache_retention: getPromptCacheRetention(compat, cacheRetention),\n\t\tprompt_cache_options: disableImplicitPromptCache ? { mode: \"explicit\" } : undefined,\n\t\tstore: false,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = Math.max(options.maxTokens, OPENAI_RESPONSES_MIN_OUTPUT_TOKENS);\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (options?.serviceTier !== undefined) {\n\t\tparams.service_tier = options.serviceTier;\n\t}\n\n\tif (toolPlacement.immediate.length > 0) {\n\t\tparams.tools = convertResponsesTools(toolPlacement.immediate, {\n\t\t\tsupportsStrictMode: compat.supportsStrictMode,\n\t\t\tsupportsOpenAIGrammarTools: compat.supportsOpenAIGrammarTools,\n\t\t});\n\t}\n\n\tif (options?.toolChoice !== undefined) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort || options?.reasoningSummary) {\n\t\t\tconst effort = options?.reasoningEffort\n\t\t\t\t? (model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort)\n\t\t\t\t: \"medium\";\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: effort as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t\tsummary: options?.reasoningSummary || \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else if (model.provider !== \"github-copilot\" && model.thinkingLevelMap?.off !== null) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: (model.thinkingLevelMap?.off ?? \"none\") as NonNullable<typeof params.reasoning>[\"effort\"],\n\t\t\t};\n\t\t}\n\t\tif (model.provider === \"xai\") params.include = [\"reasoning.encrypted_content\"];\n\t}\n\n\t// Last so custom keys override the named request fields.\n\tif (options?.samplingParams) {\n\t\tObject.assign(params, options.samplingParams);\n\t}\n\n\treturn params;\n}\n\nfunction getServiceTierCostMultiplier(\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n): number {\n\tswitch (serviceTier) {\n\t\tcase \"flex\":\n\t\t\treturn 0.5;\n\t\tcase \"priority\":\n\t\t\treturn model.id === \"gpt-5.5\" ? 2.5 : 2;\n\t\tdefault:\n\t\t\treturn 1;\n\t}\n}\n\nfunction applyServiceTierPricing(\n\tusage: Usage,\n\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\tmodel: Pick<Model<\"openai-responses\">, \"id\">,\n) {\n\tconst multiplier = getServiceTierCostMultiplier(model, 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/api/pi-messages.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"pi-messages.d.ts","sourceRoot":"","sources":["../../src/api/pi-messages.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EACX,gBAAgB,EAMhB,mBAAmB,EACnB,cAAc,EACd,aAAa,EACb,aAAa,EACb,QAAQ,EACR,MAAM,aAAa,CAAC;AAOrB,MAAM,WAAW,iBAAkB,SAAQ,aAAa;IACvD,SAAS,CAAC,EAAE,aAAa,CAAC;IAC1B,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,UAAU,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC;IAC7F,0EAA0E;IAC1E,KAAK,CAAC,EAAE,OAAO,CAAC;CAChB;AAED,KAAK,eAAe,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;AACjD,KAAK,oBAAoB,GAAG,gBAAgB,CAAC,YAAY,CAAC,CAAC;AAE3D,+EAA+E;AAC/E,MAAM,MAAM,uBAAuB,GAAG;IACrC,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;IACtB,OAAO,EAAE,OAAO,CAAC;IACjB,gBAAgB,EAAE,MAAM,CAAC;IACzB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,mBAAmB,EAAE,OAAO,CAAC;CAC7B,CAAC;AAEF,2EAA2E;AAC3E,MAAM,MAAM,eAAe,GACxB;IAAE,IAAI,EAAE,OAAO,CAAA;CAAE,GACjB;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,GAC5C;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAC3D;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,gBAAgB,CAAC,EAAE,MAAM,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,GAChD;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAC/D;IACA,IAAI,EAAE,cAAc,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,EAAE,OAAO,CAAC;CAClB,GACD;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,GAC9E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAC/D;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,QAAQ,CAAA;CAAE,GAClE;IACA,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,OAAO,CAAC,oBAAoB,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS,CAAC,CAAC;IACrE,KAAK,EAAE,eAAe,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,OAAO,CAAC,EAAE,uBAAuB,CAAC;CACjC,GACD;IACA,IAAI,EAAE,OAAO,CAAC;IACd,MAAM,EAAE,OAAO,CAAC,oBAAoB,EAAE,SAAS,GAAG,OAAO,CAAC,CAAC;IAC3D,KAAK,EAAE,eAAe,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,OAAO,CAAC,EAAE,uBAAuB,CAAC;CACjC,CAAC;AAWL,qBAAa,uBAAwB,SAAQ,KAAK;IACjD,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAEpD,YAAY,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,SAAS,EAAE,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAKhG;CACD;AAiPD,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,aAAa,EAAE,iBAAiB,CA0EnE,CAAC;AAEF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,aAAa,EAAE,mBAAmB,CAY3E,CAAC","sourcesContent":["/**\n * pi-messages API implementation.\n *\n * Streams pi's own message protocol directly to a backend: the request is a\n * single POST of `{ model, context, options }` to `<baseUrl>/messages`, the\n * response is an SSE stream of serialized assistant-message events plus a\n * terminal `done`/`error` event. This is the wire protocol spoken by the\n * Radius gateway, but any backend implementing it can be used, e.g. via a\n * models.json custom provider with `\"api\": \"pi-messages\"`.\n */\n\nimport type {\n\tAssistantMessage,\n\tAssistantMessageEvent,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tProviderEnv,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tThinkingLevel,\n\tToolCall,\n} from \"../types.ts\";\nimport { appendAssistantMessageDiagnostic, createAssistantMessageDiagnostic } from \"../utils/diagnostics.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord, providerHeadersToRecord } from \"../utils/headers.ts\";\nimport { parseStreamingJson } from \"../utils/json-parse.ts\";\nimport { getProviderEnvValue } from \"../utils/provider-env.ts\";\n\nexport interface PiMessagesOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\ttoolChoice?: \"auto\" | \"none\" | \"required\" | { type: \"function\"; function: { name: string } };\n\t/** Ask the backend for debug metadata (e.g. routing response headers). */\n\tdebug?: boolean;\n}\n\ntype PiMessagesUsage = AssistantMessage[\"usage\"];\ntype PiMessagesStopReason = AssistantMessage[\"stopReason\"];\n\n/** Impact summary of a server-side message rewrite (e.g. a gateway policy). */\nexport type PiMessagesRewriteImpact = {\n\tpolicyId: string;\n\tpolicyVersion: number;\n\tchanged: boolean;\n\ttokenCountChange: number;\n\tmessageCountChange: number;\n\tsystemPromptChanged: boolean;\n};\n\n/** Serialized assistant-message event as sent by a pi-messages backend. */\nexport type PiMessagesEvent =\n\t| { type: \"start\" }\n\t| { type: \"text_start\"; contentIndex: number }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string }\n\t| { type: \"text_end\"; contentIndex: number; content: string; contentSignature?: string }\n\t| { type: \"thinking_start\"; contentIndex: number }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string }\n\t| {\n\t\t\ttype: \"thinking_end\";\n\t\t\tcontentIndex: number;\n\t\t\tcontent: string;\n\t\t\tcontentSignature?: string;\n\t\t\tredacted?: boolean;\n\t  }\n\t| { type: \"toolcall_start\"; contentIndex: number; id: string; toolName: string }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall }\n\t| {\n\t\t\ttype: \"done\";\n\t\t\treason: Extract<PiMessagesStopReason, \"stop\" | \"length\" | \"toolUse\">;\n\t\t\tusage: PiMessagesUsage;\n\t\t\tresponseId?: string;\n\t\t\trewrite?: PiMessagesRewriteImpact;\n\t  }\n\t| {\n\t\t\ttype: \"error\";\n\t\t\treason: Extract<PiMessagesStopReason, \"aborted\" | \"error\">;\n\t\t\tusage: PiMessagesUsage;\n\t\t\terrorMessage?: string;\n\t\t\tresponseId?: string;\n\t\t\trewrite?: PiMessagesRewriteImpact;\n\t  };\n\ntype PiMessagesErrorBody = {\n\terror?: {\n\t\tmessage?: unknown;\n\t\tcode?: unknown;\n\t\tdetails?: unknown;\n\t\t[key: string]: unknown;\n\t};\n};\n\nexport class PiMessagesResponseError extends Error {\n\tcode?: string;\n\treadonly diagnosticDetails: Record<string, unknown>;\n\n\tconstructor(message: string, code: string | undefined, diagnosticDetails: Record<string, unknown>) {\n\t\tsuper(message);\n\t\tthis.name = \"PiMessagesResponseError\";\n\t\tthis.code = code;\n\t\tthis.diagnosticDetails = diagnosticDetails;\n\t}\n}\n\nfunction parsePiMessagesErrorBody(body: string): PiMessagesErrorBody | undefined {\n\ttry {\n\t\tconst parsed = JSON.parse(body) as PiMessagesErrorBody | null;\n\t\tconst error = parsed?.error;\n\t\treturn parsed && typeof error === \"object\" && error !== null && !Array.isArray(error) ? parsed : undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nfunction truncateDiagnosticString(value: string): string {\n\tconst maxLength = 8192;\n\treturn value.length > maxLength ? `${value.slice(0, maxLength)}…` : value;\n}\n\nfunction formatPiMessagesResponseError(\n\tresponse: Response,\n\tbody: string,\n\terrorBody: PiMessagesErrorBody | undefined,\n): string {\n\tconst message = typeof errorBody?.error?.message === \"string\" ? errorBody.error.message : undefined;\n\tconst code = typeof errorBody?.error?.code === \"string\" ? errorBody.error.code : undefined;\n\tconst suffix = message ?? body;\n\tconst codeSuffix = code ? ` (${code})` : \"\";\n\treturn `${response.status} ${response.statusText}: ${suffix}${codeSuffix}`;\n}\n\nfunction createPiMessagesResponseError(\n\tmodel: Model<\"pi-messages\">,\n\turl: URL,\n\tresponse: Response,\n\tbody: string,\n): PiMessagesResponseError {\n\tconst errorBody = parsePiMessagesErrorBody(body);\n\tconst code = typeof errorBody?.error?.code === \"string\" ? errorBody.error.code : undefined;\n\treturn new PiMessagesResponseError(formatPiMessagesResponseError(response, body, errorBody), code, {\n\t\tversion: 1,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\turl: url.toString(),\n\t\tstatus: response.status,\n\t\tstatusText: response.statusText,\n\t\terror: errorBody?.error,\n\t\tbody: errorBody ? undefined : truncateDiagnosticString(body),\n\t\ttimestampMs: Date.now(),\n\t});\n}\n\nfunction createEmptyUsage(): PiMessagesUsage {\n\treturn {\n\t\tinput: 0,\n\t\toutput: 0,\n\t\tcacheRead: 0,\n\t\tcacheWrite: 0,\n\t\ttotalTokens: 0,\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t};\n}\n\nfunction appendRewriteDiagnostic(message: AssistantMessage, rewrite: PiMessagesRewriteImpact | undefined): void {\n\tif (!rewrite) {\n\t\treturn;\n\t}\n\tappendAssistantMessageDiagnostic(message, {\n\t\ttype: \"pi_messages_rewrite\",\n\t\ttimestamp: Date.now(),\n\t\tdetails: { ...rewrite },\n\t});\n}\n\nfunction createEventConverter(model: Model<\"pi-messages\">) {\n\tconst partial: AssistantMessage = {\n\t\trole: \"assistant\",\n\t\tcontent: [],\n\t\tapi: model.api,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\tusage: createEmptyUsage(),\n\t\tstopReason: \"pending\",\n\t\ttimestamp: Date.now(),\n\t};\n\tconst toolJson = new Map<number, string>();\n\n\treturn (event: PiMessagesEvent): AssistantMessageEvent => {\n\t\tswitch (event.type) {\n\t\t\tcase \"done\":\n\t\t\t\tObject.assign(partial, {\n\t\t\t\t\tstopReason: event.reason,\n\t\t\t\t\tusage: event.usage,\n\t\t\t\t\tresponseId: event.responseId,\n\t\t\t\t});\n\t\t\t\tappendRewriteDiagnostic(partial, event.rewrite);\n\t\t\t\treturn { type: \"done\", reason: event.reason, message: partial };\n\t\t\tcase \"error\":\n\t\t\t\tObject.assign(partial, {\n\t\t\t\t\tstopReason: event.reason,\n\t\t\t\t\tusage: event.usage,\n\t\t\t\t\terrorMessage: event.errorMessage,\n\t\t\t\t\tresponseId: event.responseId,\n\t\t\t\t});\n\t\t\t\tappendRewriteDiagnostic(partial, event.rewrite);\n\t\t\t\treturn { type: \"error\", reason: event.reason, error: partial };\n\t\t\tcase \"start\":\n\t\t\t\tbreak;\n\t\t\tcase \"text_start\":\n\t\t\t\tpartial.content[event.contentIndex] = { type: \"text\", text: \"\" };\n\t\t\t\tbreak;\n\t\t\tcase \"text_delta\":\n\t\t\t\t(partial.content[event.contentIndex] as { text: string }).text += event.delta;\n\t\t\t\tbreak;\n\t\t\tcase \"text_end\":\n\t\t\t\tObject.assign(partial.content[event.contentIndex]!, {\n\t\t\t\t\ttext: event.content,\n\t\t\t\t\ttextSignature: event.contentSignature,\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\tcase \"thinking_start\":\n\t\t\t\tpartial.content[event.contentIndex] = { type: \"thinking\", thinking: \"\" };\n\t\t\t\tbreak;\n\t\t\tcase \"thinking_delta\":\n\t\t\t\t(partial.content[event.contentIndex] as { thinking: string }).thinking += event.delta;\n\t\t\t\tbreak;\n\t\t\tcase \"thinking_end\":\n\t\t\t\tObject.assign(partial.content[event.contentIndex]!, {\n\t\t\t\t\tthinking: event.content,\n\t\t\t\t\tthinkingSignature: event.contentSignature,\n\t\t\t\t\tredacted: event.redacted,\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\tcase \"toolcall_start\":\n\t\t\t\tpartial.content[event.contentIndex] = {\n\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\tid: event.id,\n\t\t\t\t\tname: event.toolName,\n\t\t\t\t\targuments: {},\n\t\t\t\t};\n\t\t\t\ttoolJson.set(event.contentIndex, \"\");\n\t\t\t\tbreak;\n\t\t\tcase \"toolcall_delta\": {\n\t\t\t\tconst json = `${toolJson.get(event.contentIndex) ?? \"\"}${event.delta}`;\n\t\t\t\ttoolJson.set(event.contentIndex, json);\n\t\t\t\t(partial.content[event.contentIndex] as ToolCall).arguments =\n\t\t\t\t\tparseStreamingJson<ToolCall[\"arguments\"]>(json);\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tcase \"toolcall_end\":\n\t\t\t\tObject.assign(partial.content[event.contentIndex]!, event.toolCall);\n\t\t\t\ttoolJson.delete(event.contentIndex);\n\t\t\t\treturn {\n\t\t\t\t\ttype: \"toolcall_end\",\n\t\t\t\t\tcontentIndex: event.contentIndex,\n\t\t\t\t\ttoolCall: partial.content[event.contentIndex] as ToolCall,\n\t\t\t\t\tpartial,\n\t\t\t\t};\n\t\t}\n\n\t\treturn { ...event, partial } as AssistantMessageEvent;\n\t};\n}\n\nasync function* readPiMessagesEvents(stream: ReadableStream<Uint8Array>): AsyncGenerator<PiMessagesEvent> {\n\tconst decoder = new TextDecoder();\n\tconst reader = stream.getReader();\n\tlet buffer = \"\";\n\n\ttry {\n\t\twhile (true) {\n\t\t\tconst { done, value } = await reader.read();\n\t\t\tbuffer += done ? decoder.decode() : decoder.decode(value, { stream: true });\n\t\t\tbuffer = buffer.replace(/\\r\\n/g, \"\\n\");\n\n\t\t\tlet split = buffer.indexOf(\"\\n\\n\");\n\t\t\twhile (split !== -1) {\n\t\t\t\tconst event = parsePiMessagesEvent(buffer.slice(0, split));\n\t\t\t\tif (event) {\n\t\t\t\t\tyield event;\n\t\t\t\t}\n\t\t\t\tbuffer = buffer.slice(split + 2);\n\t\t\t\tsplit = buffer.indexOf(\"\\n\\n\");\n\t\t\t}\n\n\t\t\tif (done) {\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\n\t\tif (buffer.trim()) {\n\t\t\tconst event = parsePiMessagesEvent(buffer);\n\t\t\tif (event) {\n\t\t\t\tyield event;\n\t\t\t}\n\t\t}\n\t} finally {\n\t\treader.releaseLock();\n\t}\n}\n\nfunction parsePiMessagesEvent(raw: string): PiMessagesEvent | undefined {\n\tconst data = raw\n\t\t.split(\"\\n\")\n\t\t.find((line) => line.startsWith(\"data:\"))\n\t\t?.slice(5)\n\t\t.trim();\n\n\treturn data && data !== \"[DONE]\" ? (JSON.parse(data) as PiMessagesEvent) : undefined;\n}\n\nfunction createErrorEvent(model: Model<\"pi-messages\">, error: unknown, aborted: boolean): AssistantMessageEvent {\n\tconst reason = aborted ? \"aborted\" : \"error\";\n\tconst assistantMessage: AssistantMessage = {\n\t\trole: \"assistant\",\n\t\tcontent: [],\n\t\tapi: model.api,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\tusage: createEmptyUsage(),\n\t\tstopReason: reason,\n\t\terrorMessage: error instanceof Error ? error.message : String(error),\n\t\ttimestamp: Date.now(),\n\t};\n\n\tif (!aborted && error instanceof PiMessagesResponseError) {\n\t\tappendAssistantMessageDiagnostic(\n\t\t\tassistantMessage,\n\t\t\tcreateAssistantMessageDiagnostic(\"pi_messages_response_failure\", error, error.diagnosticDetails),\n\t\t);\n\t}\n\n\treturn { type: \"error\", reason, error: assistantMessage };\n}\n\nfunction resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention | undefined {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\t// Backend defaults apply when unset; only the legacy env opt-in is mapped.\n\treturn getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\" ? \"long\" : undefined;\n}\n\nexport const stream: StreamFunction<\"pi-messages\", PiMessagesOptions> = (\n\tmodel: Model<\"pi-messages\">,\n\tcontext: Context,\n\toptions?: PiMessagesOptions,\n): AssistantMessageEventStream => {\n\tconst eventStream = new AssistantMessageEventStream();\n\tconst convertEvent = createEventConverter(model);\n\n\tvoid (async () => {\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey;\n\t\t\tif (!apiKey) {\n\t\t\t\tthrow new Error(`No API key provided for provider \"${model.provider}\"`);\n\t\t\t}\n\n\t\t\tconst url = new URL(`${model.baseUrl.replace(/\\/+$/u, \"\")}/messages`);\n\t\t\tif (options?.debug) {\n\t\t\t\turl.searchParams.set(\"debug\", \"1\");\n\t\t\t}\n\n\t\t\tlet payload: unknown = {\n\t\t\t\tmodel: model.id,\n\t\t\t\tcontext,\n\t\t\t\toptions: {\n\t\t\t\t\ttemperature: options?.temperature,\n\t\t\t\t\tmaxTokens: options?.maxTokens,\n\t\t\t\t\treasoning: options?.reasoning,\n\t\t\t\t\tcacheRetention: resolveCacheRetention(options?.cacheRetention, options?.env),\n\t\t\t\t\tsessionId: options?.sessionId,\n\t\t\t\t\ttoolChoice: options?.toolChoice,\n\t\t\t\t},\n\t\t\t};\n\t\t\tconst nextPayload = await options?.onPayload?.(payload, model);\n\t\t\tif (nextPayload !== undefined) {\n\t\t\t\tpayload = nextPayload;\n\t\t\t}\n\n\t\t\tconst response = await (options?.fetch ?? globalThis.fetch)(url, {\n\t\t\t\tmethod: \"POST\",\n\t\t\t\theaders: {\n\t\t\t\t\tauthorization: `Bearer ${apiKey}`,\n\t\t\t\t\taccept: \"text/event-stream\",\n\t\t\t\t\t\"content-type\": \"application/json\",\n\t\t\t\t\t...providerHeadersToRecord(options?.headers),\n\t\t\t\t},\n\t\t\t\tbody: JSON.stringify(payload),\n\t\t\t\tsignal: options?.signal,\n\t\t\t});\n\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\n\t\t\tif (!response.ok) {\n\t\t\t\tconst body = await response.text();\n\t\t\t\tthrow createPiMessagesResponseError(model, url, response, body);\n\t\t\t}\n\t\t\tif (!response.body) {\n\t\t\t\tthrow new Error(`${model.provider} response has no body`);\n\t\t\t}\n\n\t\t\tfor await (const piEvent of readPiMessagesEvents(response.body)) {\n\t\t\t\tconst event = convertEvent(piEvent);\n\t\t\t\teventStream.push(event);\n\t\t\t\tif (event.type === \"done\" || event.type === \"error\") {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tthrow new Error(`${model.provider} stream ended without a terminal event`);\n\t\t} catch (error) {\n\t\t\teventStream.push(createErrorEvent(model, error, options?.signal?.aborted ?? false));\n\t\t}\n\t})();\n\n\treturn eventStream;\n};\n\nexport const streamSimple: StreamFunction<\"pi-messages\", SimpleStreamOptions> = (\n\tmodel: Model<\"pi-messages\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst extra = options as PiMessagesOptions | undefined;\n\treturn stream(model, context, {\n\t\t...options,\n\t\treasoning: options?.reasoning,\n\t\ttoolChoice: extra?.toolChoice,\n\t\tdebug: extra?.debug,\n\t});\n};\n"]}
+{"version":3,"file":"pi-messages.d.ts","sourceRoot":"","sources":["../../src/api/pi-messages.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EACX,gBAAgB,EAMhB,mBAAmB,EACnB,cAAc,EACd,aAAa,EACb,aAAa,EACb,QAAQ,EACR,MAAM,aAAa,CAAC;AAQrB,MAAM,WAAW,iBAAkB,SAAQ,aAAa;IACvD,SAAS,CAAC,EAAE,aAAa,CAAC;IAC1B,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,UAAU,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC;IAC7F,0EAA0E;IAC1E,KAAK,CAAC,EAAE,OAAO,CAAC;CAChB;AAED,KAAK,eAAe,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;AACjD,KAAK,oBAAoB,GAAG,gBAAgB,CAAC,YAAY,CAAC,CAAC;AAE3D,+EAA+E;AAC/E,MAAM,MAAM,uBAAuB,GAAG;IACrC,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;IACtB,OAAO,EAAE,OAAO,CAAC;IACjB,gBAAgB,EAAE,MAAM,CAAC;IACzB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,mBAAmB,EAAE,OAAO,CAAC;CAC7B,CAAC;AAEF,2EAA2E;AAC3E,MAAM,MAAM,eAAe,GACxB;IAAE,IAAI,EAAE,OAAO,CAAA;CAAE,GACjB;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,GAC5C;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAC3D;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,gBAAgB,CAAC,EAAE,MAAM,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,GAChD;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAC/D;IACA,IAAI,EAAE,cAAc,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,EAAE,OAAO,CAAC;CAClB,GACD;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAA;CAAE,GAC9E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAC/D;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,QAAQ,CAAA;CAAE,GAClE;IACA,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,OAAO,CAAC,oBAAoB,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS,CAAC,CAAC;IACrE,KAAK,EAAE,eAAe,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,OAAO,CAAC,EAAE,uBAAuB,CAAC;CACjC,GACD;IACA,IAAI,EAAE,OAAO,CAAC;IACd,MAAM,EAAE,OAAO,CAAC,oBAAoB,EAAE,SAAS,GAAG,OAAO,CAAC,CAAC;IAC3D,KAAK,EAAE,eAAe,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,OAAO,CAAC,EAAE,uBAAuB,CAAC;CACjC,CAAC;AAWL,qBAAa,uBAAwB,SAAQ,KAAK;IACjD,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAEpD,YAAY,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,SAAS,EAAE,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAKhG;CACD;AA8PD,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,aAAa,EAAE,iBAAiB,CAiFnE,CAAC;AAEF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,aAAa,EAAE,mBAAmB,CAY3E,CAAC","sourcesContent":["/**\n * pi-messages API implementation.\n *\n * Streams pi's own message protocol directly to a backend: the request is a\n * single POST of `{ model, context, options }` to `<baseUrl>/messages`, the\n * response is an SSE stream of serialized assistant-message events plus a\n * terminal `done`/`error` event. This is the wire protocol spoken by the\n * Radius gateway, but any backend implementing it can be used, e.g. via a\n * models.json custom provider with `\"api\": \"pi-messages\"`.\n */\n\nimport type {\n\tAssistantMessage,\n\tAssistantMessageEvent,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tProviderEnv,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tThinkingLevel,\n\tToolCall,\n} from \"../types.ts\";\nimport { appendAssistantMessageDiagnostic, createAssistantMessageDiagnostic } from \"../utils/diagnostics.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { headersToRecord, providerHeadersToRecord } from \"../utils/headers.ts\";\nimport { parseStreamingJson } from \"../utils/json-parse.ts\";\nimport { createStreamDeadline, withStreamDeadline } from \"../utils/stream-deadline.ts\";\nimport { getProviderEnvValue } from \"../utils/provider-env.ts\";\n\nexport interface PiMessagesOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\ttoolChoice?: \"auto\" | \"none\" | \"required\" | { type: \"function\"; function: { name: string } };\n\t/** Ask the backend for debug metadata (e.g. routing response headers). */\n\tdebug?: boolean;\n}\n\ntype PiMessagesUsage = AssistantMessage[\"usage\"];\ntype PiMessagesStopReason = AssistantMessage[\"stopReason\"];\n\n/** Impact summary of a server-side message rewrite (e.g. a gateway policy). */\nexport type PiMessagesRewriteImpact = {\n\tpolicyId: string;\n\tpolicyVersion: number;\n\tchanged: boolean;\n\ttokenCountChange: number;\n\tmessageCountChange: number;\n\tsystemPromptChanged: boolean;\n};\n\n/** Serialized assistant-message event as sent by a pi-messages backend. */\nexport type PiMessagesEvent =\n\t| { type: \"start\" }\n\t| { type: \"text_start\"; contentIndex: number }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string }\n\t| { type: \"text_end\"; contentIndex: number; content: string; contentSignature?: string }\n\t| { type: \"thinking_start\"; contentIndex: number }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string }\n\t| {\n\t\t\ttype: \"thinking_end\";\n\t\t\tcontentIndex: number;\n\t\t\tcontent: string;\n\t\t\tcontentSignature?: string;\n\t\t\tredacted?: boolean;\n\t  }\n\t| { type: \"toolcall_start\"; contentIndex: number; id: string; toolName: string }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall }\n\t| {\n\t\t\ttype: \"done\";\n\t\t\treason: Extract<PiMessagesStopReason, \"stop\" | \"length\" | \"toolUse\">;\n\t\t\tusage: PiMessagesUsage;\n\t\t\tresponseId?: string;\n\t\t\trewrite?: PiMessagesRewriteImpact;\n\t  }\n\t| {\n\t\t\ttype: \"error\";\n\t\t\treason: Extract<PiMessagesStopReason, \"aborted\" | \"error\">;\n\t\t\tusage: PiMessagesUsage;\n\t\t\terrorMessage?: string;\n\t\t\tresponseId?: string;\n\t\t\trewrite?: PiMessagesRewriteImpact;\n\t  };\n\ntype PiMessagesErrorBody = {\n\terror?: {\n\t\tmessage?: unknown;\n\t\tcode?: unknown;\n\t\tdetails?: unknown;\n\t\t[key: string]: unknown;\n\t};\n};\n\nexport class PiMessagesResponseError extends Error {\n\tcode?: string;\n\treadonly diagnosticDetails: Record<string, unknown>;\n\n\tconstructor(message: string, code: string | undefined, diagnosticDetails: Record<string, unknown>) {\n\t\tsuper(message);\n\t\tthis.name = \"PiMessagesResponseError\";\n\t\tthis.code = code;\n\t\tthis.diagnosticDetails = diagnosticDetails;\n\t}\n}\n\nfunction parsePiMessagesErrorBody(body: string): PiMessagesErrorBody | undefined {\n\ttry {\n\t\tconst parsed = JSON.parse(body) as PiMessagesErrorBody | null;\n\t\tconst error = parsed?.error;\n\t\treturn parsed && typeof error === \"object\" && error !== null && !Array.isArray(error) ? parsed : undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nfunction truncateDiagnosticString(value: string): string {\n\tconst maxLength = 8192;\n\treturn value.length > maxLength ? `${value.slice(0, maxLength)}…` : value;\n}\n\nfunction formatPiMessagesResponseError(\n\tresponse: Response,\n\tbody: string,\n\terrorBody: PiMessagesErrorBody | undefined,\n): string {\n\tconst message = typeof errorBody?.error?.message === \"string\" ? errorBody.error.message : undefined;\n\tconst code = typeof errorBody?.error?.code === \"string\" ? errorBody.error.code : undefined;\n\tconst suffix = message ?? body;\n\tconst codeSuffix = code ? ` (${code})` : \"\";\n\treturn `${response.status} ${response.statusText}: ${suffix}${codeSuffix}`;\n}\n\nfunction createPiMessagesResponseError(\n\tmodel: Model<\"pi-messages\">,\n\turl: URL,\n\tresponse: Response,\n\tbody: string,\n): PiMessagesResponseError {\n\tconst errorBody = parsePiMessagesErrorBody(body);\n\tconst code = typeof errorBody?.error?.code === \"string\" ? errorBody.error.code : undefined;\n\treturn new PiMessagesResponseError(formatPiMessagesResponseError(response, body, errorBody), code, {\n\t\tversion: 1,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\turl: url.toString(),\n\t\tstatus: response.status,\n\t\tstatusText: response.statusText,\n\t\terror: errorBody?.error,\n\t\tbody: errorBody ? undefined : truncateDiagnosticString(body),\n\t\ttimestampMs: Date.now(),\n\t});\n}\n\nfunction createEmptyUsage(): PiMessagesUsage {\n\treturn {\n\t\tinput: 0,\n\t\toutput: 0,\n\t\tcacheRead: 0,\n\t\tcacheWrite: 0,\n\t\ttotalTokens: 0,\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t};\n}\n\nfunction appendRewriteDiagnostic(message: AssistantMessage, rewrite: PiMessagesRewriteImpact | undefined): void {\n\tif (!rewrite) {\n\t\treturn;\n\t}\n\tappendAssistantMessageDiagnostic(message, {\n\t\ttype: \"pi_messages_rewrite\",\n\t\ttimestamp: Date.now(),\n\t\tdetails: { ...rewrite },\n\t});\n}\n\nfunction createEventConverter(model: Model<\"pi-messages\">) {\n\tconst partial: AssistantMessage = {\n\t\trole: \"assistant\",\n\t\tcontent: [],\n\t\tapi: model.api,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\tusage: createEmptyUsage(),\n\t\tstopReason: \"pending\",\n\t\ttimestamp: Date.now(),\n\t};\n\tconst toolJson = new Map<number, string>();\n\n\treturn (event: PiMessagesEvent): AssistantMessageEvent => {\n\t\tswitch (event.type) {\n\t\t\tcase \"done\":\n\t\t\t\tObject.assign(partial, {\n\t\t\t\t\tstopReason: event.reason,\n\t\t\t\t\tusage: event.usage,\n\t\t\t\t\tresponseId: event.responseId,\n\t\t\t\t});\n\t\t\t\tappendRewriteDiagnostic(partial, event.rewrite);\n\t\t\t\treturn { type: \"done\", reason: event.reason, message: partial };\n\t\t\tcase \"error\":\n\t\t\t\tObject.assign(partial, {\n\t\t\t\t\tstopReason: event.reason,\n\t\t\t\t\tusage: event.usage,\n\t\t\t\t\terrorMessage: event.errorMessage,\n\t\t\t\t\tresponseId: event.responseId,\n\t\t\t\t});\n\t\t\t\tappendRewriteDiagnostic(partial, event.rewrite);\n\t\t\t\treturn { type: \"error\", reason: event.reason, error: partial };\n\t\t\tcase \"start\":\n\t\t\t\tbreak;\n\t\t\tcase \"text_start\":\n\t\t\t\tpartial.content[event.contentIndex] = { type: \"text\", text: \"\" };\n\t\t\t\tbreak;\n\t\t\tcase \"text_delta\":\n\t\t\t\t(partial.content[event.contentIndex] as { text: string }).text += event.delta;\n\t\t\t\tbreak;\n\t\t\tcase \"text_end\":\n\t\t\t\tObject.assign(partial.content[event.contentIndex]!, {\n\t\t\t\t\ttext: event.content,\n\t\t\t\t\ttextSignature: event.contentSignature,\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\tcase \"thinking_start\":\n\t\t\t\tpartial.content[event.contentIndex] = { type: \"thinking\", thinking: \"\" };\n\t\t\t\tbreak;\n\t\t\tcase \"thinking_delta\":\n\t\t\t\t(partial.content[event.contentIndex] as { thinking: string }).thinking += event.delta;\n\t\t\t\tbreak;\n\t\t\tcase \"thinking_end\":\n\t\t\t\tObject.assign(partial.content[event.contentIndex]!, {\n\t\t\t\t\tthinking: event.content,\n\t\t\t\t\tthinkingSignature: event.contentSignature,\n\t\t\t\t\tredacted: event.redacted,\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\tcase \"toolcall_start\":\n\t\t\t\tpartial.content[event.contentIndex] = {\n\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\tid: event.id,\n\t\t\t\t\tname: event.toolName,\n\t\t\t\t\targuments: {},\n\t\t\t\t};\n\t\t\t\ttoolJson.set(event.contentIndex, \"\");\n\t\t\t\tbreak;\n\t\t\tcase \"toolcall_delta\": {\n\t\t\t\tconst json = `${toolJson.get(event.contentIndex) ?? \"\"}${event.delta}`;\n\t\t\t\ttoolJson.set(event.contentIndex, json);\n\t\t\t\t(partial.content[event.contentIndex] as ToolCall).arguments =\n\t\t\t\t\tparseStreamingJson<ToolCall[\"arguments\"]>(json);\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tcase \"toolcall_end\":\n\t\t\t\tObject.assign(partial.content[event.contentIndex]!, event.toolCall);\n\t\t\t\ttoolJson.delete(event.contentIndex);\n\t\t\t\treturn {\n\t\t\t\t\ttype: \"toolcall_end\",\n\t\t\t\t\tcontentIndex: event.contentIndex,\n\t\t\t\t\ttoolCall: partial.content[event.contentIndex] as ToolCall,\n\t\t\t\t\tpartial,\n\t\t\t\t};\n\t\t}\n\n\t\treturn { ...event, partial } as AssistantMessageEvent;\n\t};\n}\n\nasync function* readPiMessagesEvents(\n\tstream: ReadableStream<Uint8Array>,\n\tsignal?: AbortSignal,\n): AsyncGenerator<PiMessagesEvent> {\n\tconst decoder = new TextDecoder();\n\tconst reader = stream.getReader();\n\tlet buffer = \"\";\n\tconst onAbort = () => {\n\t\tvoid reader.cancel().catch(() => {});\n\t};\n\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\n\ttry {\n\t\twhile (true) {\n\t\t\tif (signal?.aborted) throw signal.reason ?? new Error(\"Request was aborted\");\n\t\t\tconst { done, value } = await reader.read();\n\t\t\tif (signal?.aborted) throw signal.reason ?? new Error(\"Request was aborted\");\n\t\t\tbuffer += done ? decoder.decode() : decoder.decode(value, { stream: true });\n\t\t\tbuffer = buffer.replace(/\\r\\n/g, \"\\n\");\n\n\t\t\tlet split = buffer.indexOf(\"\\n\\n\");\n\t\t\twhile (split !== -1) {\n\t\t\t\tconst event = parsePiMessagesEvent(buffer.slice(0, split));\n\t\t\t\tif (event) {\n\t\t\t\t\tyield event;\n\t\t\t\t}\n\t\t\t\tbuffer = buffer.slice(split + 2);\n\t\t\t\tsplit = buffer.indexOf(\"\\n\\n\");\n\t\t\t}\n\n\t\t\tif (done) {\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\n\t\tif (buffer.trim()) {\n\t\t\tconst event = parsePiMessagesEvent(buffer);\n\t\t\tif (event) {\n\t\t\t\tyield event;\n\t\t\t}\n\t\t}\n\t} finally {\n\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\ttry {\n\t\t\tawait reader.cancel();\n\t\t} catch {}\n\t\treader.releaseLock();\n\t}\n}\n\nfunction parsePiMessagesEvent(raw: string): PiMessagesEvent | undefined {\n\tconst data = raw\n\t\t.split(\"\\n\")\n\t\t.find((line) => line.startsWith(\"data:\"))\n\t\t?.slice(5)\n\t\t.trim();\n\n\treturn data && data !== \"[DONE]\" ? (JSON.parse(data) as PiMessagesEvent) : undefined;\n}\n\nfunction createErrorEvent(model: Model<\"pi-messages\">, error: unknown, aborted: boolean): AssistantMessageEvent {\n\tconst reason = aborted ? \"aborted\" : \"error\";\n\tconst assistantMessage: AssistantMessage = {\n\t\trole: \"assistant\",\n\t\tcontent: [],\n\t\tapi: model.api,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\tusage: createEmptyUsage(),\n\t\tstopReason: reason,\n\t\terrorMessage: error instanceof Error ? error.message : String(error),\n\t\ttimestamp: Date.now(),\n\t};\n\n\tif (!aborted && error instanceof PiMessagesResponseError) {\n\t\tappendAssistantMessageDiagnostic(\n\t\t\tassistantMessage,\n\t\t\tcreateAssistantMessageDiagnostic(\"pi_messages_response_failure\", error, error.diagnosticDetails),\n\t\t);\n\t}\n\n\treturn { type: \"error\", reason, error: assistantMessage };\n}\n\nfunction resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention | undefined {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\t// Backend defaults apply when unset; only the legacy env opt-in is mapped.\n\treturn getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\" ? \"long\" : undefined;\n}\n\nexport const stream: StreamFunction<\"pi-messages\", PiMessagesOptions> = (\n\tmodel: Model<\"pi-messages\">,\n\tcontext: Context,\n\toptions?: PiMessagesOptions,\n): AssistantMessageEventStream => {\n\tconst eventStream = new AssistantMessageEventStream();\n\tconst convertEvent = createEventConverter(model);\n\n\tvoid (async () => {\n\t\tconst streamDeadline = createStreamDeadline(options?.streamDeadlineMs, options?.signal);\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey;\n\t\t\tif (!apiKey) {\n\t\t\t\tthrow new Error(`No API key provided for provider \"${model.provider}\"`);\n\t\t\t}\n\n\t\t\tconst url = new URL(`${model.baseUrl.replace(/\\/+$/u, \"\")}/messages`);\n\t\t\tif (options?.debug) {\n\t\t\t\turl.searchParams.set(\"debug\", \"1\");\n\t\t\t}\n\n\t\t\tlet payload: unknown = {\n\t\t\t\tmodel: model.id,\n\t\t\t\tcontext,\n\t\t\t\toptions: {\n\t\t\t\t\ttemperature: options?.temperature,\n\t\t\t\t\tmaxTokens: options?.maxTokens,\n\t\t\t\t\treasoning: options?.reasoning,\n\t\t\t\t\tcacheRetention: resolveCacheRetention(options?.cacheRetention, options?.env),\n\t\t\t\t\tsessionId: options?.sessionId,\n\t\t\t\t\ttoolChoice: options?.toolChoice,\n\t\t\t\t},\n\t\t\t};\n\t\t\tconst nextPayload = await options?.onPayload?.(payload, model);\n\t\t\tif (nextPayload !== undefined) {\n\t\t\t\tpayload = nextPayload;\n\t\t\t}\n\n\t\t\tconst response = await (options?.fetch ?? globalThis.fetch)(url, {\n\t\t\t\tmethod: \"POST\",\n\t\t\t\theaders: {\n\t\t\t\t\tauthorization: `Bearer ${apiKey}`,\n\t\t\t\t\taccept: \"text/event-stream\",\n\t\t\t\t\t\"content-type\": \"application/json\",\n\t\t\t\t\t...providerHeadersToRecord({ ...model.headers, ...options?.headers }),\n\t\t\t\t},\n\t\t\t\tbody: JSON.stringify(payload),\n\t\t\t\tsignal: streamDeadline.signal,\n\t\t\t});\n\n\t\t\tawait options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);\n\n\t\t\tif (!response.ok) {\n\t\t\t\tconst body = await response.text();\n\t\t\t\tthrow createPiMessagesResponseError(model, url, response, body);\n\t\t\t}\n\t\t\tif (!response.body) {\n\t\t\t\tthrow new Error(`${model.provider} response has no body`);\n\t\t\t}\n\n\t\t\tfor await (const piEvent of withStreamDeadline(\n\t\t\t\treadPiMessagesEvents(response.body, streamDeadline.signal),\n\t\t\t\tstreamDeadline.deadlineMs,\n\t\t\t\tstreamDeadline.abort,\n\t\t\t)) {\n\t\t\t\tconst event = convertEvent(piEvent);\n\t\t\t\teventStream.push(event);\n\t\t\t\tif (event.type === \"done\" || event.type === \"error\") {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tthrow new Error(`${model.provider} stream ended without a terminal event`);\n\t\t} catch (error) {\n\t\t\teventStream.push(createErrorEvent(model, error, options?.signal?.aborted ?? false));\n\t\t} finally {\n\t\t\tstreamDeadline.cleanup();\n\t\t}\n\t})();\n\n\treturn eventStream;\n};\n\nexport const streamSimple: StreamFunction<\"pi-messages\", SimpleStreamOptions> = (\n\tmodel: Model<\"pi-messages\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst extra = options as PiMessagesOptions | undefined;\n\treturn stream(model, context, {\n\t\t...options,\n\t\treasoning: options?.reasoning,\n\t\ttoolChoice: options?.toolChoice,\n\t\tdebug: extra?.debug,\n\t});\n};\n"]}

package/dist/api/pi-messages.js

@@ -12,6 +12,7 @@ import { appendAssistantMessageDiagnostic, createAssistantMessageDiagnostic } fr
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { headersToRecord, providerHeadersToRecord } from "../utils/headers.js";
 import { parseStreamingJson } from "../utils/json-parse.js";
+import { createStreamDeadline, withStreamDeadline } from "../utils/stream-deadline.js";
 import { getProviderEnvValue } from "../utils/provider-env.js";
 export class PiMessagesResponseError extends Error {
     code;
@@ -166,13 +167,21 @@ function createEventConverter(model) {
         return { ...event, partial };
     };
 }
-async function* readPiMessagesEvents(stream) {
+async function* readPiMessagesEvents(stream, signal) {
     const decoder = new TextDecoder();
     const reader = stream.getReader();
     let buffer = "";
+    const onAbort = () => {
+        void reader.cancel().catch(() => { });
+    };
+    signal?.addEventListener("abort", onAbort, { once: true });
     try {
         while (true) {
+            if (signal?.aborted)
+                throw signal.reason ?? new Error("Request was aborted");
             const { done, value } = await reader.read();
+            if (signal?.aborted)
+                throw signal.reason ?? new Error("Request was aborted");
             buffer += done ? decoder.decode() : decoder.decode(value, { stream: true });
             buffer = buffer.replace(/\r\n/g, "\n");
             let split = buffer.indexOf("\n\n");
@@ -196,6 +205,11 @@ async function* readPiMessagesEvents(stream) {
         }
     }
     finally {
+        signal?.removeEventListener("abort", onAbort);
+        try {
+            await reader.cancel();
+        }
+        catch { }
         reader.releaseLock();
     }
 }
@@ -236,6 +250,7 @@ export const stream = (model, context, options) => {
     const eventStream = new AssistantMessageEventStream();
     const convertEvent = createEventConverter(model);
     void (async () => {
+        const streamDeadline = createStreamDeadline(options?.streamDeadlineMs, options?.signal);
         try {
             const apiKey = options?.apiKey;
             if (!apiKey) {
@@ -267,10 +282,10 @@ export const stream = (model, context, options) => {
                     authorization: `Bearer ${apiKey}`,
                     accept: "text/event-stream",
                     "content-type": "application/json",
-                    ...providerHeadersToRecord(options?.headers),
+                    ...providerHeadersToRecord({ ...model.headers, ...options?.headers }),
                 },
                 body: JSON.stringify(payload),
-                signal: options?.signal,
+                signal: streamDeadline.signal,
             });
             await options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);
             if (!response.ok) {
@@ -280,7 +295,7 @@ export const stream = (model, context, options) => {
             if (!response.body) {
                 throw new Error(`${model.provider} response has no body`);
             }
-            for await (const piEvent of readPiMessagesEvents(response.body)) {
+            for await (const piEvent of withStreamDeadline(readPiMessagesEvents(response.body, streamDeadline.signal), streamDeadline.deadlineMs, streamDeadline.abort)) {
                 const event = convertEvent(piEvent);
                 eventStream.push(event);
                 if (event.type === "done" || event.type === "error") {
@@ -292,6 +307,9 @@ export const stream = (model, context, options) => {
         catch (error) {
             eventStream.push(createErrorEvent(model, error, options?.signal?.aborted ?? false));
         }
+        finally {
+            streamDeadline.cleanup();
+        }
     })();
     return eventStream;
 };
@@ -300,7 +318,7 @@ export const streamSimple = (model, context, options) => {
     return stream(model, context, {
         ...options,
         reasoning: options?.reasoning,
-        toolChoice: extra?.toolChoice,
+        toolChoice: options?.toolChoice,
         debug: extra?.debug,
     });
 };