npm - @mariozechner/pi-ai - Versions diffs - 0.70.1 → 0.70.2 - Mend

@mariozechner/pi-ai 0.70.1 → 0.70.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (16) hide show

package/dist/models.generated.d.ts.map +1 -1
package/dist/models.generated.js +1 -1
package/dist/models.generated.js.map +1 -1
package/dist/providers/anthropic.d.ts.map +1 -1
package/dist/providers/anthropic.js +6 -7
package/dist/providers/anthropic.js.map +1 -1
package/dist/providers/azure-openai-responses.d.ts.map +1 -1
package/dist/providers/azure-openai-responses.js +6 -7
package/dist/providers/azure-openai-responses.js.map +1 -1
package/dist/providers/openai-completions.d.ts.map +1 -1
package/dist/providers/openai-completions.js +6 -5
package/dist/providers/openai-completions.js.map +1 -1
package/dist/providers/openai-responses.d.ts.map +1 -1
package/dist/providers/openai-responses.js +6 -7
package/dist/providers/openai-responses.js.map +1 -1
package/package.json +1 -1

package/dist/providers/azure-openai-responses.d.ts.map CHANGED Viewed

	@@ -1 +1 @@
1	- {"version":3,"file":"azure-openai-responses.d.ts","sourceRoot":"","sources":["../../src/providers/azure-openai-responses.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAKX,mBAAmB,EACnB,cAAc,EACd,aAAa,EACb,MAAM,aAAa,CAAC;AA+BrB,MAAM,WAAW,2BAA4B,SAAQ,aAAa;IACjE,eAAe,CAAC,EAAE,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;IAClE,gBAAgB,CAAC,EAAE,MAAM,GAAG,UAAU,GAAG,SAAS,GAAG,IAAI,CAAC;IAC1D,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED;;GAEG;AACH,eAAO,MAAM,0BAA0B,EAAE,cAAc,CAAC,wBAAwB,EAAE,2BAA2B,~~CA0E5G~~,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,cAAc,CAAC,wBAAwB,EAAE,mBAAmB,CAiB1G,CAAC","sourcesContent":["import { AzureOpenAI } from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { supportsXhigh } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { headersToRecord } from \"../utils/headers.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\n\nconst DEFAULT_AZURE_API_VERSION = \"v1\";\nconst AZURE_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\", \"azure-openai-responses\"]);\n\nfunction parseDeploymentNameMap(value: string \| undefined): Map<string, string> {\n\tconst map = new Map<string, string>();\n\tif (!value) return map;\n\tfor (const entry of value.split(\",\")) {\n\t\tconst trimmed = entry.trim();\n\t\tif (!trimmed) continue;\n\t\tconst [modelId, deploymentName] = trimmed.split(\"=\", 2);\n\t\tif (!modelId \|\| !deploymentName) continue;\n\t\tmap.set(modelId.trim(), deploymentName.trim());\n\t}\n\treturn map;\n}\n\nfunction resolveDeploymentName(model: Model<\"azure-openai-responses\">, options?: AzureOpenAIResponsesOptions): string {\n\tif (options?.azureDeploymentName) {\n\t\treturn options.azureDeploymentName;\n\t}\n\tconst mappedDeployment = parseDeploymentNameMap(process.env.AZURE_OPENAI_DEPLOYMENT_NAME_MAP).get(model.id);\n\treturn mappedDeployment \|\| model.id;\n}\n\n// Azure OpenAI Responses-specific options\nexport interface AzureOpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" \| \"low\" \| \"medium\" \| \"high\" \| \"xhigh\";\n\treasoningSummary?: \"auto\" \| \"detailed\" \| \"concise\" \| null;\n\tazureApiVersion?: string;\n\tazureResourceName?: string;\n\tazureBaseUrl?: string;\n\tazureDeploymentName?: string;\n}\n\n/*\n Generate function for Azure OpenAI Responses API\n */\nexport const streamAzureOpenAIResponses: StreamFunction<\"azure-openai-responses\", AzureOpenAIResponsesOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions?: AzureOpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t// Start async processing\n\t(async () => {\n\t\tconst deploymentName = resolveDeploymentName(model, options);\n\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"azure-openai-responses\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create Azure OpenAI client\n\t\t\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider) \|\| \"\";\n\t\t\tconst client = createClient(model, apiKey, options);\n\t\t\tlet params = buildParams(model, context, options, deploymentName);\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 { ~~data: openaiStream, response }~~ = ~~await client.responses~~\n\t\t\t\t~~.create~~(~~params,~~ {~~\n\t\t\t\t\tsignal~~: options?.signal,\n\t\t\t\t~~\ttimeout~~: options?.timeoutMs,\n\t\t\t\t~~\tmaxRetries~~: options?.maxRetries,\n\t\t\t\t})\n\t\t\~~t\t~~.withResponse();\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\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" \|\| output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// partialJson is only a streaming scratch buffer; never persist it.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleAzureOpenAIResponses: StreamFunction<\"azure-openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\n\treturn streamAzureOpenAIResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies AzureOpenAIResponsesOptions);\n};\n\nfunction normalizeAzureBaseUrl(baseUrl: string): string {\n\treturn baseUrl.replace(/\\/+$/, \"\");\n}\n\nfunction buildDefaultBaseUrl(resourceName: string): string {\n\treturn `https://${resourceName}.openai.azure.com/openai/v1`;\n}\n\nfunction resolveAzureConfig(\n\tmodel: Model<\"azure-openai-responses\">,\n\toptions?: AzureOpenAIResponsesOptions,\n): { baseUrl: string; apiVersion: string } {\n\tconst apiVersion = options?.azureApiVersion \|\| process.env.AZURE_OPENAI_API_VERSION \|\| DEFAULT_AZURE_API_VERSION;\n\n\tconst baseUrl = options?.azureBaseUrl?.trim() \|\| process.env.AZURE_OPENAI_BASE_URL?.trim() \|\| undefined;\n\tconst resourceName = options?.azureResourceName \|\| process.env.AZURE_OPENAI_RESOURCE_NAME;\n\n\tlet resolvedBaseUrl = baseUrl;\n\n\tif (!resolvedBaseUrl && resourceName) {\n\t\tresolvedBaseUrl = buildDefaultBaseUrl(resourceName);\n\t}\n\n\tif (!resolvedBaseUrl && model.baseUrl) {\n\t\tresolvedBaseUrl = model.baseUrl;\n\t}\n\n\tif (!resolvedBaseUrl) {\n\t\tthrow new Error(\n\t\t\t\"Azure OpenAI base URL is required. Set AZURE_OPENAI_BASE_URL or AZURE_OPENAI_RESOURCE_NAME, or pass azureBaseUrl, azureResourceName, or model.baseUrl.\",\n\t\t);\n\t}\n\n\treturn {\n\t\tbaseUrl: normalizeAzureBaseUrl(resolvedBaseUrl),\n\t\tapiVersion,\n\t};\n}\n\nfunction createClient(model: Model<\"azure-openai-responses\">, apiKey: string, options?: AzureOpenAIResponsesOptions) {\n\tif (!apiKey) {\n\t\tif (!process.env.AZURE_OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"Azure OpenAI API key is required. Set AZURE_OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.AZURE_OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\n\tif (options?.headers) {\n\t\tObject.assign(headers, options.headers);\n\t}\n\n\tconst { baseUrl, apiVersion } = resolveAzureConfig(model, options);\n\n\treturn new AzureOpenAI({\n\t\tapiKey,\n\t\tapiVersion,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t\tbaseURL: baseUrl,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions: AzureOpenAIResponsesOptions \| undefined,\n\tdeploymentName: string,\n) {\n\tconst messages = convertResponsesMessages(model, context, AZURE_TOOL_CALL_PROVIDERS);\n\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: deploymentName,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: options?.sessionId,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = options?.maxTokens;\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertResponsesTools(context.tools);\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort \|\| options?.reasoningSummary) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: options?.reasoningEffort \|\| \"medium\",\n\t\t\t\tsummary: options?.reasoningSummary \|\| \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else {\n\t\t\tparams.reasoning = { effort: \"none\" };\n\t\t}\n\t}\n\n\treturn params;\n}\n"]}
1	+ {"version":3,"file":"azure-openai-responses.d.ts","sourceRoot":"","sources":["../../src/providers/azure-openai-responses.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAKX,mBAAmB,EACnB,cAAc,EACd,aAAa,EACb,MAAM,aAAa,CAAC;AA+BrB,MAAM,WAAW,2BAA4B,SAAQ,aAAa;IACjE,eAAe,CAAC,EAAE,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;IAClE,gBAAgB,CAAC,EAAE,MAAM,GAAG,UAAU,GAAG,SAAS,GAAG,IAAI,CAAC;IAC1D,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED;;GAEG;AACH,eAAO,MAAM,0BAA0B,EAAE,cAAc,CAAC,wBAAwB,EAAE,2BAA2B,CAyE5G,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,cAAc,CAAC,wBAAwB,EAAE,mBAAmB,CAiB1G,CAAC","sourcesContent":["import { AzureOpenAI } from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { supportsXhigh } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { headersToRecord } from \"../utils/headers.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\n\nconst DEFAULT_AZURE_API_VERSION = \"v1\";\nconst AZURE_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\", \"azure-openai-responses\"]);\n\nfunction parseDeploymentNameMap(value: string \| undefined): Map<string, string> {\n\tconst map = new Map<string, string>();\n\tif (!value) return map;\n\tfor (const entry of value.split(\",\")) {\n\t\tconst trimmed = entry.trim();\n\t\tif (!trimmed) continue;\n\t\tconst [modelId, deploymentName] = trimmed.split(\"=\", 2);\n\t\tif (!modelId \|\| !deploymentName) continue;\n\t\tmap.set(modelId.trim(), deploymentName.trim());\n\t}\n\treturn map;\n}\n\nfunction resolveDeploymentName(model: Model<\"azure-openai-responses\">, options?: AzureOpenAIResponsesOptions): string {\n\tif (options?.azureDeploymentName) {\n\t\treturn options.azureDeploymentName;\n\t}\n\tconst mappedDeployment = parseDeploymentNameMap(process.env.AZURE_OPENAI_DEPLOYMENT_NAME_MAP).get(model.id);\n\treturn mappedDeployment \|\| model.id;\n}\n\n// Azure OpenAI Responses-specific options\nexport interface AzureOpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" \| \"low\" \| \"medium\" \| \"high\" \| \"xhigh\";\n\treasoningSummary?: \"auto\" \| \"detailed\" \| \"concise\" \| null;\n\tazureApiVersion?: string;\n\tazureResourceName?: string;\n\tazureBaseUrl?: string;\n\tazureDeploymentName?: string;\n}\n\n/*\n Generate function for Azure OpenAI Responses API\n */\nexport const streamAzureOpenAIResponses: StreamFunction<\"azure-openai-responses\", AzureOpenAIResponsesOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions?: AzureOpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t// Start async processing\n\t(async () => {\n\t\tconst deploymentName = resolveDeploymentName(model, options);\n\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"azure-openai-responses\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create Azure OpenAI client\n\t\t\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider) \|\| \"\";\n\t\t\tconst client = createClient(model, apiKey, options);\n\t\t\tlet params = buildParams(model, context, options, deploymentName);\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\t...(options?.maxRetries !== undefined ? { maxRetries: options.maxRetries } : {}),\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await client.responses.create(params, requestOptions).withResponse();\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\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" \|\| output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// partialJson is only a streaming scratch buffer; never persist it.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleAzureOpenAIResponses: StreamFunction<\"azure-openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\n\treturn streamAzureOpenAIResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies AzureOpenAIResponsesOptions);\n};\n\nfunction normalizeAzureBaseUrl(baseUrl: string): string {\n\treturn baseUrl.replace(/\\/+$/, \"\");\n}\n\nfunction buildDefaultBaseUrl(resourceName: string): string {\n\treturn `https://${resourceName}.openai.azure.com/openai/v1`;\n}\n\nfunction resolveAzureConfig(\n\tmodel: Model<\"azure-openai-responses\">,\n\toptions?: AzureOpenAIResponsesOptions,\n): { baseUrl: string; apiVersion: string } {\n\tconst apiVersion = options?.azureApiVersion \|\| process.env.AZURE_OPENAI_API_VERSION \|\| DEFAULT_AZURE_API_VERSION;\n\n\tconst baseUrl = options?.azureBaseUrl?.trim() \|\| process.env.AZURE_OPENAI_BASE_URL?.trim() \|\| undefined;\n\tconst resourceName = options?.azureResourceName \|\| process.env.AZURE_OPENAI_RESOURCE_NAME;\n\n\tlet resolvedBaseUrl = baseUrl;\n\n\tif (!resolvedBaseUrl && resourceName) {\n\t\tresolvedBaseUrl = buildDefaultBaseUrl(resourceName);\n\t}\n\n\tif (!resolvedBaseUrl && model.baseUrl) {\n\t\tresolvedBaseUrl = model.baseUrl;\n\t}\n\n\tif (!resolvedBaseUrl) {\n\t\tthrow new Error(\n\t\t\t\"Azure OpenAI base URL is required. Set AZURE_OPENAI_BASE_URL or AZURE_OPENAI_RESOURCE_NAME, or pass azureBaseUrl, azureResourceName, or model.baseUrl.\",\n\t\t);\n\t}\n\n\treturn {\n\t\tbaseUrl: normalizeAzureBaseUrl(resolvedBaseUrl),\n\t\tapiVersion,\n\t};\n}\n\nfunction createClient(model: Model<\"azure-openai-responses\">, apiKey: string, options?: AzureOpenAIResponsesOptions) {\n\tif (!apiKey) {\n\t\tif (!process.env.AZURE_OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"Azure OpenAI API key is required. Set AZURE_OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.AZURE_OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\n\tif (options?.headers) {\n\t\tObject.assign(headers, options.headers);\n\t}\n\n\tconst { baseUrl, apiVersion } = resolveAzureConfig(model, options);\n\n\treturn new AzureOpenAI({\n\t\tapiKey,\n\t\tapiVersion,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t\tbaseURL: baseUrl,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions: AzureOpenAIResponsesOptions \| undefined,\n\tdeploymentName: string,\n) {\n\tconst messages = convertResponsesMessages(model, context, AZURE_TOOL_CALL_PROVIDERS);\n\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: deploymentName,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: options?.sessionId,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = options?.maxTokens;\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertResponsesTools(context.tools);\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort \|\| options?.reasoningSummary) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: options?.reasoningEffort \|\| \"medium\",\n\t\t\t\tsummary: options?.reasoningSummary \|\| \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else {\n\t\t\tparams.reasoning = { effort: \"none\" };\n\t\t}\n\t}\n\n\treturn params;\n}\n"]}

package/dist/providers/azure-openai-responses.js CHANGED Viewed

@@ -63,13 +63,12 @@ export const streamAzureOpenAIResponses = (model, context, options) => {
             if (nextParams !== undefined) {
                 params = nextParams;
             }
-            const { data: openaiStream, response } = await client.responses
-                .create(params, {
-                signal: options?.signal,
-                timeout: options?.timeoutMs,
-                maxRetries: options?.maxRetries,
-            })
-                .withResponse();
+            const requestOptions = {
+                ...(options?.signal ? { signal: options.signal } : {}),
+                ...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),
+                ...(options?.maxRetries !== undefined ? { maxRetries: options.maxRetries } : {}),
+            };
+            const { data: openaiStream, response } = await client.responses.create(params, requestOptions).withResponse();
             await options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);
             stream.push({ type: "start", partial: output });
             await processResponsesStream(openaiStream, output, stream, model);

package/dist/providers/azure-openai-responses.js.map CHANGED Viewed

	@@ -1 +1 @@
1	- {"version":3,"file":"azure-openai-responses.js","sourceRoot":"","sources":["../../src/providers/azure-openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,QAAQ,CAAC;AAErC,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAU7C,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAE,wBAAwB,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,8BAA8B,CAAC;AACvH,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAEvE,MAAM,yBAAyB,GAAG,IAAI,CAAC;AACvC,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,EAAE,wBAAwB,CAAC,CAAC,CAAC;AAE5G,SAAS,sBAAsB,CAAC,KAAyB,EAAuB;IAC/E,MAAM,GAAG,GAAG,IAAI,GAAG,EAAkB,CAAC;IACtC,IAAI,CAAC,KAAK;QAAE,OAAO,GAAG,CAAC;IACvB,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;QACtC,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;QAC7B,IAAI,CAAC,OAAO;YAAE,SAAS;QACvB,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;QACxD,IAAI,CAAC,OAAO,IAAI,CAAC,cAAc;YAAE,SAAS;QAC1C,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,EAAE,cAAc,CAAC,IAAI,EAAE,CAAC,CAAC;IAChD,CAAC;IACD,OAAO,GAAG,CAAC;AAAA,CACX;AAED,SAAS,qBAAqB,CAAC,KAAsC,EAAE,OAAqC,EAAU;IACrH,IAAI,OAAO,EAAE,mBAAmB,EAAE,CAAC;QAClC,OAAO,OAAO,CAAC,mBAAmB,CAAC;IACpC,CAAC;IACD,MAAM,gBAAgB,GAAG,sBAAsB,CAAC,OAAO,CAAC,GAAG,CAAC,gCAAgC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC5G,OAAO,gBAAgB,IAAI,KAAK,CAAC,EAAE,CAAC;AAAA,CACpC;AAYD;;GAEG;AACH,MAAM,CAAC,MAAM,0BAA0B,GAA0E,CAChH,KAAsC,EACtC,OAAgB,EAChB,OAAqC,EACP,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE,CAAC;QACZ,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;QAE7D,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,wBAA+B;YACpC,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,MAAM;YAClB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,6BAA6B;YAC7B,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACrE,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;YACpD,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;YAClE,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,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS;iBAC7D,MAAM,CAAC,MAAM,EAAE;gBACf,MAAM,EAAE,OAAO,EAAE,MAAM;gBACvB,OAAO,EAAE,OAAO,EAAE,SAAS;gBAC3B,UAAU,EAAE,OAAO,EAAE,UAAU;aAC/B,CAAC;iBACD,YAAY,EAAE,CAAC;YACjB,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,CAAC,CAAC;YAElE,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;YAC9C,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpC,OAAQ,KAA4B,CAAC,KAAK,CAAC;gBAC3C,oEAAoE;gBACpE,OAAQ,KAAkC,CAAC,WAAW,CAAC;YACxD,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,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YACrF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IAAA,CACD,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AAAA,CACd,CAAC;AAEF,MAAM,CAAC,MAAM,gCAAgC,GAAkE,CAC9G,KAAsC,EACtC,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,IAAI,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACtD,MAAM,eAAe,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;IAEvG,OAAO,0BAA0B,CAAC,KAAK,EAAE,OAAO,EAAE;QACjD,GAAG,IAAI;QACP,eAAe;KACuB,CAAC,CAAC;AAAA,CACzC,CAAC;AAEF,SAAS,qBAAqB,CAAC,OAAe,EAAU;IACvD,OAAO,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;AAAA,CACnC;AAED,SAAS,mBAAmB,CAAC,YAAoB,EAAU;IAC1D,OAAO,WAAW,YAAY,6BAA6B,CAAC;AAAA,CAC5D;AAED,SAAS,kBAAkB,CAC1B,KAAsC,EACtC,OAAqC,EACK;IAC1C,MAAM,UAAU,GAAG,OAAO,EAAE,eAAe,IAAI,OAAO,CAAC,GAAG,CAAC,wBAAwB,IAAI,yBAAyB,CAAC;IAEjH,MAAM,OAAO,GAAG,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,IAAI,OAAO,CAAC,GAAG,CAAC,qBAAqB,EAAE,IAAI,EAAE,IAAI,SAAS,CAAC;IACxG,MAAM,YAAY,GAAG,OAAO,EAAE,iBAAiB,IAAI,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC;IAE1F,IAAI,eAAe,GAAG,OAAO,CAAC;IAE9B,IAAI,CAAC,eAAe,IAAI,YAAY,EAAE,CAAC;QACtC,eAAe,GAAG,mBAAmB,CAAC,YAAY,CAAC,CAAC;IACrD,CAAC;IAED,IAAI,CAAC,eAAe,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;QACvC,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC;IACjC,CAAC;IAED,IAAI,CAAC,eAAe,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CACd,wJAAwJ,CACxJ,CAAC;IACH,CAAC;IAED,OAAO;QACN,OAAO,EAAE,qBAAqB,CAAC,eAAe,CAAC;QAC/C,UAAU;KACV,CAAC;AAAA,CACF;AAED,SAAS,YAAY,CAAC,KAAsC,EAAE,MAAc,EAAE,OAAqC,EAAE;IACpH,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,oBAAoB,EAAE,CAAC;YACvC,MAAM,IAAI,KAAK,CACd,4GAA4G,CAC5G,CAAC;QACH,CAAC;QACD,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC;IAC3C,CAAC;IAED,MAAM,OAAO,GAAG,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IAErC,IAAI,OAAO,EAAE,OAAO,EAAE,CAAC;QACtB,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC;IACzC,CAAC;IAED,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAEnE,OAAO,IAAI,WAAW,CAAC;QACtB,MAAM;QACN,UAAU;QACV,uBAAuB,EAAE,IAAI;QAC7B,cAAc,EAAE,OAAO;QACvB,OAAO,EAAE,OAAO;KAChB,CAAC,CAAC;AAAA,CACH;AAED,SAAS,WAAW,CACnB,KAAsC,EACtC,OAAgB,EAChB,OAAgD,EAChD,cAAsB,EACrB;IACD,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,yBAAyB,CAAC,CAAC;IAErF,MAAM,MAAM,GAAkC;QAC7C,KAAK,EAAE,cAAc;QACrB,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,gBAAgB,EAAE,OAAO,EAAE,SAAS;KACpC,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,OAAO,EAAE,SAAS,CAAC;IAC/C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IACrD,CAAC;IAED,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,OAAO,EAAE,eAAe,IAAI,OAAO,EAAE,gBAAgB,EAAE,CAAC;YAC3D,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,OAAO,EAAE,eAAe,IAAI,QAAQ;gBAC5C,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,CAAC;YACP,MAAM,CAAC,SAAS,GAAG,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;QACvC,CAAC;IACF,CAAC;IAED,OAAO,MAAM,CAAC;AAAA,CACd","sourcesContent":["import { AzureOpenAI } from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { supportsXhigh } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { headersToRecord } from \"../utils/headers.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\n\nconst DEFAULT_AZURE_API_VERSION = \"v1\";\nconst AZURE_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\", \"azure-openai-responses\"]);\n\nfunction parseDeploymentNameMap(value: string \| undefined): Map<string, string> {\n\tconst map = new Map<string, string>();\n\tif (!value) return map;\n\tfor (const entry of value.split(\",\")) {\n\t\tconst trimmed = entry.trim();\n\t\tif (!trimmed) continue;\n\t\tconst [modelId, deploymentName] = trimmed.split(\"=\", 2);\n\t\tif (!modelId \|\| !deploymentName) continue;\n\t\tmap.set(modelId.trim(), deploymentName.trim());\n\t}\n\treturn map;\n}\n\nfunction resolveDeploymentName(model: Model<\"azure-openai-responses\">, options?: AzureOpenAIResponsesOptions): string {\n\tif (options?.azureDeploymentName) {\n\t\treturn options.azureDeploymentName;\n\t}\n\tconst mappedDeployment = parseDeploymentNameMap(process.env.AZURE_OPENAI_DEPLOYMENT_NAME_MAP).get(model.id);\n\treturn mappedDeployment \|\| model.id;\n}\n\n// Azure OpenAI Responses-specific options\nexport interface AzureOpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" \| \"low\" \| \"medium\" \| \"high\" \| \"xhigh\";\n\treasoningSummary?: \"auto\" \| \"detailed\" \| \"concise\" \| null;\n\tazureApiVersion?: string;\n\tazureResourceName?: string;\n\tazureBaseUrl?: string;\n\tazureDeploymentName?: string;\n}\n\n/*\n Generate function for Azure OpenAI Responses API\n */\nexport const streamAzureOpenAIResponses: StreamFunction<\"azure-openai-responses\", AzureOpenAIResponsesOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions?: AzureOpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t// Start async processing\n\t(async () => {\n\t\tconst deploymentName = resolveDeploymentName(model, options);\n\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"azure-openai-responses\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create Azure OpenAI client\n\t\t\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider) \|\| \"\";\n\t\t\tconst client = createClient(model, apiKey, options);\n\t\t\tlet params = buildParams(model, context, options, deploymentName);\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 { data: openaiStream, response } = await client.responses\n\t\t\t\t.create(params, {\n\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t\ttimeout: options?.timeoutMs,\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t})\n\t\t\t\t.withResponse();\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\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" \|\| output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// partialJson is only a streaming scratch buffer; never persist it.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleAzureOpenAIResponses: StreamFunction<\"azure-openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\n\treturn streamAzureOpenAIResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies AzureOpenAIResponsesOptions);\n};\n\nfunction normalizeAzureBaseUrl(baseUrl: string): string {\n\treturn baseUrl.replace(/\\/+$/, \"\");\n}\n\nfunction buildDefaultBaseUrl(resourceName: string): string {\n\treturn `https://${resourceName}.openai.azure.com/openai/v1`;\n}\n\nfunction resolveAzureConfig(\n\tmodel: Model<\"azure-openai-responses\">,\n\toptions?: AzureOpenAIResponsesOptions,\n): { baseUrl: string; apiVersion: string } {\n\tconst apiVersion = options?.azureApiVersion \|\| process.env.AZURE_OPENAI_API_VERSION \|\| DEFAULT_AZURE_API_VERSION;\n\n\tconst baseUrl = options?.azureBaseUrl?.trim() \|\| process.env.AZURE_OPENAI_BASE_URL?.trim() \|\| undefined;\n\tconst resourceName = options?.azureResourceName \|\| process.env.AZURE_OPENAI_RESOURCE_NAME;\n\n\tlet resolvedBaseUrl = baseUrl;\n\n\tif (!resolvedBaseUrl && resourceName) {\n\t\tresolvedBaseUrl = buildDefaultBaseUrl(resourceName);\n\t}\n\n\tif (!resolvedBaseUrl && model.baseUrl) {\n\t\tresolvedBaseUrl = model.baseUrl;\n\t}\n\n\tif (!resolvedBaseUrl) {\n\t\tthrow new Error(\n\t\t\t\"Azure OpenAI base URL is required. Set AZURE_OPENAI_BASE_URL or AZURE_OPENAI_RESOURCE_NAME, or pass azureBaseUrl, azureResourceName, or model.baseUrl.\",\n\t\t);\n\t}\n\n\treturn {\n\t\tbaseUrl: normalizeAzureBaseUrl(resolvedBaseUrl),\n\t\tapiVersion,\n\t};\n}\n\nfunction createClient(model: Model<\"azure-openai-responses\">, apiKey: string, options?: AzureOpenAIResponsesOptions) {\n\tif (!apiKey) {\n\t\tif (!process.env.AZURE_OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"Azure OpenAI API key is required. Set AZURE_OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.AZURE_OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\n\tif (options?.headers) {\n\t\tObject.assign(headers, options.headers);\n\t}\n\n\tconst { baseUrl, apiVersion } = resolveAzureConfig(model, options);\n\n\treturn new AzureOpenAI({\n\t\tapiKey,\n\t\tapiVersion,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t\tbaseURL: baseUrl,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions: AzureOpenAIResponsesOptions \| undefined,\n\tdeploymentName: string,\n) {\n\tconst messages = convertResponsesMessages(model, context, AZURE_TOOL_CALL_PROVIDERS);\n\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: deploymentName,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: options?.sessionId,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = options?.maxTokens;\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertResponsesTools(context.tools);\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort \|\| options?.reasoningSummary) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: options?.reasoningEffort \|\| \"medium\",\n\t\t\t\tsummary: options?.reasoningSummary \|\| \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else {\n\t\t\tparams.reasoning = { effort: \"none\" };\n\t\t}\n\t}\n\n\treturn params;\n}\n"]}
1	+ {"version":3,"file":"azure-openai-responses.js","sourceRoot":"","sources":["../../src/providers/azure-openai-responses.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,QAAQ,CAAC;AAErC,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAU7C,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAE,wBAAwB,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,8BAA8B,CAAC;AACvH,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAEvE,MAAM,yBAAyB,GAAG,IAAI,CAAC;AACvC,MAAM,yBAAyB,GAAG,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,cAAc,EAAE,UAAU,EAAE,wBAAwB,CAAC,CAAC,CAAC;AAE5G,SAAS,sBAAsB,CAAC,KAAyB,EAAuB;IAC/E,MAAM,GAAG,GAAG,IAAI,GAAG,EAAkB,CAAC;IACtC,IAAI,CAAC,KAAK;QAAE,OAAO,GAAG,CAAC;IACvB,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;QACtC,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;QAC7B,IAAI,CAAC,OAAO;YAAE,SAAS;QACvB,MAAM,CAAC,OAAO,EAAE,cAAc,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;QACxD,IAAI,CAAC,OAAO,IAAI,CAAC,cAAc;YAAE,SAAS;QAC1C,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,EAAE,cAAc,CAAC,IAAI,EAAE,CAAC,CAAC;IAChD,CAAC;IACD,OAAO,GAAG,CAAC;AAAA,CACX;AAED,SAAS,qBAAqB,CAAC,KAAsC,EAAE,OAAqC,EAAU;IACrH,IAAI,OAAO,EAAE,mBAAmB,EAAE,CAAC;QAClC,OAAO,OAAO,CAAC,mBAAmB,CAAC;IACpC,CAAC;IACD,MAAM,gBAAgB,GAAG,sBAAsB,CAAC,OAAO,CAAC,GAAG,CAAC,gCAAgC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC5G,OAAO,gBAAgB,IAAI,KAAK,CAAC,EAAE,CAAC;AAAA,CACpC;AAYD;;GAEG;AACH,MAAM,CAAC,MAAM,0BAA0B,GAA0E,CAChH,KAAsC,EACtC,OAAgB,EAChB,OAAqC,EACP,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,yBAAyB;IACzB,CAAC,KAAK,IAAI,EAAE,CAAC;QACZ,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;QAE7D,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,wBAA+B;YACpC,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,MAAM;YAClB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,6BAA6B;YAC7B,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACrE,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;YACpD,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;YAClE,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,GAAG,CAAC,OAAO,EAAE,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAChF,CAAC;YACF,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,YAAY,EAAE,CAAC;YAC9G,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,CAAC,CAAC;YAElE,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;YAC9C,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpC,OAAQ,KAA4B,CAAC,KAAK,CAAC;gBAC3C,oEAAoE;gBACpE,OAAQ,KAAkC,CAAC,WAAW,CAAC;YACxD,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,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YACrF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IAAA,CACD,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AAAA,CACd,CAAC;AAEF,MAAM,CAAC,MAAM,gCAAgC,GAAkE,CAC9G,KAAsC,EACtC,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,IAAI,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACtD,MAAM,eAAe,GAAG,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;IAEvG,OAAO,0BAA0B,CAAC,KAAK,EAAE,OAAO,EAAE;QACjD,GAAG,IAAI;QACP,eAAe;KACuB,CAAC,CAAC;AAAA,CACzC,CAAC;AAEF,SAAS,qBAAqB,CAAC,OAAe,EAAU;IACvD,OAAO,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;AAAA,CACnC;AAED,SAAS,mBAAmB,CAAC,YAAoB,EAAU;IAC1D,OAAO,WAAW,YAAY,6BAA6B,CAAC;AAAA,CAC5D;AAED,SAAS,kBAAkB,CAC1B,KAAsC,EACtC,OAAqC,EACK;IAC1C,MAAM,UAAU,GAAG,OAAO,EAAE,eAAe,IAAI,OAAO,CAAC,GAAG,CAAC,wBAAwB,IAAI,yBAAyB,CAAC;IAEjH,MAAM,OAAO,GAAG,OAAO,EAAE,YAAY,EAAE,IAAI,EAAE,IAAI,OAAO,CAAC,GAAG,CAAC,qBAAqB,EAAE,IAAI,EAAE,IAAI,SAAS,CAAC;IACxG,MAAM,YAAY,GAAG,OAAO,EAAE,iBAAiB,IAAI,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC;IAE1F,IAAI,eAAe,GAAG,OAAO,CAAC;IAE9B,IAAI,CAAC,eAAe,IAAI,YAAY,EAAE,CAAC;QACtC,eAAe,GAAG,mBAAmB,CAAC,YAAY,CAAC,CAAC;IACrD,CAAC;IAED,IAAI,CAAC,eAAe,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;QACvC,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC;IACjC,CAAC;IAED,IAAI,CAAC,eAAe,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CACd,wJAAwJ,CACxJ,CAAC;IACH,CAAC;IAED,OAAO;QACN,OAAO,EAAE,qBAAqB,CAAC,eAAe,CAAC;QAC/C,UAAU;KACV,CAAC;AAAA,CACF;AAED,SAAS,YAAY,CAAC,KAAsC,EAAE,MAAc,EAAE,OAAqC,EAAE;IACpH,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,oBAAoB,EAAE,CAAC;YACvC,MAAM,IAAI,KAAK,CACd,4GAA4G,CAC5G,CAAC;QACH,CAAC;QACD,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC;IAC3C,CAAC;IAED,MAAM,OAAO,GAAG,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,CAAC;IAErC,IAAI,OAAO,EAAE,OAAO,EAAE,CAAC;QACtB,MAAM,CAAC,MAAM,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC;IACzC,CAAC;IAED,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,kBAAkB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAEnE,OAAO,IAAI,WAAW,CAAC;QACtB,MAAM;QACN,UAAU;QACV,uBAAuB,EAAE,IAAI;QAC7B,cAAc,EAAE,OAAO;QACvB,OAAO,EAAE,OAAO;KAChB,CAAC,CAAC;AAAA,CACH;AAED,SAAS,WAAW,CACnB,KAAsC,EACtC,OAAgB,EAChB,OAAgD,EAChD,cAAsB,EACrB;IACD,MAAM,QAAQ,GAAG,wBAAwB,CAAC,KAAK,EAAE,OAAO,EAAE,yBAAyB,CAAC,CAAC;IAErF,MAAM,MAAM,GAAkC;QAC7C,KAAK,EAAE,cAAc;QACrB,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,gBAAgB,EAAE,OAAO,EAAE,SAAS;KACpC,CAAC;IAEF,IAAI,OAAO,EAAE,SAAS,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,OAAO,EAAE,SAAS,CAAC;IAC/C,CAAC;IAED,IAAI,OAAO,EAAE,WAAW,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,CAAC,WAAW,GAAG,OAAO,EAAE,WAAW,CAAC;IAC3C,CAAC;IAED,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,CAAC,KAAK,GAAG,qBAAqB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IACrD,CAAC;IAED,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QACrB,IAAI,OAAO,EAAE,eAAe,IAAI,OAAO,EAAE,gBAAgB,EAAE,CAAC;YAC3D,MAAM,CAAC,SAAS,GAAG;gBAClB,MAAM,EAAE,OAAO,EAAE,eAAe,IAAI,QAAQ;gBAC5C,OAAO,EAAE,OAAO,EAAE,gBAAgB,IAAI,MAAM;aAC5C,CAAC;YACF,MAAM,CAAC,OAAO,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAClD,CAAC;aAAM,CAAC;YACP,MAAM,CAAC,SAAS,GAAG,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;QACvC,CAAC;IACF,CAAC;IAED,OAAO,MAAM,CAAC;AAAA,CACd","sourcesContent":["import { AzureOpenAI } from \"openai\";\nimport type { ResponseCreateParamsStreaming } from \"openai/resources/responses/responses.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { supportsXhigh } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { headersToRecord } from \"../utils/headers.js\";\nimport { convertResponsesMessages, convertResponsesTools, processResponsesStream } from \"./openai-responses-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\n\nconst DEFAULT_AZURE_API_VERSION = \"v1\";\nconst AZURE_TOOL_CALL_PROVIDERS = new Set([\"openai\", \"openai-codex\", \"opencode\", \"azure-openai-responses\"]);\n\nfunction parseDeploymentNameMap(value: string \| undefined): Map<string, string> {\n\tconst map = new Map<string, string>();\n\tif (!value) return map;\n\tfor (const entry of value.split(\",\")) {\n\t\tconst trimmed = entry.trim();\n\t\tif (!trimmed) continue;\n\t\tconst [modelId, deploymentName] = trimmed.split(\"=\", 2);\n\t\tif (!modelId \|\| !deploymentName) continue;\n\t\tmap.set(modelId.trim(), deploymentName.trim());\n\t}\n\treturn map;\n}\n\nfunction resolveDeploymentName(model: Model<\"azure-openai-responses\">, options?: AzureOpenAIResponsesOptions): string {\n\tif (options?.azureDeploymentName) {\n\t\treturn options.azureDeploymentName;\n\t}\n\tconst mappedDeployment = parseDeploymentNameMap(process.env.AZURE_OPENAI_DEPLOYMENT_NAME_MAP).get(model.id);\n\treturn mappedDeployment \|\| model.id;\n}\n\n// Azure OpenAI Responses-specific options\nexport interface AzureOpenAIResponsesOptions extends StreamOptions {\n\treasoningEffort?: \"minimal\" \| \"low\" \| \"medium\" \| \"high\" \| \"xhigh\";\n\treasoningSummary?: \"auto\" \| \"detailed\" \| \"concise\" \| null;\n\tazureApiVersion?: string;\n\tazureResourceName?: string;\n\tazureBaseUrl?: string;\n\tazureDeploymentName?: string;\n}\n\n/*\n Generate function for Azure OpenAI Responses API\n */\nexport const streamAzureOpenAIResponses: StreamFunction<\"azure-openai-responses\", AzureOpenAIResponsesOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions?: AzureOpenAIResponsesOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t// Start async processing\n\t(async () => {\n\t\tconst deploymentName = resolveDeploymentName(model, options);\n\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"azure-openai-responses\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\t// Create Azure OpenAI client\n\t\t\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider) \|\| \"\";\n\t\t\tconst client = createClient(model, apiKey, options);\n\t\t\tlet params = buildParams(model, context, options, deploymentName);\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\t...(options?.maxRetries !== undefined ? { maxRetries: options.maxRetries } : {}),\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await client.responses.create(params, requestOptions).withResponse();\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\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" \|\| output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t// partialJson is only a streaming scratch buffer; never persist it.\n\t\t\t\tdelete (block as { partialJson?: string }).partialJson;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleAzureOpenAIResponses: StreamFunction<\"azure-openai-responses\", SimpleStreamOptions> = (\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\n\treturn streamAzureOpenAIResponses(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t} satisfies AzureOpenAIResponsesOptions);\n};\n\nfunction normalizeAzureBaseUrl(baseUrl: string): string {\n\treturn baseUrl.replace(/\\/+$/, \"\");\n}\n\nfunction buildDefaultBaseUrl(resourceName: string): string {\n\treturn `https://${resourceName}.openai.azure.com/openai/v1`;\n}\n\nfunction resolveAzureConfig(\n\tmodel: Model<\"azure-openai-responses\">,\n\toptions?: AzureOpenAIResponsesOptions,\n): { baseUrl: string; apiVersion: string } {\n\tconst apiVersion = options?.azureApiVersion \|\| process.env.AZURE_OPENAI_API_VERSION \|\| DEFAULT_AZURE_API_VERSION;\n\n\tconst baseUrl = options?.azureBaseUrl?.trim() \|\| process.env.AZURE_OPENAI_BASE_URL?.trim() \|\| undefined;\n\tconst resourceName = options?.azureResourceName \|\| process.env.AZURE_OPENAI_RESOURCE_NAME;\n\n\tlet resolvedBaseUrl = baseUrl;\n\n\tif (!resolvedBaseUrl && resourceName) {\n\t\tresolvedBaseUrl = buildDefaultBaseUrl(resourceName);\n\t}\n\n\tif (!resolvedBaseUrl && model.baseUrl) {\n\t\tresolvedBaseUrl = model.baseUrl;\n\t}\n\n\tif (!resolvedBaseUrl) {\n\t\tthrow new Error(\n\t\t\t\"Azure OpenAI base URL is required. Set AZURE_OPENAI_BASE_URL or AZURE_OPENAI_RESOURCE_NAME, or pass azureBaseUrl, azureResourceName, or model.baseUrl.\",\n\t\t);\n\t}\n\n\treturn {\n\t\tbaseUrl: normalizeAzureBaseUrl(resolvedBaseUrl),\n\t\tapiVersion,\n\t};\n}\n\nfunction createClient(model: Model<\"azure-openai-responses\">, apiKey: string, options?: AzureOpenAIResponsesOptions) {\n\tif (!apiKey) {\n\t\tif (!process.env.AZURE_OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"Azure OpenAI API key is required. Set AZURE_OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.AZURE_OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\n\tif (options?.headers) {\n\t\tObject.assign(headers, options.headers);\n\t}\n\n\tconst { baseUrl, apiVersion } = resolveAzureConfig(model, options);\n\n\treturn new AzureOpenAI({\n\t\tapiKey,\n\t\tapiVersion,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t\tbaseURL: baseUrl,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"azure-openai-responses\">,\n\tcontext: Context,\n\toptions: AzureOpenAIResponsesOptions \| undefined,\n\tdeploymentName: string,\n) {\n\tconst messages = convertResponsesMessages(model, context, AZURE_TOOL_CALL_PROVIDERS);\n\n\tconst params: ResponseCreateParamsStreaming = {\n\t\tmodel: deploymentName,\n\t\tinput: messages,\n\t\tstream: true,\n\t\tprompt_cache_key: options?.sessionId,\n\t};\n\n\tif (options?.maxTokens) {\n\t\tparams.max_output_tokens = options?.maxTokens;\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options?.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertResponsesTools(context.tools);\n\t}\n\n\tif (model.reasoning) {\n\t\tif (options?.reasoningEffort \|\| options?.reasoningSummary) {\n\t\t\tparams.reasoning = {\n\t\t\t\teffort: options?.reasoningEffort \|\| \"medium\",\n\t\t\t\tsummary: options?.reasoningSummary \|\| \"auto\",\n\t\t\t};\n\t\t\tparams.include = [\"reasoning.encrypted_content\"];\n\t\t} else {\n\t\t\tparams.reasoning = { effort: \"none\" };\n\t\t}\n\t}\n\n\treturn params;\n}\n"]}

package/dist/providers/openai-completions.d.ts.map CHANGED Viewed

	@@ -1 +1 @@
1	- {"version":3,"file":"openai-completions.d.ts","sourceRoot":"","sources":["../../src/providers/openai-completions.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAOX,0BAA0B,EAG1B,MAAM,sCAAsC,CAAC;AAG9C,OAAO,KAAK,EAGX,OAAO,EAGP,KAAK,EACL,uBAAuB,EACvB,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAMb,MAAM,aAAa,CAAC;AA4CrB,MAAM,WAAW,wBAAyB,SAAQ,aAAa;IAC9D,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,eAAe,CAAC,EAAE,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;CAClE;AAOD,KAAK,+BAA+B,GAAG,IAAI,CAAC,QAAQ,CAAC,uBAAuB,CAAC,EAAE,oBAAoB,CAAC,GAAG;IACtG,kBAAkB,CAAC,EAAE,uBAAuB,CAAC,oBAAoB,CAAC,CAAC;CACnE,CAAC;AAsBF,eAAO,MAAM,uBAAuB,EAAE,cAAc,CAAC,oBAAoB,EAAE,wBAAwB,CA6RlG,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,cAAc,CAAC,oBAAoB,EAAE,mBAAmB,CAmBnG,CAAC;AAsRF,wBAAgB,eAAe,CAC9B,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,EAClC,OAAO,EAAE,OAAO,EAChB,MAAM,EAAE,+BAA+B,GACrC,0BAA0B,EAAE,CA6O9B","sourcesContent":["import OpenAI from \"openai\";\nimport type {\n\tChatCompletionAssistantMessageParam,\n\tChatCompletionChunk,\n\tChatCompletionContentPart,\n\tChatCompletionContentPartImage,\n\tChatCompletionContentPartText,\n\tChatCompletionDeveloperMessageParam,\n\tChatCompletionMessageParam,\n\tChatCompletionSystemMessageParam,\n\tChatCompletionToolMessageParam,\n} from \"openai/resources/chat/completions.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost, supportsXhigh } from \"../models.js\";\nimport type {\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tImageContent,\n\tMessage,\n\tModel,\n\tOpenAICompletionsCompat,\n\tSimpleStreamOptions,\n\tStopReason,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingContent,\n\tTool,\n\tToolCall,\n\tToolResultMessage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { headersToRecord } from \"../utils/headers.js\";\nimport { parseStreamingJson } from \"../utils/json-parse.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\nimport { transformMessages } from \"./transform-messages.js\";\n\n/*\n Check if conversation messages contain tool calls or tool results.\n * This is needed because Anthropic (via proxy) requires the tools param\n * to be present when messages include tool_calls or tool role messages.\n /\nfunction hasToolHistory(messages: Message[]): boolean {\n\tfor (const msg of messages) {\n\t\tif (msg.role === \"toolResult\") {\n\t\t\treturn true;\n\t\t}\n\t\tif (msg.role === \"assistant\") {\n\t\t\tif (msg.content.some((block) => block.type === \"toolCall\")) {\n\t\t\t\treturn true;\n\t\t\t}\n\t\t}\n\t}\n\treturn false;\n}\n\nfunction isTextContentBlock(block: { type: string }): block is TextContent {\n\treturn block.type === \"text\";\n}\n\nfunction isThinkingContentBlock(block: { type: string }): block is ThinkingContent {\n\treturn block.type === \"thinking\";\n}\n\nfunction isToolCallBlock(block: { type: string }): block is ToolCall {\n\treturn block.type === \"toolCall\";\n}\n\nfunction isImageContentBlock(block: { type: string }): block is ImageContent {\n\treturn block.type === \"image\";\n}\n\nexport interface OpenAICompletionsOptions extends StreamOptions {\n\ttoolChoice?: \"auto\" \| \"none\" \| \"required\" \| { type: \"function\"; function: { name: string } };\n\treasoningEffort?: \"minimal\" \| \"low\" \| \"medium\" \| \"high\" \| \"xhigh\";\n}\n\ninterface OpenAICompatCacheControl {\n\ttype: \"ephemeral\";\n\tttl?: string;\n}\n\ntype ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, \"cacheControlFormat\"> & {\n\tcacheControlFormat?: OpenAICompletionsCompat[\"cacheControlFormat\"];\n};\n\ntype ChatCompletionInstructionMessageParam = ChatCompletionDeveloperMessageParam \| ChatCompletionSystemMessageParam;\n\ntype ChatCompletionTextPartWithCacheControl = ChatCompletionContentPartText & {\n\tcache_control?: OpenAICompatCacheControl;\n};\n\ntype ChatCompletionToolWithCacheControl = OpenAI.Chat.Completions.ChatCompletionTool & {\n\tcache_control?: OpenAICompatCacheControl;\n};\n\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.PI_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nexport const streamOpenAICompletions: StreamFunction<\"openai-completions\", OpenAICompletionsOptions> = (\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\toptions?: OpenAICompletionsOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider) \|\| \"\";\n\t\t\tconst compat = getCompat(model);\n\t\t\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention);\n\t\t\tconst cacheSessionId = cacheRetention === \"none\" ? undefined : options?.sessionId;\n\t\t\tconst client = createClient(model, context, apiKey, options?.headers, cacheSessionId, compat);\n\t\t\tlet params = buildParams(model, context, options, compat, cacheRetention);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst { data: openaiStream, response } = await client.chat.completions\n\t\t\t\t.create(params, {\n\t\t\t\t\tsignal: options?.signal,\n\t\t\t\t\ttimeout: options?.timeoutMs,\n\t\t\t\t\tmaxRetries: options?.maxRetries,\n\t\t\t\t})\n\t\t\t\t.withResponse();\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\tinterface StreamingToolCallBlock extends ToolCall {\n\t\t\t\tpartialArgs?: string;\n\t\t\t\tstreamIndex?: number;\n\t\t\t}\n\n\t\t\tlet currentBlock: TextContent \| ThinkingContent \| StreamingToolCallBlock \| null = null;\n\t\t\tconst blocks = output.content;\n\t\t\tconst getContentIndex = (block: typeof currentBlock) => (block ? blocks.indexOf(block) : -1);\n\t\t\tconst currentContentIndex = () => getContentIndex(currentBlock);\n\t\t\tconst finishCurrentBlock = (block?: typeof currentBlock) => {\n\t\t\t\tif (block) {\n\t\t\t\t\tconst contentIndex = getContentIndex(block);\n\t\t\t\t\tif (contentIndex === -1) {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tif (block.type === \"text\") {\n\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\tcontentIndex,\n\t\t\t\t\t\t\tcontent: block.text,\n\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t});\n\t\t\t\t\t} else if (block.type === \"thinking\") {\n\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\tcontentIndex,\n\t\t\t\t\t\t\tcontent: block.thinking,\n\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t});\n\t\t\t\t\t} else if (block.type === \"toolCall\") {\n\t\t\t\t\t\tblock.arguments = parseStreamingJson(block.partialArgs);\n\t\t\t\t\t\t// Finalize in-place and strip the scratch buffers so replay only\n\t\t\t\t\t\t// carries parsed arguments.\n\t\t\t\t\t\tdelete block.partialArgs;\n\t\t\t\t\t\tdelete block.streamIndex;\n\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\ttype: \"toolcall_end\",\n\t\t\t\t\t\t\tcontentIndex,\n\t\t\t\t\t\t\ttoolCall: block,\n\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t};\n\n\t\t\tfor await (const chunk of openaiStream) {\n\t\t\t\tif (!chunk \|\| typeof chunk !== \"object\") continue;\n\n\t\t\t\t// OpenAI documents ChatCompletionChunk.id as the unique chat completion identifier,\n\t\t\t\t// and each chunk in a streamed completion carries the same id.\n\t\t\t\toutput.responseId \|\|= chunk.id;\n\t\t\t\tif (chunk.usage) {\n\t\t\t\t\toutput.usage = parseChunkUsage(chunk.usage, model);\n\t\t\t\t}\n\n\t\t\t\tconst choice = Array.isArray(chunk.choices) ? chunk.choices[0] : undefined;\n\t\t\t\tif (!choice) continue;\n\n\t\t\t\t// Fallback: some providers (e.g., Moonshot) return usage\n\t\t\t\t// in choice.usage instead of the standard chunk.usage\n\t\t\t\tif (!chunk.usage && (choice as any).usage) {\n\t\t\t\t\toutput.usage = parseChunkUsage((choice as any).usage, model);\n\t\t\t\t}\n\n\t\t\t\tif (choice.finish_reason) {\n\t\t\t\t\tconst finishReasonResult = mapStopReason(choice.finish_reason);\n\t\t\t\t\toutput.stopReason = finishReasonResult.stopReason;\n\t\t\t\t\tif (finishReasonResult.errorMessage) {\n\t\t\t\t\t\toutput.errorMessage = finishReasonResult.errorMessage;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tif (choice.delta) {\n\t\t\t\t\tif (\n\t\t\t\t\t\tchoice.delta.content !== null &&\n\t\t\t\t\t\tchoice.delta.content !== undefined &&\n\t\t\t\t\t\tchoice.delta.content.length > 0\n\t\t\t\t\t) {\n\t\t\t\t\t\tif (!currentBlock \|\| currentBlock.type !== \"text\") {\n\t\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: currentContentIndex(), partial: output });\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\tcurrentBlock.text += choice.delta.content;\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: currentContentIndex(),\n\t\t\t\t\t\t\t\tdelta: choice.delta.content,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\t// Some endpoints return reasoning in reasoning_content (llama.cpp),\n\t\t\t\t\t// or reasoning (other openai compatible endpoints)\n\t\t\t\t\t// Use the first non-empty reasoning field to avoid duplication\n\t\t\t\t\t// (e.g., chutes.ai returns both reasoning_content and reasoning with same content)\n\t\t\t\t\tconst reasoningFields = [\"reasoning_content\", \"reasoning\", \"reasoning_text\"];\n\t\t\t\t\tlet foundReasoningField: string \| null = null;\n\t\t\t\t\tfor (const field of reasoningFields) {\n\t\t\t\t\t\tif (\n\t\t\t\t\t\t\t(choice.delta as any)[field] !== null &&\n\t\t\t\t\t\t\t(choice.delta as any)[field] !== undefined &&\n\t\t\t\t\t\t\t(choice.delta as any)[field].length > 0\n\t\t\t\t\t\t) {\n\t\t\t\t\t\t\tif (!foundReasoningField) {\n\t\t\t\t\t\t\t\tfoundReasoningField = field;\n\t\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\tif (foundReasoningField) {\n\t\t\t\t\t\tif (!currentBlock \|\| currentBlock.type !== \"thinking\") {\n\t\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\t\tcurrentBlock = {\n\t\t\t\t\t\t\t\ttype: \"thinking\",\n\t\t\t\t\t\t\t\tthinking: \"\",\n\t\t\t\t\t\t\t\tthinkingSignature: foundReasoningField,\n\t\t\t\t\t\t\t};\n\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: currentContentIndex(), partial: output });\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tif (currentBlock.type === \"thinking\") {\n\t\t\t\t\t\t\tconst delta = (choice.delta as any)[foundReasoningField];\n\t\t\t\t\t\t\tcurrentBlock.thinking += delta;\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: currentContentIndex(),\n\t\t\t\t\t\t\t\tdelta,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\tif (choice?.delta?.tool_calls) {\n\t\t\t\t\t\tfor (const toolCall of choice.delta.tool_calls) {\n\t\t\t\t\t\t\tconst streamIndex = typeof toolCall.index === \"number\" ? toolCall.index : undefined;\n\t\t\t\t\t\t\tconst sameToolCall =\n\t\t\t\t\t\t\t\tcurrentBlock?.type === \"toolCall\" &&\n\t\t\t\t\t\t\t\t((streamIndex !== undefined && currentBlock.streamIndex === streamIndex) \|\|\n\t\t\t\t\t\t\t\t\t(streamIndex === undefined && toolCall.id && currentBlock.id === toolCall.id));\n\n\t\t\t\t\t\t\tif (!sameToolCall) {\n\t\t\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\t\t\tcurrentBlock = {\n\t\t\t\t\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\t\t\t\t\tid: toolCall.id \|\| \"\",\n\t\t\t\t\t\t\t\t\tname: toolCall.function?.name \|\| \"\",\n\t\t\t\t\t\t\t\t\targuments: {},\n\t\t\t\t\t\t\t\t\tpartialArgs: \"\",\n\t\t\t\t\t\t\t\t\tstreamIndex,\n\t\t\t\t\t\t\t\t};\n\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"toolcall_start\",\n\t\t\t\t\t\t\t\t\tcontentIndex: getContentIndex(currentBlock),\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\tconst currentToolCallBlock = currentBlock?.type === \"toolCall\" ? currentBlock : null;\n\t\t\t\t\t\t\tif (currentToolCallBlock) {\n\t\t\t\t\t\t\t\tif (!currentToolCallBlock.id && toolCall.id) currentToolCallBlock.id = toolCall.id;\n\t\t\t\t\t\t\t\tif (!currentToolCallBlock.name && toolCall.function?.name) {\n\t\t\t\t\t\t\t\t\tcurrentToolCallBlock.name = toolCall.function.name;\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tif (currentToolCallBlock.streamIndex === undefined && streamIndex !== undefined) {\n\t\t\t\t\t\t\t\t\tcurrentToolCallBlock.streamIndex = streamIndex;\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tlet delta = \"\";\n\t\t\t\t\t\t\t\tif (toolCall.function?.arguments) {\n\t\t\t\t\t\t\t\t\tdelta = toolCall.function.arguments;\n\t\t\t\t\t\t\t\t\tcurrentToolCallBlock.partialArgs += toolCall.function.arguments;\n\t\t\t\t\t\t\t\t\tcurrentToolCallBlock.arguments = parseStreamingJson(currentToolCallBlock.partialArgs);\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\t\t\t\t\t\tcontentIndex: getContentIndex(currentToolCallBlock),\n\t\t\t\t\t\t\t\t\tdelta,\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\tconst reasoningDetails = (choice.delta as any).reasoning_details;\n\t\t\t\t\tif (reasoningDetails && Array.isArray(reasoningDetails)) {\n\t\t\t\t\t\tfor (const detail of reasoningDetails) {\n\t\t\t\t\t\t\tif (detail.type === \"reasoning.encrypted\" && detail.id && detail.data) {\n\t\t\t\t\t\t\t\tconst matchingToolCall = output.content.find(\n\t\t\t\t\t\t\t\t\t(b) => b.type === \"toolCall\" && b.id === detail.id,\n\t\t\t\t\t\t\t\t) as ToolCall \| undefined;\n\t\t\t\t\t\t\t\tif (matchingToolCall) {\n\t\t\t\t\t\t\t\t\tmatchingToolCall.thoughtSignature = JSON.stringify(detail);\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\") {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage \|\| \"Provider returned an error stop reason\");\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 { partialArgs?: string }).partialArgs;\n\t\t\t\tdelete (block as { streamIndex?: number }).streamIndex;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\t// Some providers via OpenRouter give additional information in this field.\n\t\t\tconst rawMetadata = (error as any)?.error?.metadata?.raw;\n\t\t\tif (rawMetadata) output.errorMessage += `\\n${rawMetadata}`;\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 streamSimpleOpenAICompletions: StreamFunction<\"openai-completions\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\tconst toolChoice = (options as OpenAICompletionsOptions \| undefined)?.toolChoice;\n\n\treturn streamOpenAICompletions(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t\ttoolChoice,\n\t} satisfies OpenAICompletionsOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\tapiKey?: string,\n\toptionsHeaders?: Record<string, string>,\n\tsessionId?: string,\n\tcompat: ResolvedOpenAICompletionsCompat = getCompat(model),\n) {\n\tif (!apiKey) {\n\t\tif (!process.env.OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"OpenAI API key is required. Set OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = buildCopilotDynamicHeaders({\n\t\t\tmessages: context.messages,\n\t\t\thasImages,\n\t\t});\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\tif (sessionId && compat.sendSessionAffinityHeaders) {\n\t\theaders.session_id = sessionId;\n\t\theaders[\"x-client-request-id\"] = sessionId;\n\t\theaders[\"x-session-affinity\"] = sessionId;\n\t}\n\n\t// Merge options headers last so they can override defaults\n\tif (optionsHeaders) {\n\t\tObject.assign(headers, optionsHeaders);\n\t}\n\n\treturn new OpenAI({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\toptions?: OpenAICompletionsOptions,\n\tcompat: ResolvedOpenAICompletionsCompat = getCompat(model),\n\tcacheRetention: CacheRetention = resolveCacheRetention(options?.cacheRetention),\n) {\n\tconst messages = convertMessages(model, context, compat);\n\tconst cacheControl = getCompatCacheControl(compat, cacheRetention);\n\n\tconst params: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tmessages,\n\t\tstream: true,\n\t\tprompt_cache_key:\n\t\t\t(model.baseUrl.includes(\"api.openai.com\") && cacheRetention !== \"none\") \|\|\n\t\t\t(cacheRetention === \"long\" && compat.supportsLongCacheRetention)\n\t\t\t\t? options?.sessionId\n\t\t\t\t: undefined,\n\t\tprompt_cache_retention: cacheRetention === \"long\" && compat.supportsLongCacheRetention ? \"24h\" : undefined,\n\t};\n\n\tif (compat.supportsUsageInStreaming !== false) {\n\t\t(params as any).stream_options = { include_usage: true };\n\t}\n\n\tif (compat.supportsStore) {\n\t\tparams.store = false;\n\t}\n\n\tif (options?.maxTokens) {\n\t\tif (compat.maxTokensField === \"max_tokens\") {\n\t\t\t(params as any).max_tokens = options.maxTokens;\n\t\t} else {\n\t\t\tparams.max_completion_tokens = options.maxTokens;\n\t\t}\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertTools(context.tools, compat);\n\t\tif (compat.zaiToolStream) {\n\t\t\t(params as any).tool_stream = true;\n\t\t}\n\t} else if (hasToolHistory(context.messages)) {\n\t\t// Anthropic (via LiteLLM/proxy) requires tools param when conversation has tool_calls/tool_results\n\t\tparams.tools = [];\n\t}\n\n\tif (cacheControl) {\n\t\tapplyAnthropicCacheControl(messages, params.tools, cacheControl);\n\t}\n\n\tif (options?.toolChoice) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tif (compat.thinkingFormat === \"zai\" && model.reasoning) {\n\t\t(params as any).enable_thinking = !!options?.reasoningEffort;\n\t} else if (compat.thinkingFormat === \"qwen\" && model.reasoning) {\n\t\t(params as any).enable_thinking = !!options?.reasoningEffort;\n\t} else if (compat.thinkingFormat === \"qwen-chat-template\" && model.reasoning) {\n\t\t(params as any).chat_template_kwargs = {\n\t\t\tenable_thinking: !!options?.reasoningEffort,\n\t\t\tpreserve_thinking: true,\n\t\t};\n\t} else if (compat.thinkingFormat === \"deepseek\" && model.reasoning) {\n\t\t(params as any).thinking = { type: options?.reasoningEffort ? \"enabled\" : \"disabled\" };\n\t\tif (options?.reasoningEffort) {\n\t\t\t(params as any).reasoning_effort = mapReasoningEffort(options.reasoningEffort, compat.reasoningEffortMap);\n\t\t}\n\t} else if (compat.thinkingFormat === \"openrouter\" && model.reasoning) {\n\t\t// OpenRouter normalizes reasoning across providers via a nested reasoning object.\n\t\tconst openRouterParams = params as typeof params & { reasoning?: { effort?: string } };\n\t\tif (options?.reasoningEffort) {\n\t\t\topenRouterParams.reasoning = {\n\t\t\t\teffort: mapReasoningEffort(options.reasoningEffort, compat.reasoningEffortMap),\n\t\t\t};\n\t\t} else {\n\t\t\topenRouterParams.reasoning = { effort: \"none\" };\n\t\t}\n\t} else if (options?.reasoningEffort && model.reasoning && compat.supportsReasoningEffort) {\n\t\t// OpenAI-style reasoning_effort\n\t\t(params as any).reasoning_effort = mapReasoningEffort(options.reasoningEffort, compat.reasoningEffortMap);\n\t}\n\n\t// OpenRouter provider routing preferences\n\tif (model.baseUrl.includes(\"openrouter.ai\") && model.compat?.openRouterRouting) {\n\t\t(params as any).provider = model.compat.openRouterRouting;\n\t}\n\n\t// Vercel AI Gateway provider routing preferences\n\tif (model.baseUrl.includes(\"ai-gateway.vercel.sh\") && model.compat?.vercelGatewayRouting) {\n\t\tconst routing = model.compat.vercelGatewayRouting;\n\t\tif (routing.only \|\| routing.order) {\n\t\t\tconst gatewayOptions: Record<string, string[]> = {};\n\t\t\tif (routing.only) gatewayOptions.only = routing.only;\n\t\t\tif (routing.order) gatewayOptions.order = routing.order;\n\t\t\t(params as any).providerOptions = { gateway: gatewayOptions };\n\t\t}\n\t}\n\n\treturn params;\n}\n\nfunction mapReasoningEffort(\n\teffort: NonNullable<OpenAICompletionsOptions[\"reasoningEffort\"]>,\n\treasoningEffortMap: Partial<Record<NonNullable<OpenAICompletionsOptions[\"reasoningEffort\"]>, string>>,\n): string {\n\treturn reasoningEffortMap[effort] ?? effort;\n}\n\nfunction getCompatCacheControl(\n\tcompat: ResolvedOpenAICompletionsCompat,\n\tcacheRetention: CacheRetention,\n): OpenAICompatCacheControl \| undefined {\n\tif (compat.cacheControlFormat !== \"anthropic\" \|\| cacheRetention === \"none\") {\n\t\treturn undefined;\n\t}\n\n\tconst ttl = cacheRetention === \"long\" && compat.supportsLongCacheRetention ? \"1h\" : undefined;\n\treturn { type: \"ephemeral\", ...(ttl ? { ttl } : {}) };\n}\n\nfunction applyAnthropicCacheControl(\n\tmessages: ChatCompletionMessageParam[],\n\ttools: OpenAI.Chat.Completions.ChatCompletionTool[] \| undefined,\n\tcacheControl: OpenAICompatCacheControl,\n): void {\n\taddCacheControlToSystemPrompt(messages, cacheControl);\n\taddCacheControlToLastTool(tools, cacheControl);\n\taddCacheControlToLastConversationMessage(messages, cacheControl);\n}\n\nfunction addCacheControlToSystemPrompt(\n\tmessages: ChatCompletionMessageParam[],\n\tcacheControl: OpenAICompatCacheControl,\n): void {\n\tfor (const message of messages) {\n\t\tif (message.role === \"system\" \|\| message.role === \"developer\") {\n\t\t\taddCacheControlToInstructionMessage(message, cacheControl);\n\t\t\treturn;\n\t\t}\n\t}\n}\n\nfunction addCacheControlToLastConversationMessage(\n\tmessages: ChatCompletionMessageParam[],\n\tcacheControl: OpenAICompatCacheControl,\n): void {\n\tfor (let i = messages.length - 1; i >= 0; i--) {\n\t\tconst message = messages[i];\n\t\tif (message.role === \"user\" \|\| message.role === \"assistant\") {\n\t\t\tif (addCacheControlToMessage(message, cacheControl)) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t}\n\t}\n}\n\nfunction addCacheControlToLastTool(\n\ttools: OpenAI.Chat.Completions.ChatCompletionTool[] \| undefined,\n\tcacheControl: OpenAICompatCacheControl,\n): void {\n\tif (!tools \|\| tools.length === 0) {\n\t\treturn;\n\t}\n\n\tconst lastTool = tools[tools.length - 1] as ChatCompletionToolWithCacheControl;\n\tlastTool.cache_control = cacheControl;\n}\n\nfunction addCacheControlToInstructionMessage(\n\tmessage: ChatCompletionInstructionMessageParam,\n\tcacheControl: OpenAICompatCacheControl,\n): boolean {\n\treturn addCacheControlToTextContent(message, cacheControl);\n}\n\nfunction addCacheControlToMessage(\n\tmessage: ChatCompletionMessageParam,\n\tcacheControl: OpenAICompatCacheControl,\n): boolean {\n\tif (message.role === \"user\" \|\| message.role === \"assistant\") {\n\t\treturn addCacheControlToTextContent(message, cacheControl);\n\t}\n\treturn false;\n}\n\nfunction addCacheControlToTextContent(\n\tmessage:\n\t\t\| ChatCompletionInstructionMessageParam\n\t\t\| ChatCompletionAssistantMessageParam\n\t\t\| Extract<ChatCompletionMessageParam, { role: \"user\" }>,\n\tcacheControl: OpenAICompatCacheControl,\n): boolean {\n\tconst content = message.content;\n\tif (typeof content === \"string\") {\n\t\tif (content.length === 0) {\n\t\t\treturn false;\n\t\t}\n\t\tmessage.content = [\n\t\t\t{\n\t\t\t\ttype: \"text\",\n\t\t\t\ttext: content,\n\t\t\t\tcache_control: cacheControl,\n\t\t\t},\n\t\t] as ChatCompletionTextPartWithCacheControl[];\n\t\treturn true;\n\t}\n\n\tif (!Array.isArray(content)) {\n\t\treturn false;\n\t}\n\n\tfor (let i = content.length - 1; i >= 0; i--) {\n\t\tconst part = content[i];\n\t\tif (part?.type === \"text\") {\n\t\t\tconst textPart = part as ChatCompletionTextPartWithCacheControl;\n\t\t\ttextPart.cache_control = cacheControl;\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n\nexport function convertMessages(\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\tcompat: ResolvedOpenAICompletionsCompat,\n): ChatCompletionMessageParam[] {\n\tconst params: ChatCompletionMessageParam[] = [];\n\n\tconst normalizeToolCallId = (id: string): string => {\n\t\t// Handle pipe-separated IDs from OpenAI Responses API\n\t\t// Format: {call_id}\|{id} where {id} can be 400+ chars with special chars (+, /, =)\n\t\t// These come from providers like github-copilot, openai-codex, opencode\n\t\t// Extract just the call_id part and normalize it\n\t\tif (id.includes(\"\|\")) {\n\t\t\tconst [callId] = id.split(\"\|\");\n\t\t\t// Sanitize to allowed chars and truncate to 40 chars (OpenAI limit)\n\t\t\treturn callId.replace(/[^a-zA-Z0-9_-]/g, \"_\").slice(0, 40);\n\t\t}\n\n\t\tif (model.provider === \"openai\") return id.length > 40 ? id.slice(0, 40) : id;\n\t\treturn id;\n\t};\n\n\tconst transformedMessages = transformMessages(context.messages, model, (id) => normalizeToolCallId(id));\n\n\tif (context.systemPrompt) {\n\t\tconst useDeveloperRole = model.reasoning && compat.supportsDeveloperRole;\n\t\tconst role = useDeveloperRole ? \"developer\" : \"system\";\n\t\tparams.push({ role: role, content: sanitizeSurrogates(context.systemPrompt) });\n\t}\n\n\tlet lastRole: string \| null = null;\n\n\tfor (let i = 0; i < transformedMessages.length; i++) {\n\t\tconst msg = transformedMessages[i];\n\t\t// Some providers don't allow user messages directly after tool results\n\t\t// Insert a synthetic assistant message to bridge the gap\n\t\tif (compat.requiresAssistantAfterToolResult && lastRole === \"toolResult\" && msg.role === \"user\") {\n\t\t\tparams.push({\n\t\t\t\trole: \"assistant\",\n\t\t\t\tcontent: \"I have processed the tool results.\",\n\t\t\t});\n\t\t}\n\n\t\tif (msg.role === \"user\") {\n\t\t\tif (typeof msg.content === \"string\") {\n\t\t\t\tparams.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent: sanitizeSurrogates(msg.content),\n\t\t\t\t});\n\t\t\t} else {\n\t\t\t\tconst content: ChatCompletionContentPart[] = msg.content.map((item): ChatCompletionContentPart => {\n\t\t\t\t\tif (item.type === \"text\") {\n\t\t\t\t\t\treturn {\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: sanitizeSurrogates(item.text),\n\t\t\t\t\t\t} satisfies ChatCompletionContentPartText;\n\t\t\t\t\t} else {\n\t\t\t\t\t\treturn {\n\t\t\t\t\t\t\ttype: \"image_url\",\n\t\t\t\t\t\t\timage_url: {\n\t\t\t\t\t\t\t\turl: `data:${item.mimeType};base64,${item.data}`,\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t} satisfies ChatCompletionContentPartImage;\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\tif (content.length === 0) continue;\n\t\t\t\tparams.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent,\n\t\t\t\t});\n\t\t\t}\n\t\t} else if (msg.role === \"assistant\") {\n\t\t\t// Some providers don't accept null content, use empty string instead\n\t\t\tconst assistantMsg: ChatCompletionAssistantMessageParam = {\n\t\t\t\trole: \"assistant\",\n\t\t\t\tcontent: compat.requiresAssistantAfterToolResult ? \"\" : null,\n\t\t\t};\n\n\t\t\tconst assistantTextParts = msg.content\n\t\t\t\t.filter(isTextContentBlock)\n\t\t\t\t.filter((block) => block.text.trim().length > 0)\n\t\t\t\t.map(\n\t\t\t\t\t(block) =>\n\t\t\t\t\t\t({\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: sanitizeSurrogates(block.text),\n\t\t\t\t\t\t}) satisfies ChatCompletionContentPartText,\n\t\t\t\t);\n\t\t\tconst assistantText = assistantTextParts.map((part) => part.text).join(\"\");\n\n\t\t\tconst nonEmptyThinkingBlocks = msg.content\n\t\t\t\t.filter(isThinkingContentBlock)\n\t\t\t\t.filter((block) => block.thinking.trim().length > 0);\n\t\t\tif (nonEmptyThinkingBlocks.length > 0) {\n\t\t\t\tif (compat.requiresThinkingAsText) {\n\t\t\t\t\t// Convert thinking blocks to plain text (no tags to avoid model mimicking them)\n\t\t\t\t\tconst thinkingText = nonEmptyThinkingBlocks\n\t\t\t\t\t\t.map((block) => sanitizeSurrogates(block.thinking))\n\t\t\t\t\t\t.join(\"\\n\\n\");\n\t\t\t\t\tassistantMsg.content = [{ type: \"text\", text: thinkingText }, ...assistantTextParts];\n\t\t\t\t} else {\n\t\t\t\t\t// Always send assistant content as a plain string (OpenAI Chat Completions\n\t\t\t\t\t// API standard format). Sending as an array of {type:\"text\", text:\"...\"}\n\t\t\t\t\t// objects is non-standard and causes some models (e.g. DeepSeek V3.2 via\n\t\t\t\t\t// NVIDIA NIM) to mirror the content-block structure literally in their\n\t\t\t\t\t// output, producing recursive nesting like [{'type':'text','text':'[{...}]'}].\n\t\t\t\t\tif (assistantText.length > 0) {\n\t\t\t\t\t\tassistantMsg.content = assistantText;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Use the signature from the first thinking block if available (for llama.cpp server + gpt-oss)\n\t\t\t\t\tconst signature = nonEmptyThinkingBlocks[0].thinkingSignature;\n\t\t\t\t\tif (signature && signature.length > 0) {\n\t\t\t\t\t\t(assistantMsg as any)[signature] = nonEmptyThinkingBlocks.map((block) => block.thinking).join(\"\\n\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} else if (assistantText.length > 0) {\n\t\t\t\t// Always send assistant content as a plain string (OpenAI Chat Completions\n\t\t\t\t// API standard format). Sending as an array of {type:\"text\", text:\"...\"}\n\t\t\t\t// objects is non-standard and causes some models (e.g. DeepSeek V3.2 via\n\t\t\t\t// NVIDIA NIM) to mirror the content-block structure literally in their\n\t\t\t\t// output, producing recursive nesting like [{'type':'text','text':'[{...}]'}].\n\t\t\t\tassistantMsg.content = assistantText;\n\t\t\t}\n\n\t\t\tconst toolCalls = msg.content.filter(isToolCallBlock);\n\t\t\tif (toolCalls.length > 0) {\n\t\t\t\tassistantMsg.tool_calls = toolCalls.map((tc) => ({\n\t\t\t\t\tid: tc.id,\n\t\t\t\t\ttype: \"function\" as const,\n\t\t\t\t\tfunction: {\n\t\t\t\t\t\tname: tc.name,\n\t\t\t\t\t\targuments: JSON.stringify(tc.arguments),\n\t\t\t\t\t},\n\t\t\t\t}));\n\t\t\t\tconst reasoningDetails = toolCalls\n\t\t\t\t\t.filter((tc) => tc.thoughtSignature)\n\t\t\t\t\t.map((tc) => {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn JSON.parse(tc.thoughtSignature!);\n\t\t\t\t\t\t} catch {\n\t\t\t\t\t\t\treturn null;\n\t\t\t\t\t\t}\n\t\t\t\t\t})\n\t\t\t\t\t.filter(Boolean);\n\t\t\t\tif (reasoningDetails.length > 0) {\n\t\t\t\t\t(assistantMsg as any).reasoning_details = reasoningDetails;\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (\n\t\t\t\tcompat.requiresReasoningContentOnAssistantMessages &&\n\t\t\t\tmodel.reasoning &&\n\t\t\t\t(assistantMsg as { reasoning_content?: string }).reasoning_content === undefined\n\t\t\t) {\n\t\t\t\t(assistantMsg as { reasoning_content?: string }).reasoning_content = \"\";\n\t\t\t}\n\t\t\t// Skip assistant messages that have no content and no tool calls.\n\t\t\t// Some providers require \"either content or tool_calls, but not none\".\n\t\t\t// Other providers also don't accept empty assistant messages.\n\t\t\t// This handles aborted assistant responses that got no content.\n\t\t\tconst content = assistantMsg.content;\n\t\t\tconst hasContent =\n\t\t\t\tcontent !== null &&\n\t\t\t\tcontent !== undefined &&\n\t\t\t\t(typeof content === \"string\" ? content.length > 0 : content.length > 0);\n\t\t\tif (!hasContent && !assistantMsg.tool_calls) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tparams.push(assistantMsg);\n\t\t} else if (msg.role === \"toolResult\") {\n\t\t\tconst imageBlocks: Array<{ type: \"image_url\"; image_url: { url: string } }> = [];\n\t\t\tlet j = i;\n\n\t\t\tfor (; j < transformedMessages.length && transformedMessages[j].role === \"toolResult\"; j++) {\n\t\t\t\tconst toolMsg = transformedMessages[j] as ToolResultMessage;\n\n\t\t\t\t// Extract text and image content\n\t\t\t\tconst textResult = toolMsg.content\n\t\t\t\t\t.filter(isTextContentBlock)\n\t\t\t\t\t.map((block) => block.text)\n\t\t\t\t\t.join(\"\\n\");\n\t\t\t\tconst hasImages = toolMsg.content.some((c) => c.type === \"image\");\n\n\t\t\t\t// Always send tool result with text (or placeholder if only images)\n\t\t\t\tconst hasText = textResult.length > 0;\n\t\t\t\t// Some providers require the 'name' field in tool results\n\t\t\t\tconst toolResultMsg: ChatCompletionToolMessageParam = {\n\t\t\t\t\trole: \"tool\",\n\t\t\t\t\tcontent: sanitizeSurrogates(hasText ? textResult : \"(see attached image)\"),\n\t\t\t\t\ttool_call_id: toolMsg.toolCallId,\n\t\t\t\t};\n\t\t\t\tif (compat.requiresToolResultName && toolMsg.toolName) {\n\t\t\t\t\t(toolResultMsg as any).name = toolMsg.toolName;\n\t\t\t\t}\n\t\t\t\tparams.push(toolResultMsg);\n\n\t\t\t\tif (hasImages && model.input.includes(\"image\")) {\n\t\t\t\t\tfor (const block of toolMsg.content) {\n\t\t\t\t\t\tif (isImageContentBlock(block)) {\n\t\t\t\t\t\t\timageBlocks.push({\n\t\t\t\t\t\t\t\ttype: \"image_url\",\n\t\t\t\t\t\t\t\timage_url: {\n\t\t\t\t\t\t\t\t\turl: `data:${block.mimeType};base64,${block.data}`,\n\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\ti = j - 1;\n\n\t\t\tif (imageBlocks.length > 0) {\n\t\t\t\tif (compat.requiresAssistantAfterToolResult) {\n\t\t\t\t\tparams.push({\n\t\t\t\t\t\trole: \"assistant\",\n\t\t\t\t\t\tcontent: \"I have processed the tool results.\",\n\t\t\t\t\t});\n\t\t\t\t}\n\n\t\t\t\tparams.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: \"Attached image(s) from tool result:\",\n\t\t\t\t\t\t},\n\t\t\t\t\t\t...imageBlocks,\n\t\t\t\t\t],\n\t\t\t\t});\n\t\t\t\tlastRole = \"user\";\n\t\t\t} else {\n\t\t\t\tlastRole = \"toolResult\";\n\t\t\t}\n\t\t\tcontinue;\n\t\t}\n\n\t\tlastRole = msg.role;\n\t}\n\n\treturn params;\n}\n\nfunction convertTools(\n\ttools: Tool[],\n\tcompat: ResolvedOpenAICompletionsCompat,\n): OpenAI.Chat.Completions.ChatCompletionTool[] {\n\treturn tools.map((tool) => ({\n\t\ttype: \"function\",\n\t\tfunction: {\n\t\t\tname: tool.name,\n\t\t\tdescription: tool.description,\n\t\t\tparameters: tool.parameters as any, // TypeBox already generates JSON Schema\n\t\t\t// Only include strict if provider supports it. Some reject unknown fields.\n\t\t\t...(compat.supportsStrictMode !== false && { strict: false }),\n\t\t},\n\t}));\n}\n\nfunction parseChunkUsage(\n\trawUsage: {\n\t\tprompt_tokens?: number;\n\t\tcompletion_tokens?: number;\n\t\tprompt_tokens_details?: { cached_tokens?: number; cache_write_tokens?: number };\n\t},\n\tmodel: Model<\"openai-completions\">,\n): AssistantMessage[\"usage\"] {\n\tconst promptTokens = rawUsage.prompt_tokens \|\| 0;\n\tconst reportedCachedTokens = rawUsage.prompt_tokens_details?.cached_tokens \|\| 0;\n\tconst cacheWriteTokens = rawUsage.prompt_tokens_details?.cache_write_tokens \|\| 0;\n\n\t// Normalize to pi-ai semantics:\n\t// - cacheRead: hits from cache created by previous requests only\n\t// - cacheWrite: tokens written to cache in this request\n\t// Some OpenAI-compatible providers (observed on OpenRouter) report cached_tokens\n\t// as (previous hits + current writes). In that case, remove cacheWrite from cacheRead.\n\tconst cacheReadTokens =\n\t\tcacheWriteTokens > 0 ? Math.max(0, reportedCachedTokens - cacheWriteTokens) : reportedCachedTokens;\n\n\tconst input = Math.max(0, promptTokens - cacheReadTokens - cacheWriteTokens);\n\t// OpenAI completion_tokens already includes reasoning_tokens.\n\tconst outputTokens = rawUsage.completion_tokens \|\| 0;\n\tconst usage: AssistantMessage[\"usage\"] = {\n\t\tinput,\n\t\toutput: outputTokens,\n\t\tcacheRead: cacheReadTokens,\n\t\tcacheWrite: cacheWriteTokens,\n\t\ttotalTokens: input + outputTokens + cacheReadTokens + cacheWriteTokens,\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t};\n\tcalculateCost(model, usage);\n\treturn usage;\n}\n\nfunction mapStopReason(reason: ChatCompletionChunk.Choice[\"finish_reason\"] \| string): {\n\tstopReason: StopReason;\n\terrorMessage?: string;\n} {\n\tif (reason === null) return { stopReason: \"stop\" };\n\tswitch (reason) {\n\t\tcase \"stop\":\n\t\tcase \"end\":\n\t\t\treturn { stopReason: \"stop\" };\n\t\tcase \"length\":\n\t\t\treturn { stopReason: \"length\" };\n\t\tcase \"function_call\":\n\t\tcase \"tool_calls\":\n\t\t\treturn { stopReason: \"toolUse\" };\n\t\tcase \"content_filter\":\n\t\t\treturn { stopReason: \"error\", errorMessage: \"Provider finish_reason: content_filter\" };\n\t\tcase \"network_error\":\n\t\t\treturn { stopReason: \"error\", errorMessage: \"Provider finish_reason: network_error\" };\n\t\tdefault:\n\t\t\treturn {\n\t\t\t\tstopReason: \"error\",\n\t\t\t\terrorMessage: `Provider finish_reason: ${reason}`,\n\t\t\t};\n\t}\n}\n\n/\n Detect compatibility settings from provider and baseUrl for known providers.\n * Provider takes precedence over URL-based detection since it's explicitly configured.\n * Returns a fully resolved OpenAICompletionsCompat object with all fields set.\n /\nfunction detectCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst provider = model.provider;\n\tconst baseUrl = model.baseUrl;\n\n\tconst isZai = provider === \"zai\" \|\| baseUrl.includes(\"api.z.ai\");\n\n\tconst isNonStandard =\n\t\tprovider === \"cerebras\" \|\|\n\t\tbaseUrl.includes(\"cerebras.ai\") \|\|\n\t\tprovider === \"xai\" \|\|\n\t\tbaseUrl.includes(\"api.x.ai\") \|\|\n\t\tbaseUrl.includes(\"chutes.ai\") \|\|\n\t\tbaseUrl.includes(\"deepseek.com\") \|\|\n\t\tisZai \|\|\n\t\tprovider === \"opencode\" \|\|\n\t\tbaseUrl.includes(\"opencode.ai\");\n\n\tconst useMaxTokens = baseUrl.includes(\"chutes.ai\");\n\n\tconst isGrok = provider === \"xai\" \|\| baseUrl.includes(\"api.x.ai\");\n\tconst isGroq = provider === \"groq\" \|\| baseUrl.includes(\"groq.com\");\n\tconst isDeepSeek = provider === \"deepseek\" \|\| baseUrl.includes(\"deepseek.com\");\n\tconst cacheControlFormat = provider === \"openrouter\" && model.id.startsWith(\"anthropic/\") ? \"anthropic\" : undefined;\n\n\tconst reasoningEffortMap = isDeepSeek\n\t\t? {\n\t\t\t\tminimal: \"high\",\n\t\t\t\tlow: \"high\",\n\t\t\t\tmedium: \"high\",\n\t\t\t\thigh: \"high\",\n\t\t\t\txhigh: \"max\",\n\t\t\t}\n\t\t: isGroq && model.id === \"qwen/qwen3-32b\"\n\t\t\t? {\n\t\t\t\t\tminimal: \"default\",\n\t\t\t\t\tlow: \"default\",\n\t\t\t\t\tmedium: \"default\",\n\t\t\t\t\thigh: \"default\",\n\t\t\t\t\txhigh: \"default\",\n\t\t\t\t}\n\t\t\t: {};\n\treturn {\n\t\tsupportsStore: !isNonStandard,\n\t\tsupportsDeveloperRole: !isNonStandard,\n\t\tsupportsReasoningEffort: !isGrok && !isZai,\n\t\treasoningEffortMap,\n\t\tsupportsUsageInStreaming: true,\n\t\tmaxTokensField: useMaxTokens ? \"max_tokens\" : \"max_completion_tokens\",\n\t\trequiresToolResultName: false,\n\t\trequiresAssistantAfterToolResult: false,\n\t\trequiresThinkingAsText: false,\n\t\trequiresReasoningContentOnAssistantMessages: isDeepSeek,\n\t\tthinkingFormat: isDeepSeek\n\t\t\t? \"deepseek\"\n\t\t\t: isZai\n\t\t\t\t? \"zai\"\n\t\t\t\t: provider === \"openrouter\" \|\| baseUrl.includes(\"openrouter.ai\")\n\t\t\t\t\t? \"openrouter\"\n\t\t\t\t\t: \"openai\",\n\t\topenRouterRouting: {},\n\t\tvercelGatewayRouting: {},\n\t\tzaiToolStream: false,\n\t\tsupportsStrictMode: true,\n\t\tcacheControlFormat,\n\t\tsendSessionAffinityHeaders: false,\n\t\tsupportsLongCacheRetention: true,\n\t};\n}\n\n/\n Get resolved compatibility settings for a model.\n * Uses explicit model.compat if provided, otherwise auto-detects from provider/URL.\n */\nfunction getCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst detected = detectCompat(model);\n\tif (!model.compat) return detected;\n\n\treturn {\n\t\tsupportsStore: model.compat.supportsStore ?? detected.supportsStore,\n\t\tsupportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,\n\t\tsupportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,\n\t\treasoningEffortMap: model.compat.reasoningEffortMap ?? detected.reasoningEffortMap,\n\t\tsupportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? detected.supportsUsageInStreaming,\n\t\tmaxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,\n\t\trequiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,\n\t\trequiresAssistantAfterToolResult:\n\t\t\tmodel.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,\n\t\trequiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,\n\t\trequiresReasoningContentOnAssistantMessages:\n\t\t\tmodel.compat.requiresReasoningContentOnAssistantMessages ??\n\t\t\tdetected.requiresReasoningContentOnAssistantMessages,\n\t\tthinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,\n\t\topenRouterRouting: model.compat.openRouterRouting ?? {},\n\t\tvercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,\n\t\tzaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,\n\t\tsupportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,\n\t\tcacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,\n\t\tsendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,\n\t\tsupportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,\n\t};\n}\n"]}
1	+ {"version":3,"file":"openai-completions.d.ts","sourceRoot":"","sources":["../../src/providers/openai-completions.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAOX,0BAA0B,EAG1B,MAAM,sCAAsC,CAAC;AAG9C,OAAO,KAAK,EAGX,OAAO,EAGP,KAAK,EACL,uBAAuB,EACvB,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAMb,MAAM,aAAa,CAAC;AA4CrB,MAAM,WAAW,wBAAyB,SAAQ,aAAa;IAC9D,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,eAAe,CAAC,EAAE,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;CAClE;AAOD,KAAK,+BAA+B,GAAG,IAAI,CAAC,QAAQ,CAAC,uBAAuB,CAAC,EAAE,oBAAoB,CAAC,GAAG;IACtG,kBAAkB,CAAC,EAAE,uBAAuB,CAAC,oBAAoB,CAAC,CAAC;CACnE,CAAC;AAsBF,eAAO,MAAM,uBAAuB,EAAE,cAAc,CAAC,oBAAoB,EAAE,wBAAwB,CA8RlG,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,cAAc,CAAC,oBAAoB,EAAE,mBAAmB,CAmBnG,CAAC;AAsRF,wBAAgB,eAAe,CAC9B,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,EAClC,OAAO,EAAE,OAAO,EAChB,MAAM,EAAE,+BAA+B,GACrC,0BAA0B,EAAE,CA6O9B","sourcesContent":["import OpenAI from \"openai\";\nimport type {\n\tChatCompletionAssistantMessageParam,\n\tChatCompletionChunk,\n\tChatCompletionContentPart,\n\tChatCompletionContentPartImage,\n\tChatCompletionContentPartText,\n\tChatCompletionDeveloperMessageParam,\n\tChatCompletionMessageParam,\n\tChatCompletionSystemMessageParam,\n\tChatCompletionToolMessageParam,\n} from \"openai/resources/chat/completions.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost, supportsXhigh } from \"../models.js\";\nimport type {\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tImageContent,\n\tMessage,\n\tModel,\n\tOpenAICompletionsCompat,\n\tSimpleStreamOptions,\n\tStopReason,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingContent,\n\tTool,\n\tToolCall,\n\tToolResultMessage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { headersToRecord } from \"../utils/headers.js\";\nimport { parseStreamingJson } from \"../utils/json-parse.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\nimport { transformMessages } from \"./transform-messages.js\";\n\n/*\n Check if conversation messages contain tool calls or tool results.\n * This is needed because Anthropic (via proxy) requires the tools param\n * to be present when messages include tool_calls or tool role messages.\n /\nfunction hasToolHistory(messages: Message[]): boolean {\n\tfor (const msg of messages) {\n\t\tif (msg.role === \"toolResult\") {\n\t\t\treturn true;\n\t\t}\n\t\tif (msg.role === \"assistant\") {\n\t\t\tif (msg.content.some((block) => block.type === \"toolCall\")) {\n\t\t\t\treturn true;\n\t\t\t}\n\t\t}\n\t}\n\treturn false;\n}\n\nfunction isTextContentBlock(block: { type: string }): block is TextContent {\n\treturn block.type === \"text\";\n}\n\nfunction isThinkingContentBlock(block: { type: string }): block is ThinkingContent {\n\treturn block.type === \"thinking\";\n}\n\nfunction isToolCallBlock(block: { type: string }): block is ToolCall {\n\treturn block.type === \"toolCall\";\n}\n\nfunction isImageContentBlock(block: { type: string }): block is ImageContent {\n\treturn block.type === \"image\";\n}\n\nexport interface OpenAICompletionsOptions extends StreamOptions {\n\ttoolChoice?: \"auto\" \| \"none\" \| \"required\" \| { type: \"function\"; function: { name: string } };\n\treasoningEffort?: \"minimal\" \| \"low\" \| \"medium\" \| \"high\" \| \"xhigh\";\n}\n\ninterface OpenAICompatCacheControl {\n\ttype: \"ephemeral\";\n\tttl?: string;\n}\n\ntype ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, \"cacheControlFormat\"> & {\n\tcacheControlFormat?: OpenAICompletionsCompat[\"cacheControlFormat\"];\n};\n\ntype ChatCompletionInstructionMessageParam = ChatCompletionDeveloperMessageParam \| ChatCompletionSystemMessageParam;\n\ntype ChatCompletionTextPartWithCacheControl = ChatCompletionContentPartText & {\n\tcache_control?: OpenAICompatCacheControl;\n};\n\ntype ChatCompletionToolWithCacheControl = OpenAI.Chat.Completions.ChatCompletionTool & {\n\tcache_control?: OpenAICompatCacheControl;\n};\n\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.PI_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nexport const streamOpenAICompletions: StreamFunction<\"openai-completions\", OpenAICompletionsOptions> = (\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\toptions?: OpenAICompletionsOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider) \|\| \"\";\n\t\t\tconst compat = getCompat(model);\n\t\t\tconst cacheRetention = resolveCacheRetention(options?.cacheRetention);\n\t\t\tconst cacheSessionId = cacheRetention === \"none\" ? undefined : options?.sessionId;\n\t\t\tconst client = createClient(model, context, apiKey, options?.headers, cacheSessionId, compat);\n\t\t\tlet params = buildParams(model, context, options, compat, cacheRetention);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming;\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\t...(options?.maxRetries !== undefined ? { maxRetries: options.maxRetries } : {}),\n\t\t\t};\n\t\t\tconst { data: openaiStream, response } = await client.chat.completions\n\t\t\t\t.create(params, requestOptions)\n\t\t\t\t.withResponse();\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\tinterface StreamingToolCallBlock extends ToolCall {\n\t\t\t\tpartialArgs?: string;\n\t\t\t\tstreamIndex?: number;\n\t\t\t}\n\n\t\t\tlet currentBlock: TextContent \| ThinkingContent \| StreamingToolCallBlock \| null = null;\n\t\t\tconst blocks = output.content;\n\t\t\tconst getContentIndex = (block: typeof currentBlock) => (block ? blocks.indexOf(block) : -1);\n\t\t\tconst currentContentIndex = () => getContentIndex(currentBlock);\n\t\t\tconst finishCurrentBlock = (block?: typeof currentBlock) => {\n\t\t\t\tif (block) {\n\t\t\t\t\tconst contentIndex = getContentIndex(block);\n\t\t\t\t\tif (contentIndex === -1) {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tif (block.type === \"text\") {\n\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\tcontentIndex,\n\t\t\t\t\t\t\tcontent: block.text,\n\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t});\n\t\t\t\t\t} else if (block.type === \"thinking\") {\n\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\tcontentIndex,\n\t\t\t\t\t\t\tcontent: block.thinking,\n\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t});\n\t\t\t\t\t} else if (block.type === \"toolCall\") {\n\t\t\t\t\t\tblock.arguments = parseStreamingJson(block.partialArgs);\n\t\t\t\t\t\t// Finalize in-place and strip the scratch buffers so replay only\n\t\t\t\t\t\t// carries parsed arguments.\n\t\t\t\t\t\tdelete block.partialArgs;\n\t\t\t\t\t\tdelete block.streamIndex;\n\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\ttype: \"toolcall_end\",\n\t\t\t\t\t\t\tcontentIndex,\n\t\t\t\t\t\t\ttoolCall: block,\n\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t};\n\n\t\t\tfor await (const chunk of openaiStream) {\n\t\t\t\tif (!chunk \|\| typeof chunk !== \"object\") continue;\n\n\t\t\t\t// OpenAI documents ChatCompletionChunk.id as the unique chat completion identifier,\n\t\t\t\t// and each chunk in a streamed completion carries the same id.\n\t\t\t\toutput.responseId \|\|= chunk.id;\n\t\t\t\tif (chunk.usage) {\n\t\t\t\t\toutput.usage = parseChunkUsage(chunk.usage, model);\n\t\t\t\t}\n\n\t\t\t\tconst choice = Array.isArray(chunk.choices) ? chunk.choices[0] : undefined;\n\t\t\t\tif (!choice) continue;\n\n\t\t\t\t// Fallback: some providers (e.g., Moonshot) return usage\n\t\t\t\t// in choice.usage instead of the standard chunk.usage\n\t\t\t\tif (!chunk.usage && (choice as any).usage) {\n\t\t\t\t\toutput.usage = parseChunkUsage((choice as any).usage, model);\n\t\t\t\t}\n\n\t\t\t\tif (choice.finish_reason) {\n\t\t\t\t\tconst finishReasonResult = mapStopReason(choice.finish_reason);\n\t\t\t\t\toutput.stopReason = finishReasonResult.stopReason;\n\t\t\t\t\tif (finishReasonResult.errorMessage) {\n\t\t\t\t\t\toutput.errorMessage = finishReasonResult.errorMessage;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tif (choice.delta) {\n\t\t\t\t\tif (\n\t\t\t\t\t\tchoice.delta.content !== null &&\n\t\t\t\t\t\tchoice.delta.content !== undefined &&\n\t\t\t\t\t\tchoice.delta.content.length > 0\n\t\t\t\t\t) {\n\t\t\t\t\t\tif (!currentBlock \|\| currentBlock.type !== \"text\") {\n\t\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: currentContentIndex(), partial: output });\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\tcurrentBlock.text += choice.delta.content;\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: currentContentIndex(),\n\t\t\t\t\t\t\t\tdelta: choice.delta.content,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\t// Some endpoints return reasoning in reasoning_content (llama.cpp),\n\t\t\t\t\t// or reasoning (other openai compatible endpoints)\n\t\t\t\t\t// Use the first non-empty reasoning field to avoid duplication\n\t\t\t\t\t// (e.g., chutes.ai returns both reasoning_content and reasoning with same content)\n\t\t\t\t\tconst reasoningFields = [\"reasoning_content\", \"reasoning\", \"reasoning_text\"];\n\t\t\t\t\tlet foundReasoningField: string \| null = null;\n\t\t\t\t\tfor (const field of reasoningFields) {\n\t\t\t\t\t\tif (\n\t\t\t\t\t\t\t(choice.delta as any)[field] !== null &&\n\t\t\t\t\t\t\t(choice.delta as any)[field] !== undefined &&\n\t\t\t\t\t\t\t(choice.delta as any)[field].length > 0\n\t\t\t\t\t\t) {\n\t\t\t\t\t\t\tif (!foundReasoningField) {\n\t\t\t\t\t\t\t\tfoundReasoningField = field;\n\t\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\tif (foundReasoningField) {\n\t\t\t\t\t\tif (!currentBlock \|\| currentBlock.type !== \"thinking\") {\n\t\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\t\tcurrentBlock = {\n\t\t\t\t\t\t\t\ttype: \"thinking\",\n\t\t\t\t\t\t\t\tthinking: \"\",\n\t\t\t\t\t\t\t\tthinkingSignature: foundReasoningField,\n\t\t\t\t\t\t\t};\n\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: currentContentIndex(), partial: output });\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tif (currentBlock.type === \"thinking\") {\n\t\t\t\t\t\t\tconst delta = (choice.delta as any)[foundReasoningField];\n\t\t\t\t\t\t\tcurrentBlock.thinking += delta;\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: currentContentIndex(),\n\t\t\t\t\t\t\t\tdelta,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\tif (choice?.delta?.tool_calls) {\n\t\t\t\t\t\tfor (const toolCall of choice.delta.tool_calls) {\n\t\t\t\t\t\t\tconst streamIndex = typeof toolCall.index === \"number\" ? toolCall.index : undefined;\n\t\t\t\t\t\t\tconst sameToolCall =\n\t\t\t\t\t\t\t\tcurrentBlock?.type === \"toolCall\" &&\n\t\t\t\t\t\t\t\t((streamIndex !== undefined && currentBlock.streamIndex === streamIndex) \|\|\n\t\t\t\t\t\t\t\t\t(streamIndex === undefined && toolCall.id && currentBlock.id === toolCall.id));\n\n\t\t\t\t\t\t\tif (!sameToolCall) {\n\t\t\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\t\t\tcurrentBlock = {\n\t\t\t\t\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\t\t\t\t\tid: toolCall.id \|\| \"\",\n\t\t\t\t\t\t\t\t\tname: toolCall.function?.name \|\| \"\",\n\t\t\t\t\t\t\t\t\targuments: {},\n\t\t\t\t\t\t\t\t\tpartialArgs: \"\",\n\t\t\t\t\t\t\t\t\tstreamIndex,\n\t\t\t\t\t\t\t\t};\n\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"toolcall_start\",\n\t\t\t\t\t\t\t\t\tcontentIndex: getContentIndex(currentBlock),\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\tconst currentToolCallBlock = currentBlock?.type === \"toolCall\" ? currentBlock : null;\n\t\t\t\t\t\t\tif (currentToolCallBlock) {\n\t\t\t\t\t\t\t\tif (!currentToolCallBlock.id && toolCall.id) currentToolCallBlock.id = toolCall.id;\n\t\t\t\t\t\t\t\tif (!currentToolCallBlock.name && toolCall.function?.name) {\n\t\t\t\t\t\t\t\t\tcurrentToolCallBlock.name = toolCall.function.name;\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tif (currentToolCallBlock.streamIndex === undefined && streamIndex !== undefined) {\n\t\t\t\t\t\t\t\t\tcurrentToolCallBlock.streamIndex = streamIndex;\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tlet delta = \"\";\n\t\t\t\t\t\t\t\tif (toolCall.function?.arguments) {\n\t\t\t\t\t\t\t\t\tdelta = toolCall.function.arguments;\n\t\t\t\t\t\t\t\t\tcurrentToolCallBlock.partialArgs += toolCall.function.arguments;\n\t\t\t\t\t\t\t\t\tcurrentToolCallBlock.arguments = parseStreamingJson(currentToolCallBlock.partialArgs);\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\t\t\t\t\t\tcontentIndex: getContentIndex(currentToolCallBlock),\n\t\t\t\t\t\t\t\t\tdelta,\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\tconst reasoningDetails = (choice.delta as any).reasoning_details;\n\t\t\t\t\tif (reasoningDetails && Array.isArray(reasoningDetails)) {\n\t\t\t\t\t\tfor (const detail of reasoningDetails) {\n\t\t\t\t\t\t\tif (detail.type === \"reasoning.encrypted\" && detail.id && detail.data) {\n\t\t\t\t\t\t\t\tconst matchingToolCall = output.content.find(\n\t\t\t\t\t\t\t\t\t(b) => b.type === \"toolCall\" && b.id === detail.id,\n\t\t\t\t\t\t\t\t) as ToolCall \| undefined;\n\t\t\t\t\t\t\t\tif (matchingToolCall) {\n\t\t\t\t\t\t\t\t\tmatchingToolCall.thoughtSignature = JSON.stringify(detail);\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\") {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\t\t\tif (output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(output.errorMessage \|\| \"Provider returned an error stop reason\");\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 { partialArgs?: string }).partialArgs;\n\t\t\t\tdelete (block as { streamIndex?: number }).streamIndex;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\t// Some providers via OpenRouter give additional information in this field.\n\t\t\tconst rawMetadata = (error as any)?.error?.metadata?.raw;\n\t\t\tif (rawMetadata) output.errorMessage += `\\n${rawMetadata}`;\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 streamSimpleOpenAICompletions: StreamFunction<\"openai-completions\", SimpleStreamOptions> = (\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey \|\| getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);\n\tconst toolChoice = (options as OpenAICompletionsOptions \| undefined)?.toolChoice;\n\n\treturn streamOpenAICompletions(model, context, {\n\t\t...base,\n\t\treasoningEffort,\n\t\ttoolChoice,\n\t} satisfies OpenAICompletionsOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\tapiKey?: string,\n\toptionsHeaders?: Record<string, string>,\n\tsessionId?: string,\n\tcompat: ResolvedOpenAICompletionsCompat = getCompat(model),\n) {\n\tif (!apiKey) {\n\t\tif (!process.env.OPENAI_API_KEY) {\n\t\t\tthrow new Error(\n\t\t\t\t\"OpenAI API key is required. Set OPENAI_API_KEY environment variable or pass it as an argument.\",\n\t\t\t);\n\t\t}\n\t\tapiKey = process.env.OPENAI_API_KEY;\n\t}\n\n\tconst headers = { ...model.headers };\n\tif (model.provider === \"github-copilot\") {\n\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\tconst copilotHeaders = buildCopilotDynamicHeaders({\n\t\t\tmessages: context.messages,\n\t\t\thasImages,\n\t\t});\n\t\tObject.assign(headers, copilotHeaders);\n\t}\n\n\tif (sessionId && compat.sendSessionAffinityHeaders) {\n\t\theaders.session_id = sessionId;\n\t\theaders[\"x-client-request-id\"] = sessionId;\n\t\theaders[\"x-session-affinity\"] = sessionId;\n\t}\n\n\t// Merge options headers last so they can override defaults\n\tif (optionsHeaders) {\n\t\tObject.assign(headers, optionsHeaders);\n\t}\n\n\treturn new OpenAI({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: headers,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\toptions?: OpenAICompletionsOptions,\n\tcompat: ResolvedOpenAICompletionsCompat = getCompat(model),\n\tcacheRetention: CacheRetention = resolveCacheRetention(options?.cacheRetention),\n) {\n\tconst messages = convertMessages(model, context, compat);\n\tconst cacheControl = getCompatCacheControl(compat, cacheRetention);\n\n\tconst params: OpenAI.Chat.Completions.ChatCompletionCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tmessages,\n\t\tstream: true,\n\t\tprompt_cache_key:\n\t\t\t(model.baseUrl.includes(\"api.openai.com\") && cacheRetention !== \"none\") \|\|\n\t\t\t(cacheRetention === \"long\" && compat.supportsLongCacheRetention)\n\t\t\t\t? options?.sessionId\n\t\t\t\t: undefined,\n\t\tprompt_cache_retention: cacheRetention === \"long\" && compat.supportsLongCacheRetention ? \"24h\" : undefined,\n\t};\n\n\tif (compat.supportsUsageInStreaming !== false) {\n\t\t(params as any).stream_options = { include_usage: true };\n\t}\n\n\tif (compat.supportsStore) {\n\t\tparams.store = false;\n\t}\n\n\tif (options?.maxTokens) {\n\t\tif (compat.maxTokensField === \"max_tokens\") {\n\t\t\t(params as any).max_tokens = options.maxTokens;\n\t\t} else {\n\t\t\tparams.max_completion_tokens = options.maxTokens;\n\t\t}\n\t}\n\n\tif (options?.temperature !== undefined) {\n\t\tparams.temperature = options.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertTools(context.tools, compat);\n\t\tif (compat.zaiToolStream) {\n\t\t\t(params as any).tool_stream = true;\n\t\t}\n\t} else if (hasToolHistory(context.messages)) {\n\t\t// Anthropic (via LiteLLM/proxy) requires tools param when conversation has tool_calls/tool_results\n\t\tparams.tools = [];\n\t}\n\n\tif (cacheControl) {\n\t\tapplyAnthropicCacheControl(messages, params.tools, cacheControl);\n\t}\n\n\tif (options?.toolChoice) {\n\t\tparams.tool_choice = options.toolChoice;\n\t}\n\n\tif (compat.thinkingFormat === \"zai\" && model.reasoning) {\n\t\t(params as any).enable_thinking = !!options?.reasoningEffort;\n\t} else if (compat.thinkingFormat === \"qwen\" && model.reasoning) {\n\t\t(params as any).enable_thinking = !!options?.reasoningEffort;\n\t} else if (compat.thinkingFormat === \"qwen-chat-template\" && model.reasoning) {\n\t\t(params as any).chat_template_kwargs = {\n\t\t\tenable_thinking: !!options?.reasoningEffort,\n\t\t\tpreserve_thinking: true,\n\t\t};\n\t} else if (compat.thinkingFormat === \"deepseek\" && model.reasoning) {\n\t\t(params as any).thinking = { type: options?.reasoningEffort ? \"enabled\" : \"disabled\" };\n\t\tif (options?.reasoningEffort) {\n\t\t\t(params as any).reasoning_effort = mapReasoningEffort(options.reasoningEffort, compat.reasoningEffortMap);\n\t\t}\n\t} else if (compat.thinkingFormat === \"openrouter\" && model.reasoning) {\n\t\t// OpenRouter normalizes reasoning across providers via a nested reasoning object.\n\t\tconst openRouterParams = params as typeof params & { reasoning?: { effort?: string } };\n\t\tif (options?.reasoningEffort) {\n\t\t\topenRouterParams.reasoning = {\n\t\t\t\teffort: mapReasoningEffort(options.reasoningEffort, compat.reasoningEffortMap),\n\t\t\t};\n\t\t} else {\n\t\t\topenRouterParams.reasoning = { effort: \"none\" };\n\t\t}\n\t} else if (options?.reasoningEffort && model.reasoning && compat.supportsReasoningEffort) {\n\t\t// OpenAI-style reasoning_effort\n\t\t(params as any).reasoning_effort = mapReasoningEffort(options.reasoningEffort, compat.reasoningEffortMap);\n\t}\n\n\t// OpenRouter provider routing preferences\n\tif (model.baseUrl.includes(\"openrouter.ai\") && model.compat?.openRouterRouting) {\n\t\t(params as any).provider = model.compat.openRouterRouting;\n\t}\n\n\t// Vercel AI Gateway provider routing preferences\n\tif (model.baseUrl.includes(\"ai-gateway.vercel.sh\") && model.compat?.vercelGatewayRouting) {\n\t\tconst routing = model.compat.vercelGatewayRouting;\n\t\tif (routing.only \|\| routing.order) {\n\t\t\tconst gatewayOptions: Record<string, string[]> = {};\n\t\t\tif (routing.only) gatewayOptions.only = routing.only;\n\t\t\tif (routing.order) gatewayOptions.order = routing.order;\n\t\t\t(params as any).providerOptions = { gateway: gatewayOptions };\n\t\t}\n\t}\n\n\treturn params;\n}\n\nfunction mapReasoningEffort(\n\teffort: NonNullable<OpenAICompletionsOptions[\"reasoningEffort\"]>,\n\treasoningEffortMap: Partial<Record<NonNullable<OpenAICompletionsOptions[\"reasoningEffort\"]>, string>>,\n): string {\n\treturn reasoningEffortMap[effort] ?? effort;\n}\n\nfunction getCompatCacheControl(\n\tcompat: ResolvedOpenAICompletionsCompat,\n\tcacheRetention: CacheRetention,\n): OpenAICompatCacheControl \| undefined {\n\tif (compat.cacheControlFormat !== \"anthropic\" \|\| cacheRetention === \"none\") {\n\t\treturn undefined;\n\t}\n\n\tconst ttl = cacheRetention === \"long\" && compat.supportsLongCacheRetention ? \"1h\" : undefined;\n\treturn { type: \"ephemeral\", ...(ttl ? { ttl } : {}) };\n}\n\nfunction applyAnthropicCacheControl(\n\tmessages: ChatCompletionMessageParam[],\n\ttools: OpenAI.Chat.Completions.ChatCompletionTool[] \| undefined,\n\tcacheControl: OpenAICompatCacheControl,\n): void {\n\taddCacheControlToSystemPrompt(messages, cacheControl);\n\taddCacheControlToLastTool(tools, cacheControl);\n\taddCacheControlToLastConversationMessage(messages, cacheControl);\n}\n\nfunction addCacheControlToSystemPrompt(\n\tmessages: ChatCompletionMessageParam[],\n\tcacheControl: OpenAICompatCacheControl,\n): void {\n\tfor (const message of messages) {\n\t\tif (message.role === \"system\" \|\| message.role === \"developer\") {\n\t\t\taddCacheControlToInstructionMessage(message, cacheControl);\n\t\t\treturn;\n\t\t}\n\t}\n}\n\nfunction addCacheControlToLastConversationMessage(\n\tmessages: ChatCompletionMessageParam[],\n\tcacheControl: OpenAICompatCacheControl,\n): void {\n\tfor (let i = messages.length - 1; i >= 0; i--) {\n\t\tconst message = messages[i];\n\t\tif (message.role === \"user\" \|\| message.role === \"assistant\") {\n\t\t\tif (addCacheControlToMessage(message, cacheControl)) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t}\n\t}\n}\n\nfunction addCacheControlToLastTool(\n\ttools: OpenAI.Chat.Completions.ChatCompletionTool[] \| undefined,\n\tcacheControl: OpenAICompatCacheControl,\n): void {\n\tif (!tools \|\| tools.length === 0) {\n\t\treturn;\n\t}\n\n\tconst lastTool = tools[tools.length - 1] as ChatCompletionToolWithCacheControl;\n\tlastTool.cache_control = cacheControl;\n}\n\nfunction addCacheControlToInstructionMessage(\n\tmessage: ChatCompletionInstructionMessageParam,\n\tcacheControl: OpenAICompatCacheControl,\n): boolean {\n\treturn addCacheControlToTextContent(message, cacheControl);\n}\n\nfunction addCacheControlToMessage(\n\tmessage: ChatCompletionMessageParam,\n\tcacheControl: OpenAICompatCacheControl,\n): boolean {\n\tif (message.role === \"user\" \|\| message.role === \"assistant\") {\n\t\treturn addCacheControlToTextContent(message, cacheControl);\n\t}\n\treturn false;\n}\n\nfunction addCacheControlToTextContent(\n\tmessage:\n\t\t\| ChatCompletionInstructionMessageParam\n\t\t\| ChatCompletionAssistantMessageParam\n\t\t\| Extract<ChatCompletionMessageParam, { role: \"user\" }>,\n\tcacheControl: OpenAICompatCacheControl,\n): boolean {\n\tconst content = message.content;\n\tif (typeof content === \"string\") {\n\t\tif (content.length === 0) {\n\t\t\treturn false;\n\t\t}\n\t\tmessage.content = [\n\t\t\t{\n\t\t\t\ttype: \"text\",\n\t\t\t\ttext: content,\n\t\t\t\tcache_control: cacheControl,\n\t\t\t},\n\t\t] as ChatCompletionTextPartWithCacheControl[];\n\t\treturn true;\n\t}\n\n\tif (!Array.isArray(content)) {\n\t\treturn false;\n\t}\n\n\tfor (let i = content.length - 1; i >= 0; i--) {\n\t\tconst part = content[i];\n\t\tif (part?.type === \"text\") {\n\t\t\tconst textPart = part as ChatCompletionTextPartWithCacheControl;\n\t\t\ttextPart.cache_control = cacheControl;\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n\nexport function convertMessages(\n\tmodel: Model<\"openai-completions\">,\n\tcontext: Context,\n\tcompat: ResolvedOpenAICompletionsCompat,\n): ChatCompletionMessageParam[] {\n\tconst params: ChatCompletionMessageParam[] = [];\n\n\tconst normalizeToolCallId = (id: string): string => {\n\t\t// Handle pipe-separated IDs from OpenAI Responses API\n\t\t// Format: {call_id}\|{id} where {id} can be 400+ chars with special chars (+, /, =)\n\t\t// These come from providers like github-copilot, openai-codex, opencode\n\t\t// Extract just the call_id part and normalize it\n\t\tif (id.includes(\"\|\")) {\n\t\t\tconst [callId] = id.split(\"\|\");\n\t\t\t// Sanitize to allowed chars and truncate to 40 chars (OpenAI limit)\n\t\t\treturn callId.replace(/[^a-zA-Z0-9_-]/g, \"_\").slice(0, 40);\n\t\t}\n\n\t\tif (model.provider === \"openai\") return id.length > 40 ? id.slice(0, 40) : id;\n\t\treturn id;\n\t};\n\n\tconst transformedMessages = transformMessages(context.messages, model, (id) => normalizeToolCallId(id));\n\n\tif (context.systemPrompt) {\n\t\tconst useDeveloperRole = model.reasoning && compat.supportsDeveloperRole;\n\t\tconst role = useDeveloperRole ? \"developer\" : \"system\";\n\t\tparams.push({ role: role, content: sanitizeSurrogates(context.systemPrompt) });\n\t}\n\n\tlet lastRole: string \| null = null;\n\n\tfor (let i = 0; i < transformedMessages.length; i++) {\n\t\tconst msg = transformedMessages[i];\n\t\t// Some providers don't allow user messages directly after tool results\n\t\t// Insert a synthetic assistant message to bridge the gap\n\t\tif (compat.requiresAssistantAfterToolResult && lastRole === \"toolResult\" && msg.role === \"user\") {\n\t\t\tparams.push({\n\t\t\t\trole: \"assistant\",\n\t\t\t\tcontent: \"I have processed the tool results.\",\n\t\t\t});\n\t\t}\n\n\t\tif (msg.role === \"user\") {\n\t\t\tif (typeof msg.content === \"string\") {\n\t\t\t\tparams.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent: sanitizeSurrogates(msg.content),\n\t\t\t\t});\n\t\t\t} else {\n\t\t\t\tconst content: ChatCompletionContentPart[] = msg.content.map((item): ChatCompletionContentPart => {\n\t\t\t\t\tif (item.type === \"text\") {\n\t\t\t\t\t\treturn {\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: sanitizeSurrogates(item.text),\n\t\t\t\t\t\t} satisfies ChatCompletionContentPartText;\n\t\t\t\t\t} else {\n\t\t\t\t\t\treturn {\n\t\t\t\t\t\t\ttype: \"image_url\",\n\t\t\t\t\t\t\timage_url: {\n\t\t\t\t\t\t\t\turl: `data:${item.mimeType};base64,${item.data}`,\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t} satisfies ChatCompletionContentPartImage;\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\tif (content.length === 0) continue;\n\t\t\t\tparams.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent,\n\t\t\t\t});\n\t\t\t}\n\t\t} else if (msg.role === \"assistant\") {\n\t\t\t// Some providers don't accept null content, use empty string instead\n\t\t\tconst assistantMsg: ChatCompletionAssistantMessageParam = {\n\t\t\t\trole: \"assistant\",\n\t\t\t\tcontent: compat.requiresAssistantAfterToolResult ? \"\" : null,\n\t\t\t};\n\n\t\t\tconst assistantTextParts = msg.content\n\t\t\t\t.filter(isTextContentBlock)\n\t\t\t\t.filter((block) => block.text.trim().length > 0)\n\t\t\t\t.map(\n\t\t\t\t\t(block) =>\n\t\t\t\t\t\t({\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: sanitizeSurrogates(block.text),\n\t\t\t\t\t\t}) satisfies ChatCompletionContentPartText,\n\t\t\t\t);\n\t\t\tconst assistantText = assistantTextParts.map((part) => part.text).join(\"\");\n\n\t\t\tconst nonEmptyThinkingBlocks = msg.content\n\t\t\t\t.filter(isThinkingContentBlock)\n\t\t\t\t.filter((block) => block.thinking.trim().length > 0);\n\t\t\tif (nonEmptyThinkingBlocks.length > 0) {\n\t\t\t\tif (compat.requiresThinkingAsText) {\n\t\t\t\t\t// Convert thinking blocks to plain text (no tags to avoid model mimicking them)\n\t\t\t\t\tconst thinkingText = nonEmptyThinkingBlocks\n\t\t\t\t\t\t.map((block) => sanitizeSurrogates(block.thinking))\n\t\t\t\t\t\t.join(\"\\n\\n\");\n\t\t\t\t\tassistantMsg.content = [{ type: \"text\", text: thinkingText }, ...assistantTextParts];\n\t\t\t\t} else {\n\t\t\t\t\t// Always send assistant content as a plain string (OpenAI Chat Completions\n\t\t\t\t\t// API standard format). Sending as an array of {type:\"text\", text:\"...\"}\n\t\t\t\t\t// objects is non-standard and causes some models (e.g. DeepSeek V3.2 via\n\t\t\t\t\t// NVIDIA NIM) to mirror the content-block structure literally in their\n\t\t\t\t\t// output, producing recursive nesting like [{'type':'text','text':'[{...}]'}].\n\t\t\t\t\tif (assistantText.length > 0) {\n\t\t\t\t\t\tassistantMsg.content = assistantText;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Use the signature from the first thinking block if available (for llama.cpp server + gpt-oss)\n\t\t\t\t\tconst signature = nonEmptyThinkingBlocks[0].thinkingSignature;\n\t\t\t\t\tif (signature && signature.length > 0) {\n\t\t\t\t\t\t(assistantMsg as any)[signature] = nonEmptyThinkingBlocks.map((block) => block.thinking).join(\"\\n\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} else if (assistantText.length > 0) {\n\t\t\t\t// Always send assistant content as a plain string (OpenAI Chat Completions\n\t\t\t\t// API standard format). Sending as an array of {type:\"text\", text:\"...\"}\n\t\t\t\t// objects is non-standard and causes some models (e.g. DeepSeek V3.2 via\n\t\t\t\t// NVIDIA NIM) to mirror the content-block structure literally in their\n\t\t\t\t// output, producing recursive nesting like [{'type':'text','text':'[{...}]'}].\n\t\t\t\tassistantMsg.content = assistantText;\n\t\t\t}\n\n\t\t\tconst toolCalls = msg.content.filter(isToolCallBlock);\n\t\t\tif (toolCalls.length > 0) {\n\t\t\t\tassistantMsg.tool_calls = toolCalls.map((tc) => ({\n\t\t\t\t\tid: tc.id,\n\t\t\t\t\ttype: \"function\" as const,\n\t\t\t\t\tfunction: {\n\t\t\t\t\t\tname: tc.name,\n\t\t\t\t\t\targuments: JSON.stringify(tc.arguments),\n\t\t\t\t\t},\n\t\t\t\t}));\n\t\t\t\tconst reasoningDetails = toolCalls\n\t\t\t\t\t.filter((tc) => tc.thoughtSignature)\n\t\t\t\t\t.map((tc) => {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn JSON.parse(tc.thoughtSignature!);\n\t\t\t\t\t\t} catch {\n\t\t\t\t\t\t\treturn null;\n\t\t\t\t\t\t}\n\t\t\t\t\t})\n\t\t\t\t\t.filter(Boolean);\n\t\t\t\tif (reasoningDetails.length > 0) {\n\t\t\t\t\t(assistantMsg as any).reasoning_details = reasoningDetails;\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (\n\t\t\t\tcompat.requiresReasoningContentOnAssistantMessages &&\n\t\t\t\tmodel.reasoning &&\n\t\t\t\t(assistantMsg as { reasoning_content?: string }).reasoning_content === undefined\n\t\t\t) {\n\t\t\t\t(assistantMsg as { reasoning_content?: string }).reasoning_content = \"\";\n\t\t\t}\n\t\t\t// Skip assistant messages that have no content and no tool calls.\n\t\t\t// Some providers require \"either content or tool_calls, but not none\".\n\t\t\t// Other providers also don't accept empty assistant messages.\n\t\t\t// This handles aborted assistant responses that got no content.\n\t\t\tconst content = assistantMsg.content;\n\t\t\tconst hasContent =\n\t\t\t\tcontent !== null &&\n\t\t\t\tcontent !== undefined &&\n\t\t\t\t(typeof content === \"string\" ? content.length > 0 : content.length > 0);\n\t\t\tif (!hasContent && !assistantMsg.tool_calls) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tparams.push(assistantMsg);\n\t\t} else if (msg.role === \"toolResult\") {\n\t\t\tconst imageBlocks: Array<{ type: \"image_url\"; image_url: { url: string } }> = [];\n\t\t\tlet j = i;\n\n\t\t\tfor (; j < transformedMessages.length && transformedMessages[j].role === \"toolResult\"; j++) {\n\t\t\t\tconst toolMsg = transformedMessages[j] as ToolResultMessage;\n\n\t\t\t\t// Extract text and image content\n\t\t\t\tconst textResult = toolMsg.content\n\t\t\t\t\t.filter(isTextContentBlock)\n\t\t\t\t\t.map((block) => block.text)\n\t\t\t\t\t.join(\"\\n\");\n\t\t\t\tconst hasImages = toolMsg.content.some((c) => c.type === \"image\");\n\n\t\t\t\t// Always send tool result with text (or placeholder if only images)\n\t\t\t\tconst hasText = textResult.length > 0;\n\t\t\t\t// Some providers require the 'name' field in tool results\n\t\t\t\tconst toolResultMsg: ChatCompletionToolMessageParam = {\n\t\t\t\t\trole: \"tool\",\n\t\t\t\t\tcontent: sanitizeSurrogates(hasText ? textResult : \"(see attached image)\"),\n\t\t\t\t\ttool_call_id: toolMsg.toolCallId,\n\t\t\t\t};\n\t\t\t\tif (compat.requiresToolResultName && toolMsg.toolName) {\n\t\t\t\t\t(toolResultMsg as any).name = toolMsg.toolName;\n\t\t\t\t}\n\t\t\t\tparams.push(toolResultMsg);\n\n\t\t\t\tif (hasImages && model.input.includes(\"image\")) {\n\t\t\t\t\tfor (const block of toolMsg.content) {\n\t\t\t\t\t\tif (isImageContentBlock(block)) {\n\t\t\t\t\t\t\timageBlocks.push({\n\t\t\t\t\t\t\t\ttype: \"image_url\",\n\t\t\t\t\t\t\t\timage_url: {\n\t\t\t\t\t\t\t\t\turl: `data:${block.mimeType};base64,${block.data}`,\n\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\ti = j - 1;\n\n\t\t\tif (imageBlocks.length > 0) {\n\t\t\t\tif (compat.requiresAssistantAfterToolResult) {\n\t\t\t\t\tparams.push({\n\t\t\t\t\t\trole: \"assistant\",\n\t\t\t\t\t\tcontent: \"I have processed the tool results.\",\n\t\t\t\t\t});\n\t\t\t\t}\n\n\t\t\t\tparams.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent: [\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: \"Attached image(s) from tool result:\",\n\t\t\t\t\t\t},\n\t\t\t\t\t\t...imageBlocks,\n\t\t\t\t\t],\n\t\t\t\t});\n\t\t\t\tlastRole = \"user\";\n\t\t\t} else {\n\t\t\t\tlastRole = \"toolResult\";\n\t\t\t}\n\t\t\tcontinue;\n\t\t}\n\n\t\tlastRole = msg.role;\n\t}\n\n\treturn params;\n}\n\nfunction convertTools(\n\ttools: Tool[],\n\tcompat: ResolvedOpenAICompletionsCompat,\n): OpenAI.Chat.Completions.ChatCompletionTool[] {\n\treturn tools.map((tool) => ({\n\t\ttype: \"function\",\n\t\tfunction: {\n\t\t\tname: tool.name,\n\t\t\tdescription: tool.description,\n\t\t\tparameters: tool.parameters as any, // TypeBox already generates JSON Schema\n\t\t\t// Only include strict if provider supports it. Some reject unknown fields.\n\t\t\t...(compat.supportsStrictMode !== false && { strict: false }),\n\t\t},\n\t}));\n}\n\nfunction parseChunkUsage(\n\trawUsage: {\n\t\tprompt_tokens?: number;\n\t\tcompletion_tokens?: number;\n\t\tprompt_tokens_details?: { cached_tokens?: number; cache_write_tokens?: number };\n\t},\n\tmodel: Model<\"openai-completions\">,\n): AssistantMessage[\"usage\"] {\n\tconst promptTokens = rawUsage.prompt_tokens \|\| 0;\n\tconst reportedCachedTokens = rawUsage.prompt_tokens_details?.cached_tokens \|\| 0;\n\tconst cacheWriteTokens = rawUsage.prompt_tokens_details?.cache_write_tokens \|\| 0;\n\n\t// Normalize to pi-ai semantics:\n\t// - cacheRead: hits from cache created by previous requests only\n\t// - cacheWrite: tokens written to cache in this request\n\t// Some OpenAI-compatible providers (observed on OpenRouter) report cached_tokens\n\t// as (previous hits + current writes). In that case, remove cacheWrite from cacheRead.\n\tconst cacheReadTokens =\n\t\tcacheWriteTokens > 0 ? Math.max(0, reportedCachedTokens - cacheWriteTokens) : reportedCachedTokens;\n\n\tconst input = Math.max(0, promptTokens - cacheReadTokens - cacheWriteTokens);\n\t// OpenAI completion_tokens already includes reasoning_tokens.\n\tconst outputTokens = rawUsage.completion_tokens \|\| 0;\n\tconst usage: AssistantMessage[\"usage\"] = {\n\t\tinput,\n\t\toutput: outputTokens,\n\t\tcacheRead: cacheReadTokens,\n\t\tcacheWrite: cacheWriteTokens,\n\t\ttotalTokens: input + outputTokens + cacheReadTokens + cacheWriteTokens,\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t};\n\tcalculateCost(model, usage);\n\treturn usage;\n}\n\nfunction mapStopReason(reason: ChatCompletionChunk.Choice[\"finish_reason\"] \| string): {\n\tstopReason: StopReason;\n\terrorMessage?: string;\n} {\n\tif (reason === null) return { stopReason: \"stop\" };\n\tswitch (reason) {\n\t\tcase \"stop\":\n\t\tcase \"end\":\n\t\t\treturn { stopReason: \"stop\" };\n\t\tcase \"length\":\n\t\t\treturn { stopReason: \"length\" };\n\t\tcase \"function_call\":\n\t\tcase \"tool_calls\":\n\t\t\treturn { stopReason: \"toolUse\" };\n\t\tcase \"content_filter\":\n\t\t\treturn { stopReason: \"error\", errorMessage: \"Provider finish_reason: content_filter\" };\n\t\tcase \"network_error\":\n\t\t\treturn { stopReason: \"error\", errorMessage: \"Provider finish_reason: network_error\" };\n\t\tdefault:\n\t\t\treturn {\n\t\t\t\tstopReason: \"error\",\n\t\t\t\terrorMessage: `Provider finish_reason: ${reason}`,\n\t\t\t};\n\t}\n}\n\n/\n Detect compatibility settings from provider and baseUrl for known providers.\n * Provider takes precedence over URL-based detection since it's explicitly configured.\n * Returns a fully resolved OpenAICompletionsCompat object with all fields set.\n /\nfunction detectCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst provider = model.provider;\n\tconst baseUrl = model.baseUrl;\n\n\tconst isZai = provider === \"zai\" \|\| baseUrl.includes(\"api.z.ai\");\n\n\tconst isNonStandard =\n\t\tprovider === \"cerebras\" \|\|\n\t\tbaseUrl.includes(\"cerebras.ai\") \|\|\n\t\tprovider === \"xai\" \|\|\n\t\tbaseUrl.includes(\"api.x.ai\") \|\|\n\t\tbaseUrl.includes(\"chutes.ai\") \|\|\n\t\tbaseUrl.includes(\"deepseek.com\") \|\|\n\t\tisZai \|\|\n\t\tprovider === \"opencode\" \|\|\n\t\tbaseUrl.includes(\"opencode.ai\");\n\n\tconst useMaxTokens = baseUrl.includes(\"chutes.ai\");\n\n\tconst isGrok = provider === \"xai\" \|\| baseUrl.includes(\"api.x.ai\");\n\tconst isGroq = provider === \"groq\" \|\| baseUrl.includes(\"groq.com\");\n\tconst isDeepSeek = provider === \"deepseek\" \|\| baseUrl.includes(\"deepseek.com\");\n\tconst cacheControlFormat = provider === \"openrouter\" && model.id.startsWith(\"anthropic/\") ? \"anthropic\" : undefined;\n\n\tconst reasoningEffortMap = isDeepSeek\n\t\t? {\n\t\t\t\tminimal: \"high\",\n\t\t\t\tlow: \"high\",\n\t\t\t\tmedium: \"high\",\n\t\t\t\thigh: \"high\",\n\t\t\t\txhigh: \"max\",\n\t\t\t}\n\t\t: isGroq && model.id === \"qwen/qwen3-32b\"\n\t\t\t? {\n\t\t\t\t\tminimal: \"default\",\n\t\t\t\t\tlow: \"default\",\n\t\t\t\t\tmedium: \"default\",\n\t\t\t\t\thigh: \"default\",\n\t\t\t\t\txhigh: \"default\",\n\t\t\t\t}\n\t\t\t: {};\n\treturn {\n\t\tsupportsStore: !isNonStandard,\n\t\tsupportsDeveloperRole: !isNonStandard,\n\t\tsupportsReasoningEffort: !isGrok && !isZai,\n\t\treasoningEffortMap,\n\t\tsupportsUsageInStreaming: true,\n\t\tmaxTokensField: useMaxTokens ? \"max_tokens\" : \"max_completion_tokens\",\n\t\trequiresToolResultName: false,\n\t\trequiresAssistantAfterToolResult: false,\n\t\trequiresThinkingAsText: false,\n\t\trequiresReasoningContentOnAssistantMessages: isDeepSeek,\n\t\tthinkingFormat: isDeepSeek\n\t\t\t? \"deepseek\"\n\t\t\t: isZai\n\t\t\t\t? \"zai\"\n\t\t\t\t: provider === \"openrouter\" \|\| baseUrl.includes(\"openrouter.ai\")\n\t\t\t\t\t? \"openrouter\"\n\t\t\t\t\t: \"openai\",\n\t\topenRouterRouting: {},\n\t\tvercelGatewayRouting: {},\n\t\tzaiToolStream: false,\n\t\tsupportsStrictMode: true,\n\t\tcacheControlFormat,\n\t\tsendSessionAffinityHeaders: false,\n\t\tsupportsLongCacheRetention: true,\n\t};\n}\n\n/\n Get resolved compatibility settings for a model.\n * Uses explicit model.compat if provided, otherwise auto-detects from provider/URL.\n */\nfunction getCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst detected = detectCompat(model);\n\tif (!model.compat) return detected;\n\n\treturn {\n\t\tsupportsStore: model.compat.supportsStore ?? detected.supportsStore,\n\t\tsupportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,\n\t\tsupportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,\n\t\treasoningEffortMap: model.compat.reasoningEffortMap ?? detected.reasoningEffortMap,\n\t\tsupportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? detected.supportsUsageInStreaming,\n\t\tmaxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,\n\t\trequiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,\n\t\trequiresAssistantAfterToolResult:\n\t\t\tmodel.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,\n\t\trequiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,\n\t\trequiresReasoningContentOnAssistantMessages:\n\t\t\tmodel.compat.requiresReasoningContentOnAssistantMessages ??\n\t\t\tdetected.requiresReasoningContentOnAssistantMessages,\n\t\tthinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,\n\t\topenRouterRouting: model.compat.openRouterRouting ?? {},\n\t\tvercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,\n\t\tzaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,\n\t\tsupportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,\n\t\tcacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,\n\t\tsendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,\n\t\tsupportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,\n\t};\n}\n"]}

package/dist/providers/openai-completions.js CHANGED Viewed

@@ -78,12 +78,13 @@ export const streamOpenAICompletions = (model, context, options) => {
             if (nextParams !== undefined) {
                 params = nextParams;
             }
+            const requestOptions = {
+                ...(options?.signal ? { signal: options.signal } : {}),
+                ...(options?.timeoutMs !== undefined ? { timeout: options.timeoutMs } : {}),
+                ...(options?.maxRetries !== undefined ? { maxRetries: options.maxRetries } : {}),
+            };
             const { data: openaiStream, response } = await client.chat.completions
-                .create(params, {
-                signal: options?.signal,
-                timeout: options?.timeoutMs,
-                maxRetries: options?.maxRetries,
-            })
+                .create(params, requestOptions)
                 .withResponse();
             await options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);
             stream.push({ type: "start", partial: output });