@earendil-works/pi-ai 0.80.3 → 0.80.6

package/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"","sourcesContent":["import type { AnthropicOptions } from \"./api/anthropic-messages.ts\";\nimport type { AzureOpenAIResponsesOptions } from \"./api/azure-openai-responses.ts\";\nimport type { BedrockOptions } from \"./api/bedrock-converse-stream.ts\";\nimport type { GoogleOptions } from \"./api/google-generative-ai.ts\";\nimport type { GoogleVertexOptions } from \"./api/google-vertex.ts\";\nimport type { MistralOptions } from \"./api/mistral-conversations.ts\";\nimport type { OpenAICodexResponsesOptions } from \"./api/openai-codex-responses.ts\";\nimport type { OpenAICompletionsOptions } from \"./api/openai-completions.ts\";\nimport type { OpenAIResponsesOptions } from \"./api/openai-responses.ts\";\nimport type { AssistantMessageDiagnostic } from \"./utils/diagnostics.ts\";\nimport type { AssistantMessageEventStream } from \"./utils/event-stream.ts\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.ts\";\n\nexport type KnownApi =\n\t| \"openai-completions\"\n\t| \"mistral-conversations\"\n\t| \"openai-responses\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-vertex\";\n\nexport type Api = KnownApi | (string & {});\n\nexport type KnownImagesApi = \"openrouter-images\";\n\nexport type ImagesApi = KnownImagesApi | (string & {});\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"ant-ling\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex\"\n\t| \"nvidia\"\n\t| \"deepseek\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"vercel-ai-gateway\"\n\t| \"zai\"\n\t| \"zai-coding-cn\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"minimax-cn\"\n\t| \"moonshotai\"\n\t| \"moonshotai-cn\"\n\t| \"huggingface\"\n\t| \"fireworks\"\n\t| \"together\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"cloudflare-workers-ai\"\n\t| \"cloudflare-ai-gateway\"\n\t| \"xiaomi\"\n\t| \"xiaomi-token-plan-cn\"\n\t| \"xiaomi-token-plan-ams\"\n\t| \"xiaomi-token-plan-sgp\";\nexport type ProviderId = KnownProvider | string;\n\nexport type KnownImagesProvider = \"openrouter\";\n\nexport type ImagesProviderId = KnownImagesProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\nexport type ModelThinkingLevel = \"off\" | ThinkingLevel;\nexport type ThinkingLevelMap = Partial<Record<ModelThinkingLevel, string | null>>;\nexport type ChatTemplateKwargValue =\n\t| string\n\t| number\n\t| boolean\n\t| null\n\t| {\n\t\t\t$var: \"thinking.enabled\" | \"thinking.effort\";\n\t\t\tomitWhenOff?: boolean;\n\t  };\n\n/** Token budgets for each thinking level (token-based providers only) */\nexport interface ThinkingBudgets {\n\tminimal?: number;\n\tlow?: number;\n\tmedium?: number;\n\thigh?: number;\n}\n\n// Base options all providers share\nexport type CacheRetention = \"none\" | \"short\" | \"long\";\n\nexport type Transport = \"sse\" | \"websocket\" | \"websocket-cached\" | \"auto\";\n\n/** Provider-scoped environment overrides. Values take precedence over process.env. */\nexport type ProviderEnv = Record<string, string>;\nexport type ProviderHeaders = Record<string, string | null>;\n\nexport interface ProviderResponse {\n\tstatus: number;\n\theaders: Record<string, string>;\n}\n\nexport interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Preferred transport for providers that support multiple transports.\n\t * Providers that do not support this option ignore it.\n\t */\n\ttransport?: Transport;\n\t/**\n\t * Prompt cache retention preference. Providers map this to their supported values.\n\t * Default: \"short\".\n\t */\n\tcacheRetention?: CacheRetention;\n\t/**\n\t * Optional session identifier for providers that support session-based caching.\n\t * Providers can use this to enable prompt caching, request routing, or other\n\t * session-aware features. Ignored by providers that don't support it.\n\t */\n\tsessionId?: string;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: Model<Api>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional callback invoked after an HTTP response is received and before\n\t * its body stream is consumed.\n\t */\n\tonResponse?: (response: ProviderResponse, model: Model<Api>) => void | Promise<void>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; caller values override default headers.\n\t * On AWS Bedrock these are injected via a Smithy `build`-step middleware so\n\t * they are covered by SigV4 signing; reserved headers (`x-amz-*`,\n\t * `authorization`, `host`) are silently ignored to preserve SigV4 / bearer auth.\n\t * A null value suppresses a provider/API default header with the same name.\n\t */\n\theaders?: ProviderHeaders;\n\t/**\n\t * HTTP request timeout in milliseconds for providers/SDKs that support it.\n\t * For example, OpenAI and Anthropic SDK clients default to 10 minutes.\n\t */\n\ttimeoutMs?: number;\n\t/**\n\t * WebSocket connect timeout in milliseconds for providers that support\n\t * WebSocket transports. This covers the connection/open handshake only;\n\t * stream idleness after connection uses timeoutMs.\n\t */\n\twebsocketConnectTimeoutMs?: number;\n\t/**\n\t * Maximum retry attempts for providers/SDKs that support client-side retries.\n\t * For example, OpenAI and Anthropic SDK clients default to 2.\n\t */\n\tmaxRetries?: number;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.\n\t */\n\tmetadata?: Record<string, unknown>;\n\t/**\n\t * Provider-scoped environment values. These take precedence over process.env for\n\t * provider configuration such as regional settings, endpoint placeholders, and\n\t * proxy variables.\n\t */\n\tenv?: ProviderEnv;\n}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\n/**\n * Maps known APIs to their full provider-specific stream option types.\n * Type-only imports from API implementation modules are erased at emit, so\n * this is tree-shake safe.\n */\nexport interface ApiOptionsMap {\n\t\"anthropic-messages\": AnthropicOptions;\n\t\"openai-completions\": OpenAICompletionsOptions;\n\t\"openai-responses\": OpenAIResponsesOptions;\n\t\"openai-codex-responses\": OpenAICodexResponsesOptions;\n\t\"azure-openai-responses\": AzureOpenAIResponsesOptions;\n\t\"google-generative-ai\": GoogleOptions;\n\t\"google-vertex\": GoogleVertexOptions;\n\t\"mistral-conversations\": MistralOptions;\n\t\"bedrock-converse-stream\": BedrockOptions;\n}\n\n/**\n * Full stream options for an API. Known APIs resolve to their concrete option\n * type; custom API strings fall back to the generic shape.\n */\nexport type ApiStreamOptions<TApi extends Api> = TApi extends keyof ApiOptionsMap\n\t? ApiOptionsMap[TApi]\n\t: StreamOptions & Record<string, unknown>;\n\n/**\n * The uniform stream contract of an API implementation module: every module\n * under `src/api/` exports exactly `stream` and `streamSimple`, so the module\n * itself satisfies this interface. Lazy wrappers (`lazyApi()`) and provider\n * factories pass these around as values. This is the untyped dispatch shape;\n * per-API option typing lives on the implementation modules themselves and on\n * `Provider.stream()` via `ApiStreamOptions`.\n */\nexport interface ProviderStreams {\n\tstream(model: Model<Api>, context: Context, options?: StreamOptions): AssistantMessageEventStream;\n\tstreamSimple(model: Model<Api>, context: Context, options?: SimpleStreamOptions): AssistantMessageEventStream;\n}\n\n/**\n * The uniform contract of an image-generation API implementation module:\n * every image API module under `src/api/` exports exactly `generateImages`,\n * so the module itself satisfies this interface. Lazy wrappers and image\n * provider factories pass these around as values.\n */\nexport interface ProviderImages {\n\tgenerateImages(\n\t\tmodel: ImagesModel<ImagesApi>,\n\t\tcontext: ImagesContext,\n\t\toptions?: ImagesOptions,\n\t): Promise<AssistantImages>;\n}\n\nexport interface ImagesOptions {\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Provider-scoped environment values. These take precedence over process.env for\n\t * provider configuration such as endpoint placeholders and proxy variables.\n\t */\n\tenv?: ProviderEnv;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: ImagesModel<ImagesApi>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional callback invoked after an HTTP response is received.\n\t */\n\tonResponse?: (response: ProviderResponse, model: ImagesModel<ImagesApi>) => void | Promise<void>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; can override default headers.\n\t * A null value suppresses a provider/API default header with the same name.\n\t */\n\theaders?: ProviderHeaders;\n\t/**\n\t * HTTP request timeout in milliseconds for providers/SDKs that support it.\n\t */\n\ttimeoutMs?: number;\n\t/**\n\t * Maximum retry attempts for providers/SDKs that support client-side retries.\n\t */\n\tmaxRetries?: number;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t */\n\tmetadata?: Record<string, unknown>;\n}\n\nexport type ProviderImagesOptions = ImagesOptions & Record<string, unknown>;\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n}\n\n// Generic StreamFunction with typed options.\n//\n// Contract:\n// - Must return an AssistantMessageEventStream.\n// - Once invoked, request/model/runtime failures should be encoded in the\n//   returned stream, not thrown.\n// - Error termination must produce an AssistantMessage with stopReason\n//   \"error\" or \"aborted\" and errorMessage, emitted via the stream protocol.\nexport type StreamFunction<TApi extends Api = Api, TOptions extends StreamOptions = StreamOptions> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: TOptions,\n) => AssistantMessageEventStream;\n\nexport type ImagesFunction<TApi extends ImagesApi = ImagesApi, TOptions extends ImagesOptions = ImagesOptions> = (\n\tmodel: ImagesModel<TApi>,\n\tcontext: ImagesContext,\n\toptions?: TOptions,\n) => Promise<AssistantImages>;\n\nexport interface TextSignatureV1 {\n\tv: 1;\n\tid: string;\n\tphase?: \"commentary\" | \"final_answer\";\n}\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n\t/** When true, the thinking content was redacted by safety filters. The opaque\n\t *  encrypted payload is stored in `thinkingSignature` so it can be passed back\n\t *  to the API for multi-turn continuity. */\n\tredacted?: boolean;\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\t/** Subset of `cacheWrite` written with 1h retention. Only Anthropic reports this split. */\n\tcacheWrite1h?: number;\n\t/**\n\t * Reasoning/thinking tokens, when the provider reports them. This is a subset of\n\t * `output`: `output` already includes these tokens. Set to a number (possibly 0) by\n\t * providers that expose a reasoning breakdown; left undefined by providers that don't.\n\t */\n\treasoning?: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: ProviderId;\n\tmodel: string;\n\tresponseModel?: string; // Concrete `chunk.model` when different from the requested `model` (e.g. OpenRouter `auto` -> `anthropic/...`)\n\tresponseId?: string; // Provider-specific response/message identifier when the upstream API exposes one\n\tdiagnostics?: AssistantMessageDiagnostic[]; // Redacted provider/runtime diagnostics for failures and recoveries.\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nexport type ImagesInputContent = TextContent | ImageContent;\nexport type ImagesOutputContent = TextContent | ImageContent;\n\nexport interface ImagesContext {\n\tinput: ImagesInputContent[];\n}\n\nexport type ImagesStopReason = \"stop\" | \"error\" | \"aborted\";\n\nexport interface AssistantImages {\n\tapi: ImagesApi;\n\tprovider: ImagesProviderId;\n\tmodel: string;\n\toutput: ImagesOutputContent[];\n\tresponseId?: string;\n\tusage?: Usage;\n\tstopReason: ImagesStopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nimport type { TSchema } from \"typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\n/**\n * Event protocol for AssistantMessageEventStream.\n *\n * Streams should emit `start` before partial updates, then terminate with either:\n * - `done` carrying the final successful AssistantMessage, or\n * - `error` carrying the final AssistantMessage with stopReason \"error\" or \"aborted\"\n *   and errorMessage.\n */\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for OpenAI-compatible completions APIs.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompletionsCompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Whether all replayed assistant messages must include an empty reasoning_content field when reasoning is enabled. Default: auto-detected from URL. */\n\trequiresReasoningContentOnAssistantMessages?: boolean;\n\t/** Format for reasoning/thinking parameter. \"openai\" uses reasoning_effort, \"openrouter\" uses reasoning: { effort }, \"deepseek\" uses thinking: { type } plus reasoning_effort when supported, \"together\" uses reasoning: { enabled } plus reasoning_effort when supported, \"zai\" uses thinking: { type }, \"qwen\" uses top-level enable_thinking: boolean, \"qwen-chat-template\" uses chat_template_kwargs.enable_thinking and preserve_thinking, \"chat-template\" uses configurable chat_template_kwargs, \"string-thinking\" uses top-level thinking: string, and \"ant-ling\" uses reasoning: { effort } only when the mapped effort is non-null. Default: \"openai\". */\n\tthinkingFormat?:\n\t\t| \"openai\"\n\t\t| \"openrouter\"\n\t\t| \"deepseek\"\n\t\t| \"together\"\n\t\t| \"zai\"\n\t\t| \"qwen\"\n\t\t| \"chat-template\"\n\t\t| \"qwen-chat-template\"\n\t\t| \"string-thinking\"\n\t\t| \"ant-ling\";\n\t/** Kwargs to send as `chat_template_kwargs` when `thinkingFormat` is `chat-template`. Use `{ \"$var\": \"thinking.enabled\" }` or `{ \"$var\": \"thinking.effort\" }` for pi-controlled thinking values. */\n\tchatTemplateKwargs?: Record<string, ChatTemplateKwargValue>;\n\t/** OpenRouter-compatible routing preferences sent as the `provider` request field. */\n\topenRouterRouting?: OpenRouterRouting;\n\t/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */\n\tvercelGatewayRouting?: VercelGatewayRouting;\n\t/** Whether z.ai supports top-level `tool_stream: true` for streaming tool call deltas. Default: false. */\n\tzaiToolStream?: boolean;\n\t/** Whether the provider supports the `strict` field in tool definitions. Default: true. */\n\tsupportsStrictMode?: boolean;\n\t/** Cache control convention for prompt caching. \"anthropic\" applies Anthropic-style `cache_control` markers to the system prompt, last tool definition, and last user/assistant text content. */\n\tcacheControlFormat?: \"anthropic\";\n\t/** Whether to send known session-affinity headers (`session_id`, `x-client-request-id`, `x-session-affinity`) from `options.sessionId` when caching is enabled. Default: false. */\n\tsendSessionAffinityHeaders?: boolean;\n\t/** Whether the provider supports long prompt cache retention (`prompt_cache_retention: \"24h\"` or Anthropic-style `cache_control.ttl: \"1h\"`, depending on format). Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/** Compatibility settings for OpenAI Responses APIs. */\nexport interface OpenAIResponsesCompat {\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: true. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether to send the OpenAI `session_id` cache-affinity header from `options.sessionId` when caching is enabled. Default: true. */\n\tsendSessionIdHeader?: boolean;\n\t/** Whether the provider supports `prompt_cache_retention: \"24h\"`. Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/** Compatibility settings for Anthropic Messages-compatible APIs. */\nexport interface AnthropicMessagesCompat {\n\t/**\n\t * Whether the provider accepts per-tool `eager_input_streaming`.\n\t * When false, the Anthropic provider omits `tools[].eager_input_streaming`\n\t * and sends the legacy `fine-grained-tool-streaming-2025-05-14` beta header\n\t * for tool-enabled requests.\n\t * Default: true.\n\t */\n\tsupportsEagerToolInputStreaming?: boolean;\n\t/** Whether the provider supports Anthropic long cache retention (`cache_control.ttl: \"1h\"`). Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n\t/**\n\t * Whether to send the `x-session-affinity` header from `options.sessionId`\n\t * when caching is enabled. Required for providers like Fireworks that use\n\t * session affinity for prompt cache routing (requests to the same replica\n\t * maximize cache hits).\n\t * Default: false.\n\t */\n\tsendSessionAffinityHeaders?: boolean;\n\t/**\n\t * Whether the provider supports Anthropic-style `cache_control` markers on\n\t * tool definitions. When false, `cache_control` is omitted from tool params.\n\t * Some Anthropic-compatible providers (e.g., Fireworks) do not support this\n\t * field on tools and may reject or ignore it.\n\t * Default: true.\n\t */\n\tsupportsCacheControlOnTools?: boolean;\n\t/**\n\t * Whether the model accepts the Anthropic `temperature` request field.\n\t * Claude Opus 4.7+ rejects non-default temperature values.\n\t * Default: true.\n\t */\n\tsupportsTemperature?: boolean;\n\t/**\n\t * Whether to force adaptive thinking (`thinking.type: \"adaptive\"` plus\n\t * `output_config.effort`) regardless of the model id. Built-in models that\n\t * require adaptive thinking set this in generated metadata. Custom\n\t * Anthropic-compatible providers can set this to `true` for any model whose\n\t * upstream requires the adaptive format. Set to `false` to\n\t * opt out on overridden built-in models.\n\t * Default: false.\n\t */\n\tforceAdaptiveThinking?: boolean;\n\t/** Whether to replay empty thinking signatures as `signature: \"\"` instead of converting thinking to text. Default: false. */\n\tallowEmptySignature?: boolean;\n}\n\n/**\n * OpenRouter provider routing preferences.\n * Controls which upstream providers OpenRouter routes requests to.\n * Sent as the `provider` field in the OpenRouter API request body.\n * @see https://openrouter.ai/docs/guides/routing/provider-selection\n */\nexport interface OpenRouterRouting {\n\t/** Whether to allow backup providers to serve requests. Default: true. */\n\tallow_fallbacks?: boolean;\n\t/** Whether to filter providers to only those that support all parameters in the request. Default: false. */\n\trequire_parameters?: boolean;\n\t/** Data collection setting. \"allow\" (default): allow providers that may store/train on data. \"deny\": only use providers that don't collect user data. */\n\tdata_collection?: \"deny\" | \"allow\";\n\t/** Whether to restrict routing to only ZDR (Zero Data Retention) endpoints. */\n\tzdr?: boolean;\n\t/** Whether to restrict routing to only models that allow text distillation. */\n\tenforce_distillable_text?: boolean;\n\t/** An ordered list of provider names/slugs to try in sequence, falling back to the next if unavailable. */\n\torder?: string[];\n\t/** List of provider names/slugs to exclusively allow for this request. */\n\tonly?: string[];\n\t/** List of provider names/slugs to skip for this request. */\n\tignore?: string[];\n\t/** A list of quantization levels to filter providers by (e.g., [\"fp16\", \"bf16\", \"fp8\", \"fp6\", \"int8\", \"int4\", \"fp4\", \"fp32\"]). */\n\tquantizations?: string[];\n\t/** Sorting strategy. Can be a string (e.g., \"price\", \"throughput\", \"latency\") or an object with `by` and `partition`. */\n\tsort?:\n\t\t| string\n\t\t| {\n\t\t\t\t/** The sorting metric: \"price\", \"throughput\", \"latency\". */\n\t\t\t\tby?: string;\n\t\t\t\t/** Partitioning strategy: \"model\" (default) or \"none\". */\n\t\t\t\tpartition?: string | null;\n\t\t  };\n\t/** Maximum price per million tokens (USD). */\n\tmax_price?: {\n\t\t/** Price per million prompt tokens. */\n\t\tprompt?: number | string;\n\t\t/** Price per million completion tokens. */\n\t\tcompletion?: number | string;\n\t\t/** Price per image. */\n\t\timage?: number | string;\n\t\t/** Price per audio unit. */\n\t\taudio?: number | string;\n\t\t/** Price per request. */\n\t\trequest?: number | string;\n\t};\n\t/** Preferred minimum throughput (tokens/second). Can be a number (applies to p50) or an object with percentile-specific cutoffs. */\n\tpreferred_min_throughput?:\n\t\t| number\n\t\t| {\n\t\t\t\t/** Minimum tokens/second at the 50th percentile. */\n\t\t\t\tp50?: number;\n\t\t\t\t/** Minimum tokens/second at the 75th percentile. */\n\t\t\t\tp75?: number;\n\t\t\t\t/** Minimum tokens/second at the 90th percentile. */\n\t\t\t\tp90?: number;\n\t\t\t\t/** Minimum tokens/second at the 99th percentile. */\n\t\t\t\tp99?: number;\n\t\t  };\n\t/** Preferred maximum latency (seconds). Can be a number (applies to p50) or an object with percentile-specific cutoffs. */\n\tpreferred_max_latency?:\n\t\t| number\n\t\t| {\n\t\t\t\t/** Maximum latency in seconds at the 50th percentile. */\n\t\t\t\tp50?: number;\n\t\t\t\t/** Maximum latency in seconds at the 75th percentile. */\n\t\t\t\tp75?: number;\n\t\t\t\t/** Maximum latency in seconds at the 90th percentile. */\n\t\t\t\tp90?: number;\n\t\t\t\t/** Maximum latency in seconds at the 99th percentile. */\n\t\t\t\tp99?: number;\n\t\t  };\n}\n\n/**\n * Vercel AI Gateway routing preferences.\n * Controls which upstream providers the gateway routes requests to.\n * @see https://vercel.com/docs/ai-gateway/models-and-providers/provider-options\n */\nexport interface VercelGatewayRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\n// Model interface for the unified model system\nexport interface Model<TApi extends Api> {\n\tid: string;\n\tname: string;\n\tapi: TApi;\n\tprovider: ProviderId;\n\tbaseUrl: string;\n\treasoning: boolean;\n\t/**\n\t * Maps pi thinking levels to provider/model-specific values.\n\t * Missing keys use provider defaults. null marks a level as unsupported.\n\t */\n\tthinkingLevelMap?: ThinkingLevelMap;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/** Compatibility overrides for OpenAI-compatible APIs. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\"\n\t\t? OpenAICompletionsCompat\n\t\t: TApi extends \"openai-responses\"\n\t\t\t? OpenAIResponsesCompat\n\t\t\t: TApi extends \"anthropic-messages\"\n\t\t\t\t? AnthropicMessagesCompat\n\t\t\t\t: never;\n}\n\nexport interface ImagesModel<TApi extends ImagesApi>\n\textends Omit<Model<Api>, \"api\" | \"provider\" | \"reasoning\" | \"contextWindow\" | \"maxTokens\" | \"compat\"> {\n\tapi: TApi;\n\tprovider: ImagesProviderId;\n\toutput: (\"text\" | \"image\")[];\n}\n"]}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"","sourcesContent":["import type { AnthropicOptions } from \"./api/anthropic-messages.ts\";\nimport type { AzureOpenAIResponsesOptions } from \"./api/azure-openai-responses.ts\";\nimport type { BedrockOptions } from \"./api/bedrock-converse-stream.ts\";\nimport type { GoogleOptions } from \"./api/google-generative-ai.ts\";\nimport type { GoogleVertexOptions } from \"./api/google-vertex.ts\";\nimport type { MistralOptions } from \"./api/mistral-conversations.ts\";\nimport type { OpenAICodexResponsesOptions } from \"./api/openai-codex-responses.ts\";\nimport type { OpenAICompletionsOptions } from \"./api/openai-completions.ts\";\nimport type { OpenAIResponsesOptions } from \"./api/openai-responses.ts\";\nimport type { AssistantMessageDiagnostic } from \"./utils/diagnostics.ts\";\nimport type { AssistantMessageEventStream } from \"./utils/event-stream.ts\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.ts\";\n\nexport type KnownApi =\n\t| \"openai-completions\"\n\t| \"mistral-conversations\"\n\t| \"openai-responses\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-vertex\";\n\nexport type Api = KnownApi | (string & {});\n\nexport type KnownImagesApi = \"openrouter-images\";\n\nexport type ImagesApi = KnownImagesApi | (string & {});\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"ant-ling\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"azure-openai-responses\"\n\t| \"openai-codex\"\n\t| \"nvidia\"\n\t| \"deepseek\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"vercel-ai-gateway\"\n\t| \"zai\"\n\t| \"zai-coding-cn\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"minimax-cn\"\n\t| \"moonshotai\"\n\t| \"moonshotai-cn\"\n\t| \"huggingface\"\n\t| \"fireworks\"\n\t| \"together\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"cloudflare-workers-ai\"\n\t| \"cloudflare-ai-gateway\"\n\t| \"xiaomi\"\n\t| \"xiaomi-token-plan-cn\"\n\t| \"xiaomi-token-plan-ams\"\n\t| \"xiaomi-token-plan-sgp\";\nexport type ProviderId = KnownProvider | string;\n\nexport type KnownImagesProvider = \"openrouter\";\n\nexport type ImagesProviderId = KnownImagesProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\" | \"max\";\nexport type ModelThinkingLevel = \"off\" | ThinkingLevel;\nexport type ThinkingLevelMap = Partial<Record<ModelThinkingLevel, string | null>>;\nexport type ChatTemplateKwargValue =\n\t| string\n\t| number\n\t| boolean\n\t| null\n\t| {\n\t\t\t$var: \"thinking.enabled\" | \"thinking.effort\";\n\t\t\tomitWhenOff?: boolean;\n\t  };\n\n/** Token budgets for each thinking level (token-based providers only) */\nexport interface ThinkingBudgets {\n\tminimal?: number;\n\tlow?: number;\n\tmedium?: number;\n\thigh?: number;\n}\n\n// Base options all providers share\nexport type CacheRetention = \"none\" | \"short\" | \"long\";\n\nexport type Transport = \"sse\" | \"websocket\" | \"websocket-cached\" | \"auto\";\n\n/** Provider-scoped environment overrides. Values take precedence over process.env. */\nexport type ProviderEnv = Record<string, string>;\nexport type ProviderHeaders = Record<string, string | null>;\n\nexport interface ProviderResponse {\n\tstatus: number;\n\theaders: Record<string, string>;\n}\n\nexport interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Preferred transport for providers that support multiple transports.\n\t * Providers that do not support this option ignore it.\n\t */\n\ttransport?: Transport;\n\t/**\n\t * Prompt cache retention preference. Providers map this to their supported values.\n\t * Default: \"short\".\n\t */\n\tcacheRetention?: CacheRetention;\n\t/**\n\t * Optional session identifier for providers that support session-based caching.\n\t * Providers can use this to enable prompt caching, request routing, or other\n\t * session-aware features. Ignored by providers that don't support it.\n\t */\n\tsessionId?: string;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: Model<Api>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional callback invoked after an HTTP response is received and before\n\t * its body stream is consumed.\n\t */\n\tonResponse?: (response: ProviderResponse, model: Model<Api>) => void | Promise<void>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; caller values override default headers.\n\t * On AWS Bedrock these are injected via a Smithy `build`-step middleware so\n\t * they are covered by SigV4 signing; reserved headers (`x-amz-*`,\n\t * `authorization`, `host`) are silently ignored to preserve SigV4 / bearer auth.\n\t * A null value suppresses a provider/API default header with the same name.\n\t */\n\theaders?: ProviderHeaders;\n\t/**\n\t * HTTP request timeout in milliseconds for providers/SDKs that support it.\n\t * For example, OpenAI and Anthropic SDK clients default to 10 minutes.\n\t */\n\ttimeoutMs?: number;\n\t/**\n\t * WebSocket connect timeout in milliseconds for providers that support\n\t * WebSocket transports. This covers the connection/open handshake only;\n\t * stream idleness after connection uses timeoutMs.\n\t */\n\twebsocketConnectTimeoutMs?: number;\n\t/**\n\t * Maximum retry attempts for providers/SDKs that support client-side retries.\n\t * For example, OpenAI and Anthropic SDK clients default to 2.\n\t */\n\tmaxRetries?: number;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.\n\t */\n\tmetadata?: Record<string, unknown>;\n\t/**\n\t * Provider-scoped environment values. These take precedence over process.env for\n\t * provider configuration such as regional settings, endpoint placeholders, and\n\t * proxy variables.\n\t */\n\tenv?: ProviderEnv;\n}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\n/**\n * Maps known APIs to their full provider-specific stream option types.\n * Type-only imports from API implementation modules are erased at emit, so\n * this is tree-shake safe.\n */\nexport interface ApiOptionsMap {\n\t\"anthropic-messages\": AnthropicOptions;\n\t\"openai-completions\": OpenAICompletionsOptions;\n\t\"openai-responses\": OpenAIResponsesOptions;\n\t\"openai-codex-responses\": OpenAICodexResponsesOptions;\n\t\"azure-openai-responses\": AzureOpenAIResponsesOptions;\n\t\"google-generative-ai\": GoogleOptions;\n\t\"google-vertex\": GoogleVertexOptions;\n\t\"mistral-conversations\": MistralOptions;\n\t\"bedrock-converse-stream\": BedrockOptions;\n}\n\n/**\n * Full stream options for an API. Known APIs resolve to their concrete option\n * type; custom API strings fall back to the generic shape.\n */\nexport type ApiStreamOptions<TApi extends Api> = TApi extends keyof ApiOptionsMap\n\t? ApiOptionsMap[TApi]\n\t: StreamOptions & Record<string, unknown>;\n\n/**\n * The uniform stream contract of an API implementation module: every module\n * under `src/api/` exports exactly `stream` and `streamSimple`, so the module\n * itself satisfies this interface. Lazy wrappers (`lazyApi()`) and provider\n * factories pass these around as values. This is the untyped dispatch shape;\n * per-API option typing lives on the implementation modules themselves and on\n * `Provider.stream()` via `ApiStreamOptions`.\n */\nexport interface ProviderStreams {\n\tstream(model: Model<Api>, context: Context, options?: StreamOptions): AssistantMessageEventStream;\n\tstreamSimple(model: Model<Api>, context: Context, options?: SimpleStreamOptions): AssistantMessageEventStream;\n}\n\n/**\n * The uniform contract of an image-generation API implementation module:\n * every image API module under `src/api/` exports exactly `generateImages`,\n * so the module itself satisfies this interface. Lazy wrappers and image\n * provider factories pass these around as values.\n */\nexport interface ProviderImages {\n\tgenerateImages(\n\t\tmodel: ImagesModel<ImagesApi>,\n\t\tcontext: ImagesContext,\n\t\toptions?: ImagesOptions,\n\t): Promise<AssistantImages>;\n}\n\nexport interface ImagesOptions {\n\tsignal?: AbortSignal;\n\tapiKey?: string;\n\t/**\n\t * Provider-scoped environment values. These take precedence over process.env for\n\t * provider configuration such as endpoint placeholders and proxy variables.\n\t */\n\tenv?: ProviderEnv;\n\t/**\n\t * Optional callback for inspecting or replacing provider payloads before sending.\n\t * Return undefined to keep the payload unchanged.\n\t */\n\tonPayload?: (payload: unknown, model: ImagesModel<ImagesApi>) => unknown | undefined | Promise<unknown | undefined>;\n\t/**\n\t * Optional callback invoked after an HTTP response is received.\n\t */\n\tonResponse?: (response: ProviderResponse, model: ImagesModel<ImagesApi>) => void | Promise<void>;\n\t/**\n\t * Optional custom HTTP headers to include in API requests.\n\t * Merged with provider defaults; can override default headers.\n\t * A null value suppresses a provider/API default header with the same name.\n\t */\n\theaders?: ProviderHeaders;\n\t/**\n\t * HTTP request timeout in milliseconds for providers/SDKs that support it.\n\t */\n\ttimeoutMs?: number;\n\t/**\n\t * Maximum retry attempts for providers/SDKs that support client-side retries.\n\t */\n\tmaxRetries?: number;\n\t/**\n\t * Maximum delay in milliseconds to wait for a retry when the server requests a long wait.\n\t * If the server's requested delay exceeds this value, the request fails immediately\n\t * with an error containing the requested delay, allowing higher-level retry logic\n\t * to handle it with user visibility.\n\t * Default: 60000 (60 seconds). Set to 0 to disable the cap.\n\t */\n\tmaxRetryDelayMs?: number;\n\t/**\n\t * Optional metadata to include in API requests.\n\t * Providers extract the fields they understand and ignore the rest.\n\t */\n\tmetadata?: Record<string, unknown>;\n}\n\nexport type ProviderImagesOptions = ImagesOptions & Record<string, unknown>;\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n}\n\n// Generic StreamFunction with typed options.\n//\n// Contract:\n// - Must return an AssistantMessageEventStream.\n// - Once invoked, request/model/runtime failures should be encoded in the\n//   returned stream, not thrown.\n// - Error termination must produce an AssistantMessage with stopReason\n//   \"error\" or \"aborted\" and errorMessage, emitted via the stream protocol.\nexport type StreamFunction<TApi extends Api = Api, TOptions extends StreamOptions = StreamOptions> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: TOptions,\n) => AssistantMessageEventStream;\n\nexport type ImagesFunction<TApi extends ImagesApi = ImagesApi, TOptions extends ImagesOptions = ImagesOptions> = (\n\tmodel: ImagesModel<TApi>,\n\tcontext: ImagesContext,\n\toptions?: TOptions,\n) => Promise<AssistantImages>;\n\nexport interface TextSignatureV1 {\n\tv: 1;\n\tid: string;\n\tphase?: \"commentary\" | \"final_answer\";\n}\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n\t/** When true, the thinking content was redacted by safety filters. The opaque\n\t *  encrypted payload is stored in `thinkingSignature` so it can be passed back\n\t *  to the API for multi-turn continuity. */\n\tredacted?: boolean;\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\t/** Subset of `cacheWrite` written with 1h retention. Only Anthropic reports this split. */\n\tcacheWrite1h?: number;\n\t/**\n\t * Reasoning/thinking tokens, when the provider reports them. This is a subset of\n\t * `output`: `output` already includes these tokens. Set to a number (possibly 0) by\n\t * providers that expose a reasoning breakdown; left undefined by providers that don't.\n\t */\n\treasoning?: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: ProviderId;\n\tmodel: string;\n\tresponseModel?: string; // Concrete `chunk.model` when different from the requested `model` (e.g. OpenRouter `auto` -> `anthropic/...`)\n\tresponseId?: string; // Provider-specific response/message identifier when the upstream API exposes one\n\tdiagnostics?: AssistantMessageDiagnostic[]; // Redacted provider/runtime diagnostics for failures and recoveries.\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nexport type ImagesInputContent = TextContent | ImageContent;\nexport type ImagesOutputContent = TextContent | ImageContent;\n\nexport interface ImagesContext {\n\tinput: ImagesInputContent[];\n}\n\nexport type ImagesStopReason = \"stop\" | \"error\" | \"aborted\";\n\nexport interface AssistantImages {\n\tapi: ImagesApi;\n\tprovider: ImagesProviderId;\n\tmodel: string;\n\toutput: ImagesOutputContent[];\n\tresponseId?: string;\n\tusage?: Usage;\n\tstopReason: ImagesStopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nimport type { TSchema } from \"typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\n/**\n * Event protocol for AssistantMessageEventStream.\n *\n * Streams should emit `start` before partial updates, then terminate with either:\n * - `done` carrying the final successful AssistantMessage, or\n * - `error` carrying the final AssistantMessage with stopReason \"error\" or \"aborted\"\n *   and errorMessage.\n */\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for OpenAI-compatible completions APIs.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompletionsCompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Whether all replayed assistant messages must include an empty reasoning_content field when reasoning is enabled. Default: auto-detected from URL. */\n\trequiresReasoningContentOnAssistantMessages?: boolean;\n\t/** Format for reasoning/thinking parameter. \"openai\" uses reasoning_effort, \"openrouter\" uses reasoning: { effort }, \"deepseek\" uses thinking: { type } plus reasoning_effort when supported, \"together\" uses reasoning: { enabled } plus reasoning_effort when supported, \"zai\" uses thinking: { type }, \"qwen\" uses top-level enable_thinking: boolean, \"qwen-chat-template\" uses chat_template_kwargs.enable_thinking and preserve_thinking, \"chat-template\" uses configurable chat_template_kwargs, \"string-thinking\" uses top-level thinking: string, and \"ant-ling\" uses reasoning: { effort } only when the mapped effort is non-null. Default: \"openai\". */\n\tthinkingFormat?:\n\t\t| \"openai\"\n\t\t| \"openrouter\"\n\t\t| \"deepseek\"\n\t\t| \"together\"\n\t\t| \"zai\"\n\t\t| \"qwen\"\n\t\t| \"chat-template\"\n\t\t| \"qwen-chat-template\"\n\t\t| \"string-thinking\"\n\t\t| \"ant-ling\";\n\t/** Kwargs to send as `chat_template_kwargs` when `thinkingFormat` is `chat-template`. Use `{ \"$var\": \"thinking.enabled\" }` or `{ \"$var\": \"thinking.effort\" }` for pi-controlled thinking values. */\n\tchatTemplateKwargs?: Record<string, ChatTemplateKwargValue>;\n\t/** OpenRouter-compatible routing preferences sent as the `provider` request field. */\n\topenRouterRouting?: OpenRouterRouting;\n\t/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */\n\tvercelGatewayRouting?: VercelGatewayRouting;\n\t/** Whether z.ai supports top-level `tool_stream: true` for streaming tool call deltas. Default: false. */\n\tzaiToolStream?: boolean;\n\t/** Whether the provider supports the `strict` field in tool definitions. Default: true. */\n\tsupportsStrictMode?: boolean;\n\t/** Cache control convention for prompt caching. \"anthropic\" applies Anthropic-style `cache_control` markers to the system prompt, last tool definition, and last user/assistant text content. */\n\tcacheControlFormat?: \"anthropic\";\n\t/** Whether to send known session-affinity headers (`session_id`, `x-client-request-id`, `x-session-affinity`) from `options.sessionId` when caching is enabled. Default: false. */\n\tsendSessionAffinityHeaders?: boolean;\n\t/** Whether the provider supports long prompt cache retention (`prompt_cache_retention: \"24h\"` or Anthropic-style `cache_control.ttl: \"1h\"`, depending on format). Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/** Compatibility settings for OpenAI Responses APIs. */\nexport interface OpenAIResponsesCompat {\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: true. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether to send the OpenAI `session_id` cache-affinity header from `options.sessionId` when caching is enabled. Default: true. */\n\tsendSessionIdHeader?: boolean;\n\t/** Whether the provider supports `prompt_cache_retention: \"24h\"`. Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n}\n\n/** Compatibility settings for Anthropic Messages-compatible APIs. */\nexport interface AnthropicMessagesCompat {\n\t/**\n\t * Whether the provider accepts per-tool `eager_input_streaming`.\n\t * When false, the Anthropic provider omits `tools[].eager_input_streaming`\n\t * and sends the legacy `fine-grained-tool-streaming-2025-05-14` beta header\n\t * for tool-enabled requests.\n\t * Default: true.\n\t */\n\tsupportsEagerToolInputStreaming?: boolean;\n\t/** Whether the provider supports Anthropic long cache retention (`cache_control.ttl: \"1h\"`). Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n\t/**\n\t * Whether to send the `x-session-affinity` header from `options.sessionId`\n\t * when caching is enabled. Required for providers like Fireworks that use\n\t * session affinity for prompt cache routing (requests to the same replica\n\t * maximize cache hits).\n\t * Default: false.\n\t */\n\tsendSessionAffinityHeaders?: boolean;\n\t/**\n\t * Whether the provider supports Anthropic-style `cache_control` markers on\n\t * tool definitions. When false, `cache_control` is omitted from tool params.\n\t * Some Anthropic-compatible providers (e.g., Fireworks) do not support this\n\t * field on tools and may reject or ignore it.\n\t * Default: true.\n\t */\n\tsupportsCacheControlOnTools?: boolean;\n\t/**\n\t * Whether the model accepts the Anthropic `temperature` request field.\n\t * Claude Opus 4.7+ rejects non-default temperature values.\n\t * Default: true.\n\t */\n\tsupportsTemperature?: boolean;\n\t/**\n\t * Whether to force adaptive thinking (`thinking.type: \"adaptive\"` plus\n\t * `output_config.effort`) regardless of the model id. Built-in models that\n\t * require adaptive thinking set this in generated metadata. Custom\n\t * Anthropic-compatible providers can set this to `true` for any model whose\n\t * upstream requires the adaptive format. Set to `false` to\n\t * opt out on overridden built-in models.\n\t * Default: false.\n\t */\n\tforceAdaptiveThinking?: boolean;\n\t/** Whether to replay empty thinking signatures as `signature: \"\"` instead of converting thinking to text. Default: false. */\n\tallowEmptySignature?: boolean;\n}\n\n/**\n * OpenRouter provider routing preferences.\n * Controls which upstream providers OpenRouter routes requests to.\n * Sent as the `provider` field in the OpenRouter API request body.\n * @see https://openrouter.ai/docs/guides/routing/provider-selection\n */\nexport interface OpenRouterRouting {\n\t/** Whether to allow backup providers to serve requests. Default: true. */\n\tallow_fallbacks?: boolean;\n\t/** Whether to filter providers to only those that support all parameters in the request. Default: false. */\n\trequire_parameters?: boolean;\n\t/** Data collection setting. \"allow\" (default): allow providers that may store/train on data. \"deny\": only use providers that don't collect user data. */\n\tdata_collection?: \"deny\" | \"allow\";\n\t/** Whether to restrict routing to only ZDR (Zero Data Retention) endpoints. */\n\tzdr?: boolean;\n\t/** Whether to restrict routing to only models that allow text distillation. */\n\tenforce_distillable_text?: boolean;\n\t/** An ordered list of provider names/slugs to try in sequence, falling back to the next if unavailable. */\n\torder?: string[];\n\t/** List of provider names/slugs to exclusively allow for this request. */\n\tonly?: string[];\n\t/** List of provider names/slugs to skip for this request. */\n\tignore?: string[];\n\t/** A list of quantization levels to filter providers by (e.g., [\"fp16\", \"bf16\", \"fp8\", \"fp6\", \"int8\", \"int4\", \"fp4\", \"fp32\"]). */\n\tquantizations?: string[];\n\t/** Sorting strategy. Can be a string (e.g., \"price\", \"throughput\", \"latency\") or an object with `by` and `partition`. */\n\tsort?:\n\t\t| string\n\t\t| {\n\t\t\t\t/** The sorting metric: \"price\", \"throughput\", \"latency\". */\n\t\t\t\tby?: string;\n\t\t\t\t/** Partitioning strategy: \"model\" (default) or \"none\". */\n\t\t\t\tpartition?: string | null;\n\t\t  };\n\t/** Maximum price per million tokens (USD). */\n\tmax_price?: {\n\t\t/** Price per million prompt tokens. */\n\t\tprompt?: number | string;\n\t\t/** Price per million completion tokens. */\n\t\tcompletion?: number | string;\n\t\t/** Price per image. */\n\t\timage?: number | string;\n\t\t/** Price per audio unit. */\n\t\taudio?: number | string;\n\t\t/** Price per request. */\n\t\trequest?: number | string;\n\t};\n\t/** Preferred minimum throughput (tokens/second). Can be a number (applies to p50) or an object with percentile-specific cutoffs. */\n\tpreferred_min_throughput?:\n\t\t| number\n\t\t| {\n\t\t\t\t/** Minimum tokens/second at the 50th percentile. */\n\t\t\t\tp50?: number;\n\t\t\t\t/** Minimum tokens/second at the 75th percentile. */\n\t\t\t\tp75?: number;\n\t\t\t\t/** Minimum tokens/second at the 90th percentile. */\n\t\t\t\tp90?: number;\n\t\t\t\t/** Minimum tokens/second at the 99th percentile. */\n\t\t\t\tp99?: number;\n\t\t  };\n\t/** Preferred maximum latency (seconds). Can be a number (applies to p50) or an object with percentile-specific cutoffs. */\n\tpreferred_max_latency?:\n\t\t| number\n\t\t| {\n\t\t\t\t/** Maximum latency in seconds at the 50th percentile. */\n\t\t\t\tp50?: number;\n\t\t\t\t/** Maximum latency in seconds at the 75th percentile. */\n\t\t\t\tp75?: number;\n\t\t\t\t/** Maximum latency in seconds at the 90th percentile. */\n\t\t\t\tp90?: number;\n\t\t\t\t/** Maximum latency in seconds at the 99th percentile. */\n\t\t\t\tp99?: number;\n\t\t  };\n}\n\n/**\n * Vercel AI Gateway routing preferences.\n * Controls which upstream providers the gateway routes requests to.\n * @see https://vercel.com/docs/ai-gateway/models-and-providers/provider-options\n */\nexport interface VercelGatewayRouting {\n\t/** List of provider slugs to exclusively use for this request (e.g., [\"bedrock\", \"anthropic\"]). */\n\tonly?: string[];\n\t/** List of provider slugs to try in order (e.g., [\"anthropic\", \"openai\"]). */\n\torder?: string[];\n}\n\nexport interface ModelCostRates {\n\tinput: number; // $/million tokens\n\toutput: number; // $/million tokens\n\tcacheRead: number; // $/million tokens\n\tcacheWrite: number; // $/million tokens\n}\n\nexport interface ModelCostTier extends ModelCostRates {\n\t/** Use this tier for requests whose total input usage exceeds this token count. */\n\tinputTokensAbove: number;\n}\n\nexport interface ModelCost extends ModelCostRates {\n\t/** Request-wide pricing tiers. The highest matching input threshold applies to the full request. */\n\ttiers?: ModelCostTier[];\n}\n\n// Model interface for the unified model system\nexport interface Model<TApi extends Api> {\n\tid: string;\n\tname: string;\n\tapi: TApi;\n\tprovider: ProviderId;\n\tbaseUrl: string;\n\treasoning: boolean;\n\t/**\n\t * Maps pi thinking levels to provider/model-specific values.\n\t * Missing keys use provider defaults. null marks a level as unsupported.\n\t */\n\tthinkingLevelMap?: ThinkingLevelMap;\n\tinput: (\"text\" | \"image\")[];\n\tcost: ModelCost;\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/** Compatibility overrides for OpenAI-compatible APIs. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\"\n\t\t? OpenAICompletionsCompat\n\t\t: TApi extends \"openai-responses\"\n\t\t\t? OpenAIResponsesCompat\n\t\t\t: TApi extends \"anthropic-messages\"\n\t\t\t\t? AnthropicMessagesCompat\n\t\t\t\t: never;\n}\n\nexport interface ImagesModel<TApi extends ImagesApi>\n\textends Omit<Model<Api>, \"api\" | \"provider\" | \"reasoning\" | \"contextWindow\" | \"maxTokens\" | \"compat\"> {\n\tapi: TApi;\n\tprovider: ImagesProviderId;\n\toutput: (\"text\" | \"image\")[];\n}\n"]}

package/dist/utils/estimate.d.ts

@@ -2,11 +2,11 @@ import type { Context, ImageContent, Message, TextContent, Usage } from "../type
 export interface ContextUsageEstimate {
     /** Estimated total context tokens. */
     tokens: number;
-    /** Tokens reported by the most recent assistant usage block. */
+    /** Tokens reported by the most recent applicable assistant usage block. */
     usageTokens: number;
-    /** Estimated tokens after the most recent assistant usage block. */
+    /** Estimated tokens after the most recent applicable assistant usage block. */
     trailingTokens: number;
-    /** Index of the message that provided usage, or null when none exists. */
+    /** Index of the applicable message that provided usage, or null when none exists. */
     lastUsageIndex: number | null;
 }
 export declare function calculateContextTokens(usage: Usage): number;

package/dist/utils/estimate.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"estimate.d.ts","sourceRoot":"","sources":["../../src/utils/estimate.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAoB,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AAExG,MAAM,WAAW,oBAAoB;IACpC,sCAAsC;IACtC,MAAM,EAAE,MAAM,CAAC;IACf,gEAAgE;IAChE,WAAW,EAAE,MAAM,CAAC;IACpB,oEAAoE;IACpE,cAAc,EAAE,MAAM,CAAC;IACvB,0EAA0E;IAC1E,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAKD,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,KAAK,GAAG,MAAM,CAE3D;AAkBD,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEvD;AAED,wBAAgB,iCAAiC,CAAC,OAAO,EAAE,MAAM,GAAG,KAAK,CAAC,WAAW,GAAG,YAAY,CAAC,GAAG,MAAM,CAE7G;AAED,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,OAAO,GAAG,MAAM,CAgB9D;AAiCD,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,OAAO,GAAG,SAAS,OAAO,EAAE,GAAG,oBAAoB,CAiBjG","sourcesContent":["import type { AssistantMessage, Context, ImageContent, Message, TextContent, Usage } from \"../types.ts\";\n\nexport interface ContextUsageEstimate {\n\t/** Estimated total context tokens. */\n\ttokens: number;\n\t/** Tokens reported by the most recent assistant usage block. */\n\tusageTokens: number;\n\t/** Estimated tokens after the most recent assistant usage block. */\n\ttrailingTokens: number;\n\t/** Index of the message that provided usage, or null when none exists. */\n\tlastUsageIndex: number | null;\n}\n\nconst CHARS_PER_TOKEN = 4;\nconst ESTIMATED_IMAGE_CHARS = 4800;\n\nexport function calculateContextTokens(usage: Usage): number {\n\treturn usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite;\n}\n\nfunction safeJsonStringify(value: unknown): string {\n\ttry {\n\t\treturn JSON.stringify(value) ?? \"undefined\";\n\t} catch {\n\t\treturn \"[unserializable]\";\n\t}\n}\n\nfunction estimateTextAndImageContentChars(content: string | Array<TextContent | ImageContent>): number {\n\tif (typeof content === \"string\") return content.length;\n\n\tlet chars = 0;\n\tfor (const block of content) chars += block.type === \"text\" ? block.text.length : ESTIMATED_IMAGE_CHARS;\n\treturn chars;\n}\n\nexport function estimateTextTokens(text: string): number {\n\treturn Math.ceil(text.length / CHARS_PER_TOKEN);\n}\n\nexport function estimateTextAndImageContentTokens(content: string | Array<TextContent | ImageContent>): number {\n\treturn Math.ceil(estimateTextAndImageContentChars(content) / CHARS_PER_TOKEN);\n}\n\nexport function estimateMessageTokens(message: Message): number {\n\tlet chars = 0;\n\n\tif (message.role === \"user\") return estimateTextAndImageContentTokens(message.content);\n\tif (message.role === \"toolResult\") return estimateTextAndImageContentTokens(message.content);\n\n\tfor (const block of message.content) {\n\t\tif (block.type === \"text\") {\n\t\t\tchars += block.text.length;\n\t\t} else if (block.type === \"thinking\") {\n\t\t\tchars += block.thinking.length;\n\t\t} else {\n\t\t\tchars += block.name.length + safeJsonStringify(block.arguments).length;\n\t\t}\n\t}\n\treturn Math.ceil(chars / CHARS_PER_TOKEN);\n}\n\nfunction getLastAssistantUsageInfo(messages: readonly Message[]): { usage: Usage; index: number } | undefined {\n\tfor (let i = messages.length - 1; i >= 0; i--) {\n\t\tconst message = messages[i];\n\t\tif (message.role !== \"assistant\") continue;\n\t\tconst assistant = message as AssistantMessage;\n\t\tif (assistant.stopReason === \"aborted\" || assistant.stopReason === \"error\") continue;\n\t\tif (calculateContextTokens(assistant.usage) > 0) return { usage: assistant.usage, index: i };\n\t}\n\treturn undefined;\n}\n\nfunction estimateMessages(messages: readonly Message[]): ContextUsageEstimate {\n\tconst usageInfo = getLastAssistantUsageInfo(messages);\n\tif (usageInfo) {\n\t\tconst usageTokens = calculateContextTokens(usageInfo.usage);\n\t\tlet trailingTokens = 0;\n\t\tfor (let i = usageInfo.index + 1; i < messages.length; i++) {\n\t\t\ttrailingTokens += estimateMessageTokens(messages[i]);\n\t\t}\n\t\treturn { tokens: usageTokens + trailingTokens, usageTokens, trailingTokens, lastUsageIndex: usageInfo.index };\n\t}\n\n\tlet tokens = 0;\n\tfor (const message of messages) tokens += estimateMessageTokens(message);\n\treturn { tokens, usageTokens: 0, trailingTokens: tokens, lastUsageIndex: null };\n}\n\nfunction isMessageArray(value: Context | readonly Message[]): value is readonly Message[] {\n\treturn Array.isArray(value);\n}\n\nexport function estimateContextTokens(context: Context | readonly Message[]): ContextUsageEstimate {\n\tif (isMessageArray(context)) return estimateMessages(context);\n\n\tconst estimate = estimateMessages(context.messages);\n\tif (estimate.lastUsageIndex !== null) return estimate;\n\n\tlet prefixTokens = context.systemPrompt ? estimateTextTokens(context.systemPrompt) : 0;\n\tif (context.tools && context.tools.length > 0) {\n\t\tprefixTokens += estimateTextTokens(safeJsonStringify(context.tools));\n\t}\n\n\treturn {\n\t\ttokens: estimate.tokens + prefixTokens,\n\t\tusageTokens: estimate.usageTokens,\n\t\ttrailingTokens: estimate.trailingTokens + prefixTokens,\n\t\tlastUsageIndex: estimate.lastUsageIndex,\n\t};\n}\n"]}
+{"version":3,"file":"estimate.d.ts","sourceRoot":"","sources":["../../src/utils/estimate.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAoB,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AAExG,MAAM,WAAW,oBAAoB;IACpC,sCAAsC;IACtC,MAAM,EAAE,MAAM,CAAC;IACf,2EAA2E;IAC3E,WAAW,EAAE,MAAM,CAAC;IACpB,+EAA+E;IAC/E,cAAc,EAAE,MAAM,CAAC;IACvB,qFAAqF;IACrF,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAKD,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,KAAK,GAAG,MAAM,CAE3D;AAkBD,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEvD;AAED,wBAAgB,iCAAiC,CAAC,OAAO,EAAE,MAAM,GAAG,KAAK,CAAC,WAAW,GAAG,YAAY,CAAC,GAAG,MAAM,CAE7G;AAED,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,OAAO,GAAG,MAAM,CAgB9D;AAgDD,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,OAAO,GAAG,SAAS,OAAO,EAAE,GAAG,oBAAoB,CAiBjG","sourcesContent":["import type { AssistantMessage, Context, ImageContent, Message, TextContent, Usage } from \"../types.ts\";\n\nexport interface ContextUsageEstimate {\n\t/** Estimated total context tokens. */\n\ttokens: number;\n\t/** Tokens reported by the most recent applicable assistant usage block. */\n\tusageTokens: number;\n\t/** Estimated tokens after the most recent applicable assistant usage block. */\n\ttrailingTokens: number;\n\t/** Index of the applicable message that provided usage, or null when none exists. */\n\tlastUsageIndex: number | null;\n}\n\nconst CHARS_PER_TOKEN = 4;\nconst ESTIMATED_IMAGE_CHARS = 4800;\n\nexport function calculateContextTokens(usage: Usage): number {\n\treturn usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite;\n}\n\nfunction safeJsonStringify(value: unknown): string {\n\ttry {\n\t\treturn JSON.stringify(value) ?? \"undefined\";\n\t} catch {\n\t\treturn \"[unserializable]\";\n\t}\n}\n\nfunction estimateTextAndImageContentChars(content: string | Array<TextContent | ImageContent>): number {\n\tif (typeof content === \"string\") return content.length;\n\n\tlet chars = 0;\n\tfor (const block of content) chars += block.type === \"text\" ? block.text.length : ESTIMATED_IMAGE_CHARS;\n\treturn chars;\n}\n\nexport function estimateTextTokens(text: string): number {\n\treturn Math.ceil(text.length / CHARS_PER_TOKEN);\n}\n\nexport function estimateTextAndImageContentTokens(content: string | Array<TextContent | ImageContent>): number {\n\treturn Math.ceil(estimateTextAndImageContentChars(content) / CHARS_PER_TOKEN);\n}\n\nexport function estimateMessageTokens(message: Message): number {\n\tlet chars = 0;\n\n\tif (message.role === \"user\") return estimateTextAndImageContentTokens(message.content);\n\tif (message.role === \"toolResult\") return estimateTextAndImageContentTokens(message.content);\n\n\tfor (const block of message.content) {\n\t\tif (block.type === \"text\") {\n\t\t\tchars += block.text.length;\n\t\t} else if (block.type === \"thinking\") {\n\t\t\tchars += block.thinking.length;\n\t\t} else {\n\t\t\tchars += block.name.length + safeJsonStringify(block.arguments).length;\n\t\t}\n\t}\n\treturn Math.ceil(chars / CHARS_PER_TOKEN);\n}\n\nfunction getLastAssistantUsageInfo(messages: readonly Message[]): { usage: Usage; index: number } | undefined {\n\tlet latestPrefixTimestamp = Number.NEGATIVE_INFINITY;\n\tlet usageInfo: { usage: Usage; index: number } | undefined;\n\n\tfor (let i = 0; i < messages.length; i++) {\n\t\tconst message = messages[i];\n\t\tif (message.role === \"assistant\") {\n\t\t\tconst assistant = message as AssistantMessage;\n\t\t\t// A newer prefix message was inserted after this response (for example, a\n\t\t\t// compaction summary), so its usage cannot describe the current prefix.\n\t\t\tconst usageAppliesToPrefix = assistant.timestamp >= latestPrefixTimestamp;\n\t\t\tif (\n\t\t\t\tusageAppliesToPrefix &&\n\t\t\t\tassistant.stopReason !== \"aborted\" &&\n\t\t\t\tassistant.stopReason !== \"error\" &&\n\t\t\t\tcalculateContextTokens(assistant.usage) > 0\n\t\t\t) {\n\t\t\t\tusageInfo = { usage: assistant.usage, index: i };\n\t\t\t}\n\t\t}\n\t\tlatestPrefixTimestamp = Math.max(latestPrefixTimestamp, message.timestamp);\n\t}\n\n\treturn usageInfo;\n}\n\nfunction estimateMessages(messages: readonly Message[]): ContextUsageEstimate {\n\tconst usageInfo = getLastAssistantUsageInfo(messages);\n\tif (usageInfo) {\n\t\tconst usageTokens = calculateContextTokens(usageInfo.usage);\n\t\tlet trailingTokens = 0;\n\t\tfor (let i = usageInfo.index + 1; i < messages.length; i++) {\n\t\t\ttrailingTokens += estimateMessageTokens(messages[i]);\n\t\t}\n\t\treturn { tokens: usageTokens + trailingTokens, usageTokens, trailingTokens, lastUsageIndex: usageInfo.index };\n\t}\n\n\tlet tokens = 0;\n\tfor (const message of messages) tokens += estimateMessageTokens(message);\n\treturn { tokens, usageTokens: 0, trailingTokens: tokens, lastUsageIndex: null };\n}\n\nfunction isMessageArray(value: Context | readonly Message[]): value is readonly Message[] {\n\treturn Array.isArray(value);\n}\n\nexport function estimateContextTokens(context: Context | readonly Message[]): ContextUsageEstimate {\n\tif (isMessageArray(context)) return estimateMessages(context);\n\n\tconst estimate = estimateMessages(context.messages);\n\tif (estimate.lastUsageIndex !== null) return estimate;\n\n\tlet prefixTokens = context.systemPrompt ? estimateTextTokens(context.systemPrompt) : 0;\n\tif (context.tools && context.tools.length > 0) {\n\t\tprefixTokens += estimateTextTokens(safeJsonStringify(context.tools));\n\t}\n\n\treturn {\n\t\ttokens: estimate.tokens + prefixTokens,\n\t\tusageTokens: estimate.usageTokens,\n\t\ttrailingTokens: estimate.trailingTokens + prefixTokens,\n\t\tlastUsageIndex: estimate.lastUsageIndex,\n\t};\n}\n"]}

package/dist/utils/estimate.js

@@ -45,17 +45,25 @@ export function estimateMessageTokens(message) {
     return Math.ceil(chars / CHARS_PER_TOKEN);
 }
 function getLastAssistantUsageInfo(messages) {
-    for (let i = messages.length - 1; i >= 0; i--) {
+    let latestPrefixTimestamp = Number.NEGATIVE_INFINITY;
+    let usageInfo;
+    for (let i = 0; i < messages.length; i++) {
         const message = messages[i];
-        if (message.role !== "assistant")
-            continue;
-        const assistant = message;
-        if (assistant.stopReason === "aborted" || assistant.stopReason === "error")
-            continue;
-        if (calculateContextTokens(assistant.usage) > 0)
-            return { usage: assistant.usage, index: i };
+        if (message.role === "assistant") {
+            const assistant = message;
+            // A newer prefix message was inserted after this response (for example, a
+            // compaction summary), so its usage cannot describe the current prefix.
+            const usageAppliesToPrefix = assistant.timestamp >= latestPrefixTimestamp;
+            if (usageAppliesToPrefix &&
+                assistant.stopReason !== "aborted" &&
+                assistant.stopReason !== "error" &&
+                calculateContextTokens(assistant.usage) > 0) {
+                usageInfo = { usage: assistant.usage, index: i };
+            }
+        }
+        latestPrefixTimestamp = Math.max(latestPrefixTimestamp, message.timestamp);
     }
-    return undefined;
+    return usageInfo;
 }
 function estimateMessages(messages) {
     const usageInfo = getLastAssistantUsageInfo(messages);

package/dist/utils/estimate.js.map

@@ -1 +1 @@
-{"version":3,"file":"estimate.js","sourceRoot":"","sources":["../../src/utils/estimate.ts"],"names":[],"mappings":"AAaA,MAAM,eAAe,GAAG,CAAC,CAAC;AAC1B,MAAM,qBAAqB,GAAG,IAAI,CAAC;AAEnC,MAAM,UAAU,sBAAsB,CAAC,KAAY,EAAU;IAC5D,OAAO,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC;AAAA,CAC5F;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAU;IAClD,IAAI,CAAC;QACJ,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,WAAW,CAAC;IAC7C,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,kBAAkB,CAAC;IAC3B,CAAC;AAAA,CACD;AAED,SAAS,gCAAgC,CAAC,OAAmD,EAAU;IACtG,IAAI,OAAO,OAAO,KAAK,QAAQ;QAAE,OAAO,OAAO,CAAC,MAAM,CAAC;IAEvD,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,KAAK,IAAI,OAAO;QAAE,KAAK,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,qBAAqB,CAAC;IACxG,OAAO,KAAK,CAAC;AAAA,CACb;AAED,MAAM,UAAU,kBAAkB,CAAC,IAAY,EAAU;IACxD,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,eAAe,CAAC,CAAC;AAAA,CAChD;AAED,MAAM,UAAU,iCAAiC,CAAC,OAAmD,EAAU;IAC9G,OAAO,IAAI,CAAC,IAAI,CAAC,gCAAgC,CAAC,OAAO,CAAC,GAAG,eAAe,CAAC,CAAC;AAAA,CAC9E;AAED,MAAM,UAAU,qBAAqB,CAAC,OAAgB,EAAU;IAC/D,IAAI,KAAK,GAAG,CAAC,CAAC;IAEd,IAAI,OAAO,CAAC,IAAI,KAAK,MAAM;QAAE,OAAO,iCAAiC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IACvF,IAAI,OAAO,CAAC,IAAI,KAAK,YAAY;QAAE,OAAO,iCAAiC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IAE7F,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;QACrC,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;YAC3B,KAAK,IAAI,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC;QAC5B,CAAC;aAAM,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;YACtC,KAAK,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC;QAChC,CAAC;aAAM,CAAC;YACP,KAAK,IAAI,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,iBAAiB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC;QACxE,CAAC;IACF,CAAC;IACD,OAAO,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,eAAe,CAAC,CAAC;AAAA,CAC1C;AAED,SAAS,yBAAyB,CAAC,QAA4B,EAA+C;IAC7G,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC/C,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;QAC5B,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW;YAAE,SAAS;QAC3C,MAAM,SAAS,GAAG,OAA2B,CAAC;QAC9C,IAAI,SAAS,CAAC,UAAU,KAAK,SAAS,IAAI,SAAS,CAAC,UAAU,KAAK,OAAO;YAAE,SAAS;QACrF,IAAI,sBAAsB,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC;YAAE,OAAO,EAAE,KAAK,EAAE,SAAS,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;IAC9F,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,SAAS,gBAAgB,CAAC,QAA4B,EAAwB;IAC7E,MAAM,SAAS,GAAG,yBAAyB,CAAC,QAAQ,CAAC,CAAC;IACtD,IAAI,SAAS,EAAE,CAAC;QACf,MAAM,WAAW,GAAG,sBAAsB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;QAC5D,IAAI,cAAc,GAAG,CAAC,CAAC;QACvB,KAAK,IAAI,CAAC,GAAG,SAAS,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5D,cAAc,IAAI,qBAAqB,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;QACtD,CAAC;QACD,OAAO,EAAE,MAAM,EAAE,WAAW,GAAG,cAAc,EAAE,WAAW,EAAE,cAAc,EAAE,cAAc,EAAE,SAAS,CAAC,KAAK,EAAE,CAAC;IAC/G,CAAC;IAED,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,KAAK,MAAM,OAAO,IAAI,QAAQ;QAAE,MAAM,IAAI,qBAAqB,CAAC,OAAO,CAAC,CAAC;IACzE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC,EAAE,cAAc,EAAE,MAAM,EAAE,cAAc,EAAE,IAAI,EAAE,CAAC;AAAA,CAChF;AAED,SAAS,cAAc,CAAC,KAAmC,EAA+B;IACzF,OAAO,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;AAAA,CAC5B;AAED,MAAM,UAAU,qBAAqB,CAAC,OAAqC,EAAwB;IAClG,IAAI,cAAc,CAAC,OAAO,CAAC;QAAE,OAAO,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAE9D,MAAM,QAAQ,GAAG,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;IACpD,IAAI,QAAQ,CAAC,cAAc,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IAEtD,IAAI,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,kBAAkB,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACvF,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC/C,YAAY,IAAI,kBAAkB,CAAC,iBAAiB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC;IACtE,CAAC;IAED,OAAO;QACN,MAAM,EAAE,QAAQ,CAAC,MAAM,GAAG,YAAY;QACtC,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,cAAc,EAAE,QAAQ,CAAC,cAAc,GAAG,YAAY;QACtD,cAAc,EAAE,QAAQ,CAAC,cAAc;KACvC,CAAC;AAAA,CACF","sourcesContent":["import type { AssistantMessage, Context, ImageContent, Message, TextContent, Usage } from \"../types.ts\";\n\nexport interface ContextUsageEstimate {\n\t/** Estimated total context tokens. */\n\ttokens: number;\n\t/** Tokens reported by the most recent assistant usage block. */\n\tusageTokens: number;\n\t/** Estimated tokens after the most recent assistant usage block. */\n\ttrailingTokens: number;\n\t/** Index of the message that provided usage, or null when none exists. */\n\tlastUsageIndex: number | null;\n}\n\nconst CHARS_PER_TOKEN = 4;\nconst ESTIMATED_IMAGE_CHARS = 4800;\n\nexport function calculateContextTokens(usage: Usage): number {\n\treturn usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite;\n}\n\nfunction safeJsonStringify(value: unknown): string {\n\ttry {\n\t\treturn JSON.stringify(value) ?? \"undefined\";\n\t} catch {\n\t\treturn \"[unserializable]\";\n\t}\n}\n\nfunction estimateTextAndImageContentChars(content: string | Array<TextContent | ImageContent>): number {\n\tif (typeof content === \"string\") return content.length;\n\n\tlet chars = 0;\n\tfor (const block of content) chars += block.type === \"text\" ? block.text.length : ESTIMATED_IMAGE_CHARS;\n\treturn chars;\n}\n\nexport function estimateTextTokens(text: string): number {\n\treturn Math.ceil(text.length / CHARS_PER_TOKEN);\n}\n\nexport function estimateTextAndImageContentTokens(content: string | Array<TextContent | ImageContent>): number {\n\treturn Math.ceil(estimateTextAndImageContentChars(content) / CHARS_PER_TOKEN);\n}\n\nexport function estimateMessageTokens(message: Message): number {\n\tlet chars = 0;\n\n\tif (message.role === \"user\") return estimateTextAndImageContentTokens(message.content);\n\tif (message.role === \"toolResult\") return estimateTextAndImageContentTokens(message.content);\n\n\tfor (const block of message.content) {\n\t\tif (block.type === \"text\") {\n\t\t\tchars += block.text.length;\n\t\t} else if (block.type === \"thinking\") {\n\t\t\tchars += block.thinking.length;\n\t\t} else {\n\t\t\tchars += block.name.length + safeJsonStringify(block.arguments).length;\n\t\t}\n\t}\n\treturn Math.ceil(chars / CHARS_PER_TOKEN);\n}\n\nfunction getLastAssistantUsageInfo(messages: readonly Message[]): { usage: Usage; index: number } | undefined {\n\tfor (let i = messages.length - 1; i >= 0; i--) {\n\t\tconst message = messages[i];\n\t\tif (message.role !== \"assistant\") continue;\n\t\tconst assistant = message as AssistantMessage;\n\t\tif (assistant.stopReason === \"aborted\" || assistant.stopReason === \"error\") continue;\n\t\tif (calculateContextTokens(assistant.usage) > 0) return { usage: assistant.usage, index: i };\n\t}\n\treturn undefined;\n}\n\nfunction estimateMessages(messages: readonly Message[]): ContextUsageEstimate {\n\tconst usageInfo = getLastAssistantUsageInfo(messages);\n\tif (usageInfo) {\n\t\tconst usageTokens = calculateContextTokens(usageInfo.usage);\n\t\tlet trailingTokens = 0;\n\t\tfor (let i = usageInfo.index + 1; i < messages.length; i++) {\n\t\t\ttrailingTokens += estimateMessageTokens(messages[i]);\n\t\t}\n\t\treturn { tokens: usageTokens + trailingTokens, usageTokens, trailingTokens, lastUsageIndex: usageInfo.index };\n\t}\n\n\tlet tokens = 0;\n\tfor (const message of messages) tokens += estimateMessageTokens(message);\n\treturn { tokens, usageTokens: 0, trailingTokens: tokens, lastUsageIndex: null };\n}\n\nfunction isMessageArray(value: Context | readonly Message[]): value is readonly Message[] {\n\treturn Array.isArray(value);\n}\n\nexport function estimateContextTokens(context: Context | readonly Message[]): ContextUsageEstimate {\n\tif (isMessageArray(context)) return estimateMessages(context);\n\n\tconst estimate = estimateMessages(context.messages);\n\tif (estimate.lastUsageIndex !== null) return estimate;\n\n\tlet prefixTokens = context.systemPrompt ? estimateTextTokens(context.systemPrompt) : 0;\n\tif (context.tools && context.tools.length > 0) {\n\t\tprefixTokens += estimateTextTokens(safeJsonStringify(context.tools));\n\t}\n\n\treturn {\n\t\ttokens: estimate.tokens + prefixTokens,\n\t\tusageTokens: estimate.usageTokens,\n\t\ttrailingTokens: estimate.trailingTokens + prefixTokens,\n\t\tlastUsageIndex: estimate.lastUsageIndex,\n\t};\n}\n"]}
+{"version":3,"file":"estimate.js","sourceRoot":"","sources":["../../src/utils/estimate.ts"],"names":[],"mappings":"AAaA,MAAM,eAAe,GAAG,CAAC,CAAC;AAC1B,MAAM,qBAAqB,GAAG,IAAI,CAAC;AAEnC,MAAM,UAAU,sBAAsB,CAAC,KAAY,EAAU;IAC5D,OAAO,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC;AAAA,CAC5F;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAU;IAClD,IAAI,CAAC;QACJ,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,WAAW,CAAC;IAC7C,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,kBAAkB,CAAC;IAC3B,CAAC;AAAA,CACD;AAED,SAAS,gCAAgC,CAAC,OAAmD,EAAU;IACtG,IAAI,OAAO,OAAO,KAAK,QAAQ;QAAE,OAAO,OAAO,CAAC,MAAM,CAAC;IAEvD,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,KAAK,IAAI,OAAO;QAAE,KAAK,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,qBAAqB,CAAC;IACxG,OAAO,KAAK,CAAC;AAAA,CACb;AAED,MAAM,UAAU,kBAAkB,CAAC,IAAY,EAAU;IACxD,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,eAAe,CAAC,CAAC;AAAA,CAChD;AAED,MAAM,UAAU,iCAAiC,CAAC,OAAmD,EAAU;IAC9G,OAAO,IAAI,CAAC,IAAI,CAAC,gCAAgC,CAAC,OAAO,CAAC,GAAG,eAAe,CAAC,CAAC;AAAA,CAC9E;AAED,MAAM,UAAU,qBAAqB,CAAC,OAAgB,EAAU;IAC/D,IAAI,KAAK,GAAG,CAAC,CAAC;IAEd,IAAI,OAAO,CAAC,IAAI,KAAK,MAAM;QAAE,OAAO,iCAAiC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IACvF,IAAI,OAAO,CAAC,IAAI,KAAK,YAAY;QAAE,OAAO,iCAAiC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IAE7F,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;QACrC,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;YAC3B,KAAK,IAAI,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC;QAC5B,CAAC;aAAM,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;YACtC,KAAK,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC;QAChC,CAAC;aAAM,CAAC;YACP,KAAK,IAAI,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,iBAAiB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC;QACxE,CAAC;IACF,CAAC;IACD,OAAO,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,eAAe,CAAC,CAAC;AAAA,CAC1C;AAED,SAAS,yBAAyB,CAAC,QAA4B,EAA+C;IAC7G,IAAI,qBAAqB,GAAG,MAAM,CAAC,iBAAiB,CAAC;IACrD,IAAI,SAAsD,CAAC;IAE3D,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC1C,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;QAC5B,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YAClC,MAAM,SAAS,GAAG,OAA2B,CAAC;YAC9C,0EAA0E;YAC1E,wEAAwE;YACxE,MAAM,oBAAoB,GAAG,SAAS,CAAC,SAAS,IAAI,qBAAqB,CAAC;YAC1E,IACC,oBAAoB;gBACpB,SAAS,CAAC,UAAU,KAAK,SAAS;gBAClC,SAAS,CAAC,UAAU,KAAK,OAAO;gBAChC,sBAAsB,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,EAC1C,CAAC;gBACF,SAAS,GAAG,EAAE,KAAK,EAAE,SAAS,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;YAClD,CAAC;QACF,CAAC;QACD,qBAAqB,GAAG,IAAI,CAAC,GAAG,CAAC,qBAAqB,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC;IAC5E,CAAC;IAED,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,SAAS,gBAAgB,CAAC,QAA4B,EAAwB;IAC7E,MAAM,SAAS,GAAG,yBAAyB,CAAC,QAAQ,CAAC,CAAC;IACtD,IAAI,SAAS,EAAE,CAAC;QACf,MAAM,WAAW,GAAG,sBAAsB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;QAC5D,IAAI,cAAc,GAAG,CAAC,CAAC;QACvB,KAAK,IAAI,CAAC,GAAG,SAAS,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5D,cAAc,IAAI,qBAAqB,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;QACtD,CAAC;QACD,OAAO,EAAE,MAAM,EAAE,WAAW,GAAG,cAAc,EAAE,WAAW,EAAE,cAAc,EAAE,cAAc,EAAE,SAAS,CAAC,KAAK,EAAE,CAAC;IAC/G,CAAC;IAED,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,KAAK,MAAM,OAAO,IAAI,QAAQ;QAAE,MAAM,IAAI,qBAAqB,CAAC,OAAO,CAAC,CAAC;IACzE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC,EAAE,cAAc,EAAE,MAAM,EAAE,cAAc,EAAE,IAAI,EAAE,CAAC;AAAA,CAChF;AAED,SAAS,cAAc,CAAC,KAAmC,EAA+B;IACzF,OAAO,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;AAAA,CAC5B;AAED,MAAM,UAAU,qBAAqB,CAAC,OAAqC,EAAwB;IAClG,IAAI,cAAc,CAAC,OAAO,CAAC;QAAE,OAAO,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAE9D,MAAM,QAAQ,GAAG,gBAAgB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;IACpD,IAAI,QAAQ,CAAC,cAAc,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IAEtD,IAAI,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,kBAAkB,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACvF,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC/C,YAAY,IAAI,kBAAkB,CAAC,iBAAiB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC;IACtE,CAAC;IAED,OAAO;QACN,MAAM,EAAE,QAAQ,CAAC,MAAM,GAAG,YAAY;QACtC,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,cAAc,EAAE,QAAQ,CAAC,cAAc,GAAG,YAAY;QACtD,cAAc,EAAE,QAAQ,CAAC,cAAc;KACvC,CAAC;AAAA,CACF","sourcesContent":["import type { AssistantMessage, Context, ImageContent, Message, TextContent, Usage } from \"../types.ts\";\n\nexport interface ContextUsageEstimate {\n\t/** Estimated total context tokens. */\n\ttokens: number;\n\t/** Tokens reported by the most recent applicable assistant usage block. */\n\tusageTokens: number;\n\t/** Estimated tokens after the most recent applicable assistant usage block. */\n\ttrailingTokens: number;\n\t/** Index of the applicable message that provided usage, or null when none exists. */\n\tlastUsageIndex: number | null;\n}\n\nconst CHARS_PER_TOKEN = 4;\nconst ESTIMATED_IMAGE_CHARS = 4800;\n\nexport function calculateContextTokens(usage: Usage): number {\n\treturn usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite;\n}\n\nfunction safeJsonStringify(value: unknown): string {\n\ttry {\n\t\treturn JSON.stringify(value) ?? \"undefined\";\n\t} catch {\n\t\treturn \"[unserializable]\";\n\t}\n}\n\nfunction estimateTextAndImageContentChars(content: string | Array<TextContent | ImageContent>): number {\n\tif (typeof content === \"string\") return content.length;\n\n\tlet chars = 0;\n\tfor (const block of content) chars += block.type === \"text\" ? block.text.length : ESTIMATED_IMAGE_CHARS;\n\treturn chars;\n}\n\nexport function estimateTextTokens(text: string): number {\n\treturn Math.ceil(text.length / CHARS_PER_TOKEN);\n}\n\nexport function estimateTextAndImageContentTokens(content: string | Array<TextContent | ImageContent>): number {\n\treturn Math.ceil(estimateTextAndImageContentChars(content) / CHARS_PER_TOKEN);\n}\n\nexport function estimateMessageTokens(message: Message): number {\n\tlet chars = 0;\n\n\tif (message.role === \"user\") return estimateTextAndImageContentTokens(message.content);\n\tif (message.role === \"toolResult\") return estimateTextAndImageContentTokens(message.content);\n\n\tfor (const block of message.content) {\n\t\tif (block.type === \"text\") {\n\t\t\tchars += block.text.length;\n\t\t} else if (block.type === \"thinking\") {\n\t\t\tchars += block.thinking.length;\n\t\t} else {\n\t\t\tchars += block.name.length + safeJsonStringify(block.arguments).length;\n\t\t}\n\t}\n\treturn Math.ceil(chars / CHARS_PER_TOKEN);\n}\n\nfunction getLastAssistantUsageInfo(messages: readonly Message[]): { usage: Usage; index: number } | undefined {\n\tlet latestPrefixTimestamp = Number.NEGATIVE_INFINITY;\n\tlet usageInfo: { usage: Usage; index: number } | undefined;\n\n\tfor (let i = 0; i < messages.length; i++) {\n\t\tconst message = messages[i];\n\t\tif (message.role === \"assistant\") {\n\t\t\tconst assistant = message as AssistantMessage;\n\t\t\t// A newer prefix message was inserted after this response (for example, a\n\t\t\t// compaction summary), so its usage cannot describe the current prefix.\n\t\t\tconst usageAppliesToPrefix = assistant.timestamp >= latestPrefixTimestamp;\n\t\t\tif (\n\t\t\t\tusageAppliesToPrefix &&\n\t\t\t\tassistant.stopReason !== \"aborted\" &&\n\t\t\t\tassistant.stopReason !== \"error\" &&\n\t\t\t\tcalculateContextTokens(assistant.usage) > 0\n\t\t\t) {\n\t\t\t\tusageInfo = { usage: assistant.usage, index: i };\n\t\t\t}\n\t\t}\n\t\tlatestPrefixTimestamp = Math.max(latestPrefixTimestamp, message.timestamp);\n\t}\n\n\treturn usageInfo;\n}\n\nfunction estimateMessages(messages: readonly Message[]): ContextUsageEstimate {\n\tconst usageInfo = getLastAssistantUsageInfo(messages);\n\tif (usageInfo) {\n\t\tconst usageTokens = calculateContextTokens(usageInfo.usage);\n\t\tlet trailingTokens = 0;\n\t\tfor (let i = usageInfo.index + 1; i < messages.length; i++) {\n\t\t\ttrailingTokens += estimateMessageTokens(messages[i]);\n\t\t}\n\t\treturn { tokens: usageTokens + trailingTokens, usageTokens, trailingTokens, lastUsageIndex: usageInfo.index };\n\t}\n\n\tlet tokens = 0;\n\tfor (const message of messages) tokens += estimateMessageTokens(message);\n\treturn { tokens, usageTokens: 0, trailingTokens: tokens, lastUsageIndex: null };\n}\n\nfunction isMessageArray(value: Context | readonly Message[]): value is readonly Message[] {\n\treturn Array.isArray(value);\n}\n\nexport function estimateContextTokens(context: Context | readonly Message[]): ContextUsageEstimate {\n\tif (isMessageArray(context)) return estimateMessages(context);\n\n\tconst estimate = estimateMessages(context.messages);\n\tif (estimate.lastUsageIndex !== null) return estimate;\n\n\tlet prefixTokens = context.systemPrompt ? estimateTextTokens(context.systemPrompt) : 0;\n\tif (context.tools && context.tools.length > 0) {\n\t\tprefixTokens += estimateTextTokens(safeJsonStringify(context.tools));\n\t}\n\n\treturn {\n\t\ttokens: estimate.tokens + prefixTokens,\n\t\tusageTokens: estimate.usageTokens,\n\t\ttrailingTokens: estimate.trailingTokens + prefixTokens,\n\t\tlastUsageIndex: estimate.lastUsageIndex,\n\t};\n}\n"]}

package/dist/utils/oauth/device-code.d.ts

@@ -2,6 +2,7 @@ type OAuthDeviceCodeIncompletePollResult = {
     status: "pending";
 } | {
     status: "slow_down";
+    intervalSeconds?: number;
 } | {
     status: "failed";
     message: string;
@@ -13,6 +14,7 @@ export type OAuthDeviceCodePollResult<T> = OAuthDeviceCodeIncompletePollResult |
 export type OAuthDeviceCodePollOptions<T> = {
     intervalSeconds?: number;
     expiresInSeconds?: number;
+    waitBeforeFirstPoll?: boolean;
     poll: () => Promise<OAuthDeviceCodePollResult<T>>;
     signal?: AbortSignal;
 };

package/dist/utils/oauth/device-code.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"device-code.d.ts","sourceRoot":"","sources":["../../../src/utils/oauth/device-code.ts"],"names":[],"mappings":"AAUA,KAAK,mCAAmC,GACrC;IAAE,MAAM,EAAE,SAAS,CAAA;CAAE,GACrB;IAAE,MAAM,EAAE,WAAW,CAAA;CAAE,GACvB;IAAE,MAAM,EAAE,QAAQ,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC;AAEzC,MAAM,MAAM,yBAAyB,CAAC,CAAC,IAAI,mCAAmC,GAAG;IAAE,MAAM,EAAE,UAAU,CAAC;IAAC,KAAK,EAAE,CAAC,CAAA;CAAE,CAAC;AAElH,MAAM,MAAM,0BAA0B,CAAC,CAAC,IAAI;IAC3C,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,IAAI,EAAE,MAAM,OAAO,CAAC,yBAAyB,CAAC,CAAC,CAAC,CAAC,CAAC;IAClD,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB,CAAC;AAsBF,wBAAsB,uBAAuB,CAAC,CAAC,EAAE,OAAO,EAAE,0BAA0B,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAsCnG","sourcesContent":["const CANCEL_MESSAGE = \"Login cancelled\";\nconst TIMEOUT_MESSAGE = \"Device flow timed out\";\nconst SLOW_DOWN_TIMEOUT_MESSAGE =\n\t\"Device flow timed out after one or more slow_down responses. This is often caused by clock drift in WSL or VM environments. Please sync or restart the VM clock and try again.\";\nconst MINIMUM_INTERVAL_MS = 1000;\n// RFC 8628 section 3.2: if the authorization server omits `interval`, the client must use 5 seconds.\nconst DEFAULT_POLL_INTERVAL_SECONDS = 5;\n// RFC 8628 section 3.5: `slow_down` means the polling interval must increase by 5 seconds.\nconst SLOW_DOWN_INTERVAL_INCREMENT_MS = 5000;\n\ntype OAuthDeviceCodeIncompletePollResult =\n\t| { status: \"pending\" }\n\t| { status: \"slow_down\" }\n\t| { status: \"failed\"; message: string };\n\nexport type OAuthDeviceCodePollResult<T> = OAuthDeviceCodeIncompletePollResult | { status: \"complete\"; value: T };\n\nexport type OAuthDeviceCodePollOptions<T> = {\n\tintervalSeconds?: number;\n\texpiresInSeconds?: number;\n\tpoll: () => Promise<OAuthDeviceCodePollResult<T>>;\n\tsignal?: AbortSignal;\n};\n\nfunction abortableSleep(ms: number, signal: AbortSignal | undefined, cancelMessage: string): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new Error(cancelMessage));\n\t\t\treturn;\n\t\t}\n\n\t\tconst onAbort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(new Error(cancelMessage));\n\t\t};\n\t\tconst timeout = setTimeout(() => {\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\tresolve();\n\t\t}, ms);\n\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t});\n}\n\nexport async function pollOAuthDeviceCodeFlow<T>(options: OAuthDeviceCodePollOptions<T>): Promise<T> {\n\tconst deadline =\n\t\ttypeof options.expiresInSeconds === \"number\"\n\t\t\t? Date.now() + options.expiresInSeconds * 1000\n\t\t\t: Number.POSITIVE_INFINITY;\n\tlet intervalMs = Math.max(\n\t\tMINIMUM_INTERVAL_MS,\n\t\tMath.floor((options.intervalSeconds ?? DEFAULT_POLL_INTERVAL_SECONDS) * 1000),\n\t);\n\n\tlet slowDownResponses = 0;\n\twhile (Date.now() < deadline) {\n\t\tif (options.signal?.aborted) {\n\t\t\tthrow new Error(CANCEL_MESSAGE);\n\t\t}\n\n\t\tconst result = await options.poll();\n\t\tif (result.status === \"complete\") {\n\t\t\treturn result.value;\n\t\t}\n\t\tif (result.status === \"failed\") {\n\t\t\tthrow new Error(result.message);\n\t\t}\n\t\tif (result.status === \"slow_down\") {\n\t\t\tslowDownResponses += 1;\n\t\t\t// RFC 8628 section 3.5: apply this increase to this and all subsequent requests.\n\t\t\tintervalMs = Math.max(MINIMUM_INTERVAL_MS, intervalMs + SLOW_DOWN_INTERVAL_INCREMENT_MS);\n\t\t}\n\n\t\tconst remainingMs = deadline - Date.now();\n\t\tif (remainingMs <= 0) {\n\t\t\tbreak;\n\t\t}\n\n\t\tawait abortableSleep(Math.min(intervalMs, remainingMs), options.signal, CANCEL_MESSAGE);\n\t}\n\n\tthrow new Error(slowDownResponses > 0 ? SLOW_DOWN_TIMEOUT_MESSAGE : TIMEOUT_MESSAGE);\n}\n"]}
+{"version":3,"file":"device-code.d.ts","sourceRoot":"","sources":["../../../src/utils/oauth/device-code.ts"],"names":[],"mappings":"AAUA,KAAK,mCAAmC,GACrC;IAAE,MAAM,EAAE,SAAS,CAAA;CAAE,GACrB;IAAE,MAAM,EAAE,WAAW,CAAC;IAAC,eAAe,CAAC,EAAE,MAAM,CAAA;CAAE,GACjD;IAAE,MAAM,EAAE,QAAQ,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC;AAEzC,MAAM,MAAM,yBAAyB,CAAC,CAAC,IAAI,mCAAmC,GAAG;IAAE,MAAM,EAAE,UAAU,CAAC;IAAC,KAAK,EAAE,CAAC,CAAA;CAAE,CAAC;AAElH,MAAM,MAAM,0BAA0B,CAAC,CAAC,IAAI;IAC3C,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,IAAI,EAAE,MAAM,OAAO,CAAC,yBAAyB,CAAC,CAAC,CAAC,CAAC,CAAC;IAClD,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB,CAAC;AAsBF,wBAAsB,uBAAuB,CAAC,CAAC,EAAE,OAAO,EAAE,0BAA0B,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAoDnG","sourcesContent":["const CANCEL_MESSAGE = \"Login cancelled\";\nconst TIMEOUT_MESSAGE = \"Device flow timed out\";\nconst SLOW_DOWN_TIMEOUT_MESSAGE =\n\t\"Device flow timed out after one or more slow_down responses. This is often caused by clock drift in WSL or VM environments. Please sync or restart the VM clock and try again.\";\nconst MINIMUM_INTERVAL_MS = 1000;\n// RFC 8628 section 3.2: if the authorization server omits `interval`, the client must use 5 seconds.\nconst DEFAULT_POLL_INTERVAL_SECONDS = 5;\n// RFC 8628 section 3.5: `slow_down` means the polling interval must increase by 5 seconds.\nconst SLOW_DOWN_INTERVAL_INCREMENT_MS = 5000;\n\ntype OAuthDeviceCodeIncompletePollResult =\n\t| { status: \"pending\" }\n\t| { status: \"slow_down\"; intervalSeconds?: number }\n\t| { status: \"failed\"; message: string };\n\nexport type OAuthDeviceCodePollResult<T> = OAuthDeviceCodeIncompletePollResult | { status: \"complete\"; value: T };\n\nexport type OAuthDeviceCodePollOptions<T> = {\n\tintervalSeconds?: number;\n\texpiresInSeconds?: number;\n\twaitBeforeFirstPoll?: boolean;\n\tpoll: () => Promise<OAuthDeviceCodePollResult<T>>;\n\tsignal?: AbortSignal;\n};\n\nfunction abortableSleep(ms: number, signal: AbortSignal | undefined, cancelMessage: string): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new Error(cancelMessage));\n\t\t\treturn;\n\t\t}\n\n\t\tconst onAbort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(new Error(cancelMessage));\n\t\t};\n\t\tconst timeout = setTimeout(() => {\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\tresolve();\n\t\t}, ms);\n\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t});\n}\n\nexport async function pollOAuthDeviceCodeFlow<T>(options: OAuthDeviceCodePollOptions<T>): Promise<T> {\n\tconst deadline =\n\t\ttypeof options.expiresInSeconds === \"number\"\n\t\t\t? Date.now() + options.expiresInSeconds * 1000\n\t\t\t: Number.POSITIVE_INFINITY;\n\tlet intervalMs = Math.max(\n\t\tMINIMUM_INTERVAL_MS,\n\t\tMath.floor((options.intervalSeconds ?? DEFAULT_POLL_INTERVAL_SECONDS) * 1000),\n\t);\n\n\tlet slowDownResponses = 0;\n\tif (options.waitBeforeFirstPoll) {\n\t\tconst remainingMs = deadline - Date.now();\n\t\tif (remainingMs > 0) {\n\t\t\tawait abortableSleep(Math.min(intervalMs, remainingMs), options.signal, CANCEL_MESSAGE);\n\t\t}\n\t}\n\n\twhile (Date.now() < deadline) {\n\t\tif (options.signal?.aborted) {\n\t\t\tthrow new Error(CANCEL_MESSAGE);\n\t\t}\n\n\t\tconst result = await options.poll();\n\t\tif (result.status === \"complete\") {\n\t\t\treturn result.value;\n\t\t}\n\t\tif (result.status === \"failed\") {\n\t\t\tthrow new Error(result.message);\n\t\t}\n\t\tif (result.status === \"slow_down\") {\n\t\t\tslowDownResponses += 1;\n\t\t\t// Use the server-provided interval when given (GitHub reports the new required minimum\n\t\t\t// in `interval`); trusting only a client-tracked value risks polling early forever under\n\t\t\t// WSL/VM clock drift. Otherwise apply RFC 8628 section 3.5: increase by 5 seconds.\n\t\t\tintervalMs =\n\t\t\t\ttypeof result.intervalSeconds === \"number\" &&\n\t\t\t\tNumber.isFinite(result.intervalSeconds) &&\n\t\t\t\tresult.intervalSeconds > 0\n\t\t\t\t\t? Math.max(MINIMUM_INTERVAL_MS, Math.floor(result.intervalSeconds * 1000))\n\t\t\t\t\t: Math.max(MINIMUM_INTERVAL_MS, intervalMs + SLOW_DOWN_INTERVAL_INCREMENT_MS);\n\t\t}\n\n\t\tconst remainingMs = deadline - Date.now();\n\t\tif (remainingMs <= 0) {\n\t\t\tbreak;\n\t\t}\n\n\t\tawait abortableSleep(Math.min(intervalMs, remainingMs), options.signal, CANCEL_MESSAGE);\n\t}\n\n\tthrow new Error(slowDownResponses > 0 ? SLOW_DOWN_TIMEOUT_MESSAGE : TIMEOUT_MESSAGE);\n}\n"]}

package/dist/utils/oauth/device-code.js

@@ -29,6 +29,12 @@ export async function pollOAuthDeviceCodeFlow(options) {
         : Number.POSITIVE_INFINITY;
     let intervalMs = Math.max(MINIMUM_INTERVAL_MS, Math.floor((options.intervalSeconds ?? DEFAULT_POLL_INTERVAL_SECONDS) * 1000));
     let slowDownResponses = 0;
+    if (options.waitBeforeFirstPoll) {
+        const remainingMs = deadline - Date.now();
+        if (remainingMs > 0) {
+            await abortableSleep(Math.min(intervalMs, remainingMs), options.signal, CANCEL_MESSAGE);
+        }
+    }
     while (Date.now() < deadline) {
         if (options.signal?.aborted) {
             throw new Error(CANCEL_MESSAGE);
@@ -42,8 +48,15 @@ export async function pollOAuthDeviceCodeFlow(options) {
         }
         if (result.status === "slow_down") {
             slowDownResponses += 1;
-            // RFC 8628 section 3.5: apply this increase to this and all subsequent requests.
-            intervalMs = Math.max(MINIMUM_INTERVAL_MS, intervalMs + SLOW_DOWN_INTERVAL_INCREMENT_MS);
+            // Use the server-provided interval when given (GitHub reports the new required minimum
+            // in `interval`); trusting only a client-tracked value risks polling early forever under
+            // WSL/VM clock drift. Otherwise apply RFC 8628 section 3.5: increase by 5 seconds.
+            intervalMs =
+                typeof result.intervalSeconds === "number" &&
+                    Number.isFinite(result.intervalSeconds) &&
+                    result.intervalSeconds > 0
+                    ? Math.max(MINIMUM_INTERVAL_MS, Math.floor(result.intervalSeconds * 1000))
+                    : Math.max(MINIMUM_INTERVAL_MS, intervalMs + SLOW_DOWN_INTERVAL_INCREMENT_MS);
         }
         const remainingMs = deadline - Date.now();
         if (remainingMs <= 0) {

package/dist/utils/oauth/device-code.js.map

@@ -1 +1 @@
-{"version":3,"file":"device-code.js","sourceRoot":"","sources":["../../../src/utils/oauth/device-code.ts"],"names":[],"mappings":"AAAA,MAAM,cAAc,GAAG,iBAAiB,CAAC;AACzC,MAAM,eAAe,GAAG,uBAAuB,CAAC;AAChD,MAAM,yBAAyB,GAC9B,gLAAgL,CAAC;AAClL,MAAM,mBAAmB,GAAG,IAAI,CAAC;AACjC,qGAAqG;AACrG,MAAM,6BAA6B,GAAG,CAAC,CAAC;AACxC,2FAA2F;AAC3F,MAAM,+BAA+B,GAAG,IAAI,CAAC;AAgB7C,SAAS,cAAc,CAAC,EAAU,EAAE,MAA+B,EAAE,aAAqB,EAAiB;IAC1G,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,CAAC;QACvC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,MAAM,CAAC,IAAI,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC;YACjC,OAAO;QACR,CAAC;QAED,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;YACrB,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,IAAI,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC;QAAA,CACjC,CAAC;QACF,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC;YAChC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,OAAO,EAAE,CAAC;QAAA,CACV,EAAE,EAAE,CAAC,CAAC;QAEP,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IAAA,CAC3D,CAAC,CAAC;AAAA,CACH;AAED,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAAI,OAAsC,EAAc;IACpG,MAAM,QAAQ,GACb,OAAO,OAAO,CAAC,gBAAgB,KAAK,QAAQ;QAC3C,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,CAAC,gBAAgB,GAAG,IAAI;QAC9C,CAAC,CAAC,MAAM,CAAC,iBAAiB,CAAC;IAC7B,IAAI,UAAU,GAAG,IAAI,CAAC,GAAG,CACxB,mBAAmB,EACnB,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,eAAe,IAAI,6BAA6B,CAAC,GAAG,IAAI,CAAC,CAC7E,CAAC;IAEF,IAAI,iBAAiB,GAAG,CAAC,CAAC;IAC1B,OAAO,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,CAAC;QAC9B,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,EAAE,CAAC;YAC7B,MAAM,IAAI,KAAK,CAAC,cAAc,CAAC,CAAC;QACjC,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;QACpC,IAAI,MAAM,CAAC,MAAM,KAAK,UAAU,EAAE,CAAC;YAClC,OAAO,MAAM,CAAC,KAAK,CAAC;QACrB,CAAC;QACD,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAChC,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACjC,CAAC;QACD,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YACnC,iBAAiB,IAAI,CAAC,CAAC;YACvB,iFAAiF;YACjF,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,mBAAmB,EAAE,UAAU,GAAG,+BAA+B,CAAC,CAAC;QAC1F,CAAC;QAED,MAAM,WAAW,GAAG,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC1C,IAAI,WAAW,IAAI,CAAC,EAAE,CAAC;YACtB,MAAM;QACP,CAAC;QAED,MAAM,cAAc,CAAC,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,WAAW,CAAC,EAAE,OAAO,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACzF,CAAC;IAED,MAAM,IAAI,KAAK,CAAC,iBAAiB,GAAG,CAAC,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC;AAAA,CACrF","sourcesContent":["const CANCEL_MESSAGE = \"Login cancelled\";\nconst TIMEOUT_MESSAGE = \"Device flow timed out\";\nconst SLOW_DOWN_TIMEOUT_MESSAGE =\n\t\"Device flow timed out after one or more slow_down responses. This is often caused by clock drift in WSL or VM environments. Please sync or restart the VM clock and try again.\";\nconst MINIMUM_INTERVAL_MS = 1000;\n// RFC 8628 section 3.2: if the authorization server omits `interval`, the client must use 5 seconds.\nconst DEFAULT_POLL_INTERVAL_SECONDS = 5;\n// RFC 8628 section 3.5: `slow_down` means the polling interval must increase by 5 seconds.\nconst SLOW_DOWN_INTERVAL_INCREMENT_MS = 5000;\n\ntype OAuthDeviceCodeIncompletePollResult =\n\t| { status: \"pending\" }\n\t| { status: \"slow_down\" }\n\t| { status: \"failed\"; message: string };\n\nexport type OAuthDeviceCodePollResult<T> = OAuthDeviceCodeIncompletePollResult | { status: \"complete\"; value: T };\n\nexport type OAuthDeviceCodePollOptions<T> = {\n\tintervalSeconds?: number;\n\texpiresInSeconds?: number;\n\tpoll: () => Promise<OAuthDeviceCodePollResult<T>>;\n\tsignal?: AbortSignal;\n};\n\nfunction abortableSleep(ms: number, signal: AbortSignal | undefined, cancelMessage: string): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new Error(cancelMessage));\n\t\t\treturn;\n\t\t}\n\n\t\tconst onAbort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(new Error(cancelMessage));\n\t\t};\n\t\tconst timeout = setTimeout(() => {\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\tresolve();\n\t\t}, ms);\n\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t});\n}\n\nexport async function pollOAuthDeviceCodeFlow<T>(options: OAuthDeviceCodePollOptions<T>): Promise<T> {\n\tconst deadline =\n\t\ttypeof options.expiresInSeconds === \"number\"\n\t\t\t? Date.now() + options.expiresInSeconds * 1000\n\t\t\t: Number.POSITIVE_INFINITY;\n\tlet intervalMs = Math.max(\n\t\tMINIMUM_INTERVAL_MS,\n\t\tMath.floor((options.intervalSeconds ?? DEFAULT_POLL_INTERVAL_SECONDS) * 1000),\n\t);\n\n\tlet slowDownResponses = 0;\n\twhile (Date.now() < deadline) {\n\t\tif (options.signal?.aborted) {\n\t\t\tthrow new Error(CANCEL_MESSAGE);\n\t\t}\n\n\t\tconst result = await options.poll();\n\t\tif (result.status === \"complete\") {\n\t\t\treturn result.value;\n\t\t}\n\t\tif (result.status === \"failed\") {\n\t\t\tthrow new Error(result.message);\n\t\t}\n\t\tif (result.status === \"slow_down\") {\n\t\t\tslowDownResponses += 1;\n\t\t\t// RFC 8628 section 3.5: apply this increase to this and all subsequent requests.\n\t\t\tintervalMs = Math.max(MINIMUM_INTERVAL_MS, intervalMs + SLOW_DOWN_INTERVAL_INCREMENT_MS);\n\t\t}\n\n\t\tconst remainingMs = deadline - Date.now();\n\t\tif (remainingMs <= 0) {\n\t\t\tbreak;\n\t\t}\n\n\t\tawait abortableSleep(Math.min(intervalMs, remainingMs), options.signal, CANCEL_MESSAGE);\n\t}\n\n\tthrow new Error(slowDownResponses > 0 ? SLOW_DOWN_TIMEOUT_MESSAGE : TIMEOUT_MESSAGE);\n}\n"]}
+{"version":3,"file":"device-code.js","sourceRoot":"","sources":["../../../src/utils/oauth/device-code.ts"],"names":[],"mappings":"AAAA,MAAM,cAAc,GAAG,iBAAiB,CAAC;AACzC,MAAM,eAAe,GAAG,uBAAuB,CAAC;AAChD,MAAM,yBAAyB,GAC9B,gLAAgL,CAAC;AAClL,MAAM,mBAAmB,GAAG,IAAI,CAAC;AACjC,qGAAqG;AACrG,MAAM,6BAA6B,GAAG,CAAC,CAAC;AACxC,2FAA2F;AAC3F,MAAM,+BAA+B,GAAG,IAAI,CAAC;AAiB7C,SAAS,cAAc,CAAC,EAAU,EAAE,MAA+B,EAAE,aAAqB,EAAiB;IAC1G,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,CAAC;QACvC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,MAAM,CAAC,IAAI,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC;YACjC,OAAO;QACR,CAAC;QAED,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;YACrB,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,IAAI,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC;QAAA,CACjC,CAAC;QACF,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC;YAChC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,OAAO,EAAE,CAAC;QAAA,CACV,EAAE,EAAE,CAAC,CAAC;QAEP,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IAAA,CAC3D,CAAC,CAAC;AAAA,CACH;AAED,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAAI,OAAsC,EAAc;IACpG,MAAM,QAAQ,GACb,OAAO,OAAO,CAAC,gBAAgB,KAAK,QAAQ;QAC3C,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,CAAC,gBAAgB,GAAG,IAAI;QAC9C,CAAC,CAAC,MAAM,CAAC,iBAAiB,CAAC;IAC7B,IAAI,UAAU,GAAG,IAAI,CAAC,GAAG,CACxB,mBAAmB,EACnB,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,eAAe,IAAI,6BAA6B,CAAC,GAAG,IAAI,CAAC,CAC7E,CAAC;IAEF,IAAI,iBAAiB,GAAG,CAAC,CAAC;IAC1B,IAAI,OAAO,CAAC,mBAAmB,EAAE,CAAC;QACjC,MAAM,WAAW,GAAG,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC1C,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;YACrB,MAAM,cAAc,CAAC,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,WAAW,CAAC,EAAE,OAAO,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;QACzF,CAAC;IACF,CAAC;IAED,OAAO,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,CAAC;QAC9B,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,EAAE,CAAC;YAC7B,MAAM,IAAI,KAAK,CAAC,cAAc,CAAC,CAAC;QACjC,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;QACpC,IAAI,MAAM,CAAC,MAAM,KAAK,UAAU,EAAE,CAAC;YAClC,OAAO,MAAM,CAAC,KAAK,CAAC;QACrB,CAAC;QACD,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAChC,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACjC,CAAC;QACD,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YACnC,iBAAiB,IAAI,CAAC,CAAC;YACvB,uFAAuF;YACvF,yFAAyF;YACzF,mFAAmF;YACnF,UAAU;gBACT,OAAO,MAAM,CAAC,eAAe,KAAK,QAAQ;oBAC1C,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,eAAe,CAAC;oBACvC,MAAM,CAAC,eAAe,GAAG,CAAC;oBACzB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,mBAAmB,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,eAAe,GAAG,IAAI,CAAC,CAAC;oBAC1E,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,mBAAmB,EAAE,UAAU,GAAG,+BAA+B,CAAC,CAAC;QACjF,CAAC;QAED,MAAM,WAAW,GAAG,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC1C,IAAI,WAAW,IAAI,CAAC,EAAE,CAAC;YACtB,MAAM;QACP,CAAC;QAED,MAAM,cAAc,CAAC,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,WAAW,CAAC,EAAE,OAAO,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACzF,CAAC;IAED,MAAM,IAAI,KAAK,CAAC,iBAAiB,GAAG,CAAC,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC;AAAA,CACrF","sourcesContent":["const CANCEL_MESSAGE = \"Login cancelled\";\nconst TIMEOUT_MESSAGE = \"Device flow timed out\";\nconst SLOW_DOWN_TIMEOUT_MESSAGE =\n\t\"Device flow timed out after one or more slow_down responses. This is often caused by clock drift in WSL or VM environments. Please sync or restart the VM clock and try again.\";\nconst MINIMUM_INTERVAL_MS = 1000;\n// RFC 8628 section 3.2: if the authorization server omits `interval`, the client must use 5 seconds.\nconst DEFAULT_POLL_INTERVAL_SECONDS = 5;\n// RFC 8628 section 3.5: `slow_down` means the polling interval must increase by 5 seconds.\nconst SLOW_DOWN_INTERVAL_INCREMENT_MS = 5000;\n\ntype OAuthDeviceCodeIncompletePollResult =\n\t| { status: \"pending\" }\n\t| { status: \"slow_down\"; intervalSeconds?: number }\n\t| { status: \"failed\"; message: string };\n\nexport type OAuthDeviceCodePollResult<T> = OAuthDeviceCodeIncompletePollResult | { status: \"complete\"; value: T };\n\nexport type OAuthDeviceCodePollOptions<T> = {\n\tintervalSeconds?: number;\n\texpiresInSeconds?: number;\n\twaitBeforeFirstPoll?: boolean;\n\tpoll: () => Promise<OAuthDeviceCodePollResult<T>>;\n\tsignal?: AbortSignal;\n};\n\nfunction abortableSleep(ms: number, signal: AbortSignal | undefined, cancelMessage: string): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new Error(cancelMessage));\n\t\t\treturn;\n\t\t}\n\n\t\tconst onAbort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(new Error(cancelMessage));\n\t\t};\n\t\tconst timeout = setTimeout(() => {\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\tresolve();\n\t\t}, ms);\n\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t});\n}\n\nexport async function pollOAuthDeviceCodeFlow<T>(options: OAuthDeviceCodePollOptions<T>): Promise<T> {\n\tconst deadline =\n\t\ttypeof options.expiresInSeconds === \"number\"\n\t\t\t? Date.now() + options.expiresInSeconds * 1000\n\t\t\t: Number.POSITIVE_INFINITY;\n\tlet intervalMs = Math.max(\n\t\tMINIMUM_INTERVAL_MS,\n\t\tMath.floor((options.intervalSeconds ?? DEFAULT_POLL_INTERVAL_SECONDS) * 1000),\n\t);\n\n\tlet slowDownResponses = 0;\n\tif (options.waitBeforeFirstPoll) {\n\t\tconst remainingMs = deadline - Date.now();\n\t\tif (remainingMs > 0) {\n\t\t\tawait abortableSleep(Math.min(intervalMs, remainingMs), options.signal, CANCEL_MESSAGE);\n\t\t}\n\t}\n\n\twhile (Date.now() < deadline) {\n\t\tif (options.signal?.aborted) {\n\t\t\tthrow new Error(CANCEL_MESSAGE);\n\t\t}\n\n\t\tconst result = await options.poll();\n\t\tif (result.status === \"complete\") {\n\t\t\treturn result.value;\n\t\t}\n\t\tif (result.status === \"failed\") {\n\t\t\tthrow new Error(result.message);\n\t\t}\n\t\tif (result.status === \"slow_down\") {\n\t\t\tslowDownResponses += 1;\n\t\t\t// Use the server-provided interval when given (GitHub reports the new required minimum\n\t\t\t// in `interval`); trusting only a client-tracked value risks polling early forever under\n\t\t\t// WSL/VM clock drift. Otherwise apply RFC 8628 section 3.5: increase by 5 seconds.\n\t\t\tintervalMs =\n\t\t\t\ttypeof result.intervalSeconds === \"number\" &&\n\t\t\t\tNumber.isFinite(result.intervalSeconds) &&\n\t\t\t\tresult.intervalSeconds > 0\n\t\t\t\t\t? Math.max(MINIMUM_INTERVAL_MS, Math.floor(result.intervalSeconds * 1000))\n\t\t\t\t\t: Math.max(MINIMUM_INTERVAL_MS, intervalMs + SLOW_DOWN_INTERVAL_INCREMENT_MS);\n\t\t}\n\n\t\tconst remainingMs = deadline - Date.now();\n\t\tif (remainingMs <= 0) {\n\t\t\tbreak;\n\t\t}\n\n\t\tawait abortableSleep(Math.min(intervalMs, remainingMs), options.signal, CANCEL_MESSAGE);\n\t}\n\n\tthrow new Error(slowDownResponses > 0 ? SLOW_DOWN_TIMEOUT_MESSAGE : TIMEOUT_MESSAGE);\n}\n"]}

package/dist/utils/oauth/github-copilot.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"github-copilot.d.ts","sourceRoot":"","sources":["../../../src/utils/oauth/github-copilot.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,SAAS,EAAmB,MAAM,qBAAqB,CAAC;AAItE,OAAO,KAAK,EAAE,gBAAgB,EAAE,mBAAmB,EAAuB,sBAAsB,EAAE,MAAM,YAAY,CAAC;AAqCrH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAS5D;AA4BD,wBAAgB,uBAAuB,CAAC,KAAK,CAAC,EAAE,MAAM,EAAE,gBAAgB,CAAC,EAAE,MAAM,GAAG,MAAM,CASzF;AA+LD;;GAEG;AACH,wBAAsB,yBAAyB,CAC9C,YAAY,EAAE,MAAM,EACpB,gBAAgB,CAAC,EAAE,MAAM,GACvB,OAAO,CAAC,gBAAgB,CAAC,CAM3B;AA8CD;;;;;;;GAOG;AACH,wBAAsB,kBAAkB,CAAC,OAAO,EAAE;IACjD,YAAY,EAAE,CAAC,IAAI,EAAE,mBAAmB,KAAK,IAAI,CAAC;IAClD,QAAQ,EAAE,CAAC,MAAM,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAC;QAAC,UAAU,CAAC,EAAE,OAAO,CAAA;KAAE,KAAK,OAAO,CAAC,MAAM,CAAC,CAAC;IACvG,UAAU,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;IACvC,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAuC5B;AAQD,eAAO,MAAM,kBAAkB,EAAE,SA4BhC,CAAC;AAEF,eAAO,MAAM,0BAA0B,EAAE,sBAoCxC,CAAC","sourcesContent":["/**\n * GitHub Copilot OAuth flow\n */\n\nimport type { OAuthAuth, OAuthCredential } from \"../../auth/types.ts\";\nimport { GITHUB_COPILOT_MODELS } from \"../../providers/github-copilot.models.ts\";\nimport type { Api, Model } from \"../../types.ts\";\nimport { pollOAuthDeviceCodeFlow } from \"./device-code.ts\";\nimport type { OAuthCredentials, OAuthDeviceCodeInfo, OAuthLoginCallbacks, OAuthProviderInterface } from \"./types.ts\";\n\ntype CopilotCredentials = OAuthCredentials & {\n\tenterpriseUrl?: string;\n\tavailableModelIds: string[];\n};\n\nconst decode = (s: string) => atob(s);\nconst CLIENT_ID = decode(\"SXYxLmI1MDdhMDhjODdlY2ZlOTg=\");\n\nconst COPILOT_HEADERS = {\n\t\"User-Agent\": \"GitHubCopilotChat/0.35.0\",\n\t\"Editor-Version\": \"vscode/1.107.0\",\n\t\"Editor-Plugin-Version\": \"copilot-chat/0.35.0\",\n\t\"Copilot-Integration-Id\": \"vscode-chat\",\n} as const;\nconst COPILOT_API_VERSION = \"2026-06-01\";\n\ntype DeviceCodeResponse = {\n\tdevice_code: string;\n\tuser_code: string;\n\tverification_uri: string;\n\tinterval?: number;\n\texpires_in: number;\n};\n\ntype DeviceTokenSuccessResponse = {\n\taccess_token: string;\n\ttoken_type?: string;\n\tscope?: string;\n};\n\ntype DeviceTokenErrorResponse = {\n\terror: string;\n\terror_description?: string;\n};\n\nexport function normalizeDomain(input: string): string | null {\n\tconst trimmed = input.trim();\n\tif (!trimmed) return null;\n\ttry {\n\t\tconst url = trimmed.includes(\"://\") ? new URL(trimmed) : new URL(`https://${trimmed}`);\n\t\treturn url.hostname;\n\t} catch {\n\t\treturn null;\n\t}\n}\n\nfunction getUrls(domain: string): {\n\tdeviceCodeUrl: string;\n\taccessTokenUrl: string;\n\tcopilotTokenUrl: string;\n} {\n\treturn {\n\t\tdeviceCodeUrl: `https://${domain}/login/device/code`,\n\t\taccessTokenUrl: `https://${domain}/login/oauth/access_token`,\n\t\tcopilotTokenUrl: `https://api.${domain}/copilot_internal/v2/token`,\n\t};\n}\n\n/**\n * Parse the proxy-ep from a Copilot token and convert to API base URL.\n * Token format: tid=...;exp=...;proxy-ep=proxy.individual.githubcopilot.com;...\n * Returns API URL like https://api.individual.githubcopilot.com\n */\nfunction getBaseUrlFromToken(token: string): string | null {\n\tconst match = token.match(/proxy-ep=([^;]+)/);\n\tif (!match) return null;\n\tconst proxyHost = match[1];\n\t// Convert proxy.xxx to api.xxx\n\tconst apiHost = proxyHost.replace(/^proxy\\./, \"api.\");\n\treturn `https://${apiHost}`;\n}\n\nexport function getGitHubCopilotBaseUrl(token?: string, enterpriseDomain?: string): string {\n\t// If we have a token, extract the base URL from proxy-ep\n\tif (token) {\n\t\tconst urlFromToken = getBaseUrlFromToken(token);\n\t\tif (urlFromToken) return urlFromToken;\n\t}\n\t// Fallback for enterprise or if token parsing fails\n\tif (enterpriseDomain) return `https://copilot-api.${enterpriseDomain}`;\n\treturn \"https://api.individual.githubcopilot.com\";\n}\n\nfunction asRecord(value: unknown): Record<string, unknown> | undefined {\n\treturn value && typeof value === \"object\" ? (value as Record<string, unknown>) : undefined;\n}\n\nfunction isSelectableCopilotModel(item: Record<string, unknown>): boolean {\n\tconst policy = asRecord(item.policy);\n\tconst capabilities = asRecord(item.capabilities);\n\tconst supports = asRecord(capabilities?.supports);\n\treturn item.model_picker_enabled === true && policy?.state !== \"disabled\" && supports?.tool_calls !== false;\n}\n\nfunction parseAvailableCopilotModelIds(raw: unknown): string[] {\n\tconst data = asRecord(raw)?.data;\n\tif (!Array.isArray(data)) {\n\t\tthrow new Error(\"Invalid Copilot models response\");\n\t}\n\n\tconst ids: string[] = [];\n\tfor (const rawItem of data) {\n\t\tconst item = asRecord(rawItem);\n\t\tconst id = item?.id;\n\t\tif (typeof id === \"string\" && item && isSelectableCopilotModel(item)) {\n\t\t\tids.push(id);\n\t\t}\n\t}\n\treturn ids;\n}\n\nasync function fetchAvailableGitHubCopilotModelIds(copilotToken: string, enterpriseDomain?: string): Promise<string[]> {\n\tconst baseUrl = getGitHubCopilotBaseUrl(copilotToken, enterpriseDomain);\n\tconst raw = await fetchJson(`${baseUrl}/models`, {\n\t\theaders: {\n\t\t\tAccept: \"application/json\",\n\t\t\tAuthorization: `Bearer ${copilotToken}`,\n\t\t\t...COPILOT_HEADERS,\n\t\t\t\"X-GitHub-Api-Version\": COPILOT_API_VERSION,\n\t\t},\n\t\tsignal: AbortSignal.timeout(5000),\n\t});\n\treturn parseAvailableCopilotModelIds(raw);\n}\n\nasync function fetchJson(url: string, init: RequestInit): Promise<unknown> {\n\tconst response = await fetch(url, init);\n\tif (!response.ok) {\n\t\tconst text = await response.text();\n\t\tthrow new Error(`${response.status} ${response.statusText}: ${text}`);\n\t}\n\treturn response.json();\n}\n\nasync function startDeviceFlow(domain: string): Promise<DeviceCodeResponse> {\n\tconst urls = getUrls(domain);\n\tconst data = await fetchJson(urls.deviceCodeUrl, {\n\t\tmethod: \"POST\",\n\t\theaders: {\n\t\t\tAccept: \"application/json\",\n\t\t\t\"Content-Type\": \"application/x-www-form-urlencoded\",\n\t\t\t\"User-Agent\": \"GitHubCopilotChat/0.35.0\",\n\t\t},\n\t\tbody: new URLSearchParams({\n\t\t\tclient_id: CLIENT_ID,\n\t\t\tscope: \"read:user\",\n\t\t}),\n\t});\n\n\tif (!data || typeof data !== \"object\") {\n\t\tthrow new Error(\"Invalid device code response\");\n\t}\n\n\tconst deviceCode = (data as Record<string, unknown>).device_code;\n\tconst userCode = (data as Record<string, unknown>).user_code;\n\tconst verificationUri = (data as Record<string, unknown>).verification_uri;\n\tconst interval = (data as Record<string, unknown>).interval;\n\tconst expiresIn = (data as Record<string, unknown>).expires_in;\n\n\tif (\n\t\ttypeof deviceCode !== \"string\" ||\n\t\ttypeof userCode !== \"string\" ||\n\t\ttypeof verificationUri !== \"string\" ||\n\t\t(interval !== undefined && typeof interval !== \"number\") ||\n\t\ttypeof expiresIn !== \"number\"\n\t) {\n\t\tthrow new Error(\"Invalid device code response fields\");\n\t}\n\n\t// The verification URI is opened in the user's browser and to prevent `open` from\n\t// opening an executable or similar, we force it to be a URL.\n\tlet parsedUri: URL;\n\ttry {\n\t\tparsedUri = new URL(verificationUri);\n\t} catch {\n\t\tthrow new Error(\"Untrusted verification_uri in device code response\");\n\t}\n\tif (parsedUri.protocol !== \"https:\" && parsedUri.protocol !== \"http:\") {\n\t\tthrow new Error(\"Untrusted verification_uri in device code response\");\n\t}\n\n\treturn {\n\t\tdevice_code: deviceCode,\n\t\tuser_code: userCode,\n\t\tverification_uri: parsedUri.href,\n\t\tinterval,\n\t\texpires_in: expiresIn,\n\t};\n}\n\nasync function pollForGitHubAccessToken(\n\tdomain: string,\n\tdevice: DeviceCodeResponse,\n\tsignal?: AbortSignal,\n): Promise<string> {\n\tconst urls = getUrls(domain);\n\treturn pollOAuthDeviceCodeFlow<string>({\n\t\tintervalSeconds: device.interval,\n\t\texpiresInSeconds: device.expires_in,\n\t\tsignal,\n\t\tpoll: async () => {\n\t\t\tconst raw = await fetchJson(urls.accessTokenUrl, {\n\t\t\t\tmethod: \"POST\",\n\t\t\t\theaders: {\n\t\t\t\t\tAccept: \"application/json\",\n\t\t\t\t\t\"Content-Type\": \"application/x-www-form-urlencoded\",\n\t\t\t\t\t\"User-Agent\": \"GitHubCopilotChat/0.35.0\",\n\t\t\t\t},\n\t\t\t\tbody: new URLSearchParams({\n\t\t\t\t\tclient_id: CLIENT_ID,\n\t\t\t\t\tdevice_code: device.device_code,\n\t\t\t\t\tgrant_type: \"urn:ietf:params:oauth:grant-type:device_code\",\n\t\t\t\t}),\n\t\t\t});\n\n\t\t\tif (raw && typeof raw === \"object\" && typeof (raw as DeviceTokenSuccessResponse).access_token === \"string\") {\n\t\t\t\treturn { status: \"complete\", value: (raw as DeviceTokenSuccessResponse).access_token };\n\t\t\t}\n\n\t\t\tif (raw && typeof raw === \"object\" && typeof (raw as DeviceTokenErrorResponse).error === \"string\") {\n\t\t\t\tconst { error, error_description: description } = raw as DeviceTokenErrorResponse;\n\t\t\t\tif (error === \"authorization_pending\") {\n\t\t\t\t\treturn { status: \"pending\" };\n\t\t\t\t}\n\n\t\t\t\tif (error === \"slow_down\") {\n\t\t\t\t\treturn { status: \"slow_down\" };\n\t\t\t\t}\n\n\t\t\t\tconst descriptionSuffix = description ? `: ${description}` : \"\";\n\t\t\t\treturn { status: \"failed\", message: `Device flow failed: ${error}${descriptionSuffix}` };\n\t\t\t}\n\n\t\t\treturn { status: \"failed\", message: \"Invalid device token response\" };\n\t\t},\n\t});\n}\n\nasync function refreshGitHubCopilotAccessToken(\n\trefreshToken: string,\n\tenterpriseDomain?: string,\n): Promise<OAuthCredentials> {\n\tconst domain = enterpriseDomain || \"github.com\";\n\tconst urls = getUrls(domain);\n\n\tconst raw = await fetchJson(urls.copilotTokenUrl, {\n\t\theaders: {\n\t\t\tAccept: \"application/json\",\n\t\t\tAuthorization: `Bearer ${refreshToken}`,\n\t\t\t...COPILOT_HEADERS,\n\t\t},\n\t});\n\n\tif (!raw || typeof raw !== \"object\") {\n\t\tthrow new Error(\"Invalid Copilot token response\");\n\t}\n\n\tconst token = (raw as Record<string, unknown>).token;\n\tconst expiresAt = (raw as Record<string, unknown>).expires_at;\n\n\tif (typeof token !== \"string\" || typeof expiresAt !== \"number\") {\n\t\tthrow new Error(\"Invalid Copilot token response fields\");\n\t}\n\n\treturn {\n\t\trefresh: refreshToken,\n\t\taccess: token,\n\t\texpires: expiresAt * 1000 - 5 * 60 * 1000,\n\t\tenterpriseUrl: enterpriseDomain,\n\t};\n}\n\n/**\n * Refresh GitHub Copilot token\n */\nexport async function refreshGitHubCopilotToken(\n\trefreshToken: string,\n\tenterpriseDomain?: string,\n): Promise<OAuthCredentials> {\n\tconst credentials = await refreshGitHubCopilotAccessToken(refreshToken, enterpriseDomain);\n\treturn {\n\t\t...credentials,\n\t\tavailableModelIds: await fetchAvailableGitHubCopilotModelIds(credentials.access, enterpriseDomain),\n\t};\n}\n\n/**\n * Enable a model for the user's GitHub Copilot account.\n * This is required for some models (like Claude, Grok) before they can be used.\n */\nasync function enableGitHubCopilotModel(token: string, modelId: string, enterpriseDomain?: string): Promise<boolean> {\n\tconst baseUrl = getGitHubCopilotBaseUrl(token, enterpriseDomain);\n\tconst url = `${baseUrl}/models/${modelId}/policy`;\n\n\ttry {\n\t\tconst response = await fetch(url, {\n\t\t\tmethod: \"POST\",\n\t\t\theaders: {\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t\tAuthorization: `Bearer ${token}`,\n\t\t\t\t...COPILOT_HEADERS,\n\t\t\t\t\"openai-intent\": \"chat-policy\",\n\t\t\t\t\"x-interaction-type\": \"chat-policy\",\n\t\t\t},\n\t\t\tbody: JSON.stringify({ state: \"enabled\" }),\n\t\t});\n\t\treturn response.ok;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\n/**\n * Enable all known GitHub Copilot models that may require policy acceptance.\n * Called after successful login to ensure all models are available.\n */\nasync function enableAllGitHubCopilotModels(\n\ttoken: string,\n\tenterpriseDomain?: string,\n\tonProgress?: (model: string, success: boolean) => void,\n): Promise<void> {\n\tconst models = Object.values(GITHUB_COPILOT_MODELS);\n\tawait Promise.all(\n\t\tmodels.map(async (model) => {\n\t\t\tconst success = await enableGitHubCopilotModel(token, model.id, enterpriseDomain);\n\t\t\tonProgress?.(model.id, success);\n\t\t}),\n\t);\n}\n\n/**\n * Login with GitHub Copilot OAuth (device code flow)\n *\n * @param options.onDeviceCode - Callback with URL and user code\n * @param options.onPrompt - Callback to prompt user for input\n * @param options.onProgress - Optional progress callback\n * @param options.signal - Optional AbortSignal for cancellation\n */\nexport async function loginGitHubCopilot(options: {\n\tonDeviceCode: (info: OAuthDeviceCodeInfo) => void;\n\tonPrompt: (prompt: { message: string; placeholder?: string; allowEmpty?: boolean }) => Promise<string>;\n\tonProgress?: (message: string) => void;\n\tsignal?: AbortSignal;\n}): Promise<OAuthCredentials> {\n\tconst input = await options.onPrompt({\n\t\tmessage: \"GitHub Enterprise URL/domain (blank for github.com)\",\n\t\tplaceholder: \"company.ghe.com\",\n\t\tallowEmpty: true,\n\t});\n\n\tif (options.signal?.aborted) {\n\t\tthrow new Error(\"Login cancelled\");\n\t}\n\n\tconst trimmed = input.trim();\n\tconst enterpriseDomain = normalizeDomain(input);\n\tif (trimmed && !enterpriseDomain) {\n\t\tthrow new Error(\"Invalid GitHub Enterprise URL/domain\");\n\t}\n\tconst domain = enterpriseDomain || \"github.com\";\n\n\tconst device = await startDeviceFlow(domain);\n\toptions.onDeviceCode({\n\t\tuserCode: device.user_code,\n\t\tverificationUri: device.verification_uri,\n\t\tintervalSeconds: device.interval,\n\t\texpiresInSeconds: device.expires_in,\n\t});\n\n\tconst githubAccessToken = await pollForGitHubAccessToken(domain, device, options.signal);\n\tconst credentials = await refreshGitHubCopilotAccessToken(githubAccessToken, enterpriseDomain ?? undefined);\n\n\t// Enable all models after successful login\n\toptions.onProgress?.(\"Enabling models...\");\n\tawait enableAllGitHubCopilotModels(credentials.access, enterpriseDomain ?? undefined);\n\n\t// Fetch availability after policy enable so newly enabled models are included,\n\t// while unavailable models are still filtered out.\n\treturn {\n\t\t...credentials,\n\t\tavailableModelIds: await fetchAvailableGitHubCopilotModelIds(credentials.access, enterpriseDomain ?? undefined),\n\t};\n}\n\nfunction copilotEnterpriseDomain(credential: OAuthCredential): string | undefined {\n\tconst enterpriseUrl = credential.enterpriseUrl;\n\tif (typeof enterpriseUrl !== \"string\" || !enterpriseUrl) return undefined;\n\treturn normalizeDomain(enterpriseUrl) ?? undefined;\n}\n\nexport const githubCopilotOAuth: OAuthAuth = {\n\tname: \"GitHub Copilot\",\n\n\tasync login(callbacks) {\n\t\tconst credentials = await loginGitHubCopilot({\n\t\t\tonDeviceCode: (info) => callbacks.notify({ type: \"device_code\", ...info }),\n\t\t\tonPrompt: (prompt) =>\n\t\t\t\tcallbacks.prompt({ type: \"text\", message: prompt.message, placeholder: prompt.placeholder }),\n\t\t\tonProgress: (message) => callbacks.notify({ type: \"progress\", message }),\n\t\t\tsignal: callbacks.signal,\n\t\t});\n\t\treturn { ...credentials, type: \"oauth\" };\n\t},\n\n\tasync refresh(credential) {\n\t\treturn {\n\t\t\t...(await refreshGitHubCopilotToken(credential.refresh, copilotEnterpriseDomain(credential))),\n\t\t\ttype: \"oauth\",\n\t\t};\n\t},\n\n\t/** Per-credential baseUrl from the token's proxy endpoint replaces the old `modifyModels` rewriting. */\n\tasync toAuth(credential) {\n\t\treturn {\n\t\t\tapiKey: credential.access,\n\t\t\tbaseUrl: getGitHubCopilotBaseUrl(credential.access, copilotEnterpriseDomain(credential)),\n\t\t};\n\t},\n};\n\nexport const githubCopilotOAuthProvider: OAuthProviderInterface = {\n\tid: \"github-copilot\",\n\tname: \"GitHub Copilot\",\n\n\tasync login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials> {\n\t\treturn loginGitHubCopilot({\n\t\t\tonDeviceCode: callbacks.onDeviceCode,\n\t\t\tonPrompt: callbacks.onPrompt,\n\t\t\tonProgress: callbacks.onProgress,\n\t\t\tsignal: callbacks.signal,\n\t\t});\n\t},\n\n\tasync refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials> {\n\t\tconst creds = credentials as CopilotCredentials;\n\t\treturn refreshGitHubCopilotToken(creds.refresh, creds.enterpriseUrl);\n\t},\n\n\tgetApiKey(credentials: OAuthCredentials): string {\n\t\treturn credentials.access;\n\t},\n\n\tmodifyModels(models: Model<Api>[], credentials: OAuthCredentials): Model<Api>[] {\n\t\tconst creds = credentials as CopilotCredentials;\n\t\tconst domain = creds.enterpriseUrl ? (normalizeDomain(creds.enterpriseUrl) ?? undefined) : undefined;\n\t\tconst baseUrl = getGitHubCopilotBaseUrl(creds.access, domain);\n\t\t// Older stored Pi auth entries do not have account-specific model IDs yet;\n\t\t// keep their existing generated-catalog behavior until the next refresh/login.\n\t\tconst availableModelIds = \"availableModelIds\" in creds ? new Set(creds.availableModelIds) : undefined;\n\n\t\treturn models.flatMap((m) => {\n\t\t\tif (m.provider !== \"github-copilot\") return [m];\n\t\t\tif (availableModelIds && !availableModelIds.has(m.id)) return [];\n\t\t\treturn [{ ...m, baseUrl }];\n\t\t});\n\t},\n};\n"]}
+{"version":3,"file":"github-copilot.d.ts","sourceRoot":"","sources":["../../../src/utils/oauth/github-copilot.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,SAAS,EAAmB,MAAM,qBAAqB,CAAC;AAItE,OAAO,KAAK,EAAE,gBAAgB,EAAE,mBAAmB,EAAuB,sBAAsB,EAAE,MAAM,YAAY,CAAC;AAsCrH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAS5D;AA4BD,wBAAgB,uBAAuB,CAAC,KAAK,CAAC,EAAE,MAAM,EAAE,gBAAgB,CAAC,EAAE,MAAM,GAAG,MAAM,CASzF;AAgMD;;GAEG;AACH,wBAAsB,yBAAyB,CAC9C,YAAY,EAAE,MAAM,EACpB,gBAAgB,CAAC,EAAE,MAAM,GACvB,OAAO,CAAC,gBAAgB,CAAC,CAM3B;AA8CD;;;;;;;GAOG;AACH,wBAAsB,kBAAkB,CAAC,OAAO,EAAE;IACjD,YAAY,EAAE,CAAC,IAAI,EAAE,mBAAmB,KAAK,IAAI,CAAC;IAClD,QAAQ,EAAE,CAAC,MAAM,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAC;QAAC,UAAU,CAAC,EAAE,OAAO,CAAA;KAAE,KAAK,OAAO,CAAC,MAAM,CAAC,CAAC;IACvG,UAAU,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;IACvC,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAuC5B;AAQD,eAAO,MAAM,kBAAkB,EAAE,SA4BhC,CAAC;AAEF,eAAO,MAAM,0BAA0B,EAAE,sBAoCxC,CAAC","sourcesContent":["/**\n * GitHub Copilot OAuth flow\n */\n\nimport type { OAuthAuth, OAuthCredential } from \"../../auth/types.ts\";\nimport { GITHUB_COPILOT_MODELS } from \"../../providers/github-copilot.models.ts\";\nimport type { Api, Model } from \"../../types.ts\";\nimport { pollOAuthDeviceCodeFlow } from \"./device-code.ts\";\nimport type { OAuthCredentials, OAuthDeviceCodeInfo, OAuthLoginCallbacks, OAuthProviderInterface } from \"./types.ts\";\n\ntype CopilotCredentials = OAuthCredentials & {\n\tenterpriseUrl?: string;\n\tavailableModelIds: string[];\n};\n\nconst decode = (s: string) => atob(s);\nconst CLIENT_ID = decode(\"SXYxLmI1MDdhMDhjODdlY2ZlOTg=\");\n\nconst COPILOT_HEADERS = {\n\t\"User-Agent\": \"GitHubCopilotChat/0.35.0\",\n\t\"Editor-Version\": \"vscode/1.107.0\",\n\t\"Editor-Plugin-Version\": \"copilot-chat/0.35.0\",\n\t\"Copilot-Integration-Id\": \"vscode-chat\",\n} as const;\nconst COPILOT_API_VERSION = \"2026-06-01\";\n\ntype DeviceCodeResponse = {\n\tdevice_code: string;\n\tuser_code: string;\n\tverification_uri: string;\n\tinterval?: number;\n\texpires_in: number;\n};\n\ntype DeviceTokenSuccessResponse = {\n\taccess_token: string;\n\ttoken_type?: string;\n\tscope?: string;\n};\n\ntype DeviceTokenErrorResponse = {\n\terror: string;\n\terror_description?: string;\n\tinterval?: number;\n};\n\nexport function normalizeDomain(input: string): string | null {\n\tconst trimmed = input.trim();\n\tif (!trimmed) return null;\n\ttry {\n\t\tconst url = trimmed.includes(\"://\") ? new URL(trimmed) : new URL(`https://${trimmed}`);\n\t\treturn url.hostname;\n\t} catch {\n\t\treturn null;\n\t}\n}\n\nfunction getUrls(domain: string): {\n\tdeviceCodeUrl: string;\n\taccessTokenUrl: string;\n\tcopilotTokenUrl: string;\n} {\n\treturn {\n\t\tdeviceCodeUrl: `https://${domain}/login/device/code`,\n\t\taccessTokenUrl: `https://${domain}/login/oauth/access_token`,\n\t\tcopilotTokenUrl: `https://api.${domain}/copilot_internal/v2/token`,\n\t};\n}\n\n/**\n * Parse the proxy-ep from a Copilot token and convert to API base URL.\n * Token format: tid=...;exp=...;proxy-ep=proxy.individual.githubcopilot.com;...\n * Returns API URL like https://api.individual.githubcopilot.com\n */\nfunction getBaseUrlFromToken(token: string): string | null {\n\tconst match = token.match(/proxy-ep=([^;]+)/);\n\tif (!match) return null;\n\tconst proxyHost = match[1];\n\t// Convert proxy.xxx to api.xxx\n\tconst apiHost = proxyHost.replace(/^proxy\\./, \"api.\");\n\treturn `https://${apiHost}`;\n}\n\nexport function getGitHubCopilotBaseUrl(token?: string, enterpriseDomain?: string): string {\n\t// If we have a token, extract the base URL from proxy-ep\n\tif (token) {\n\t\tconst urlFromToken = getBaseUrlFromToken(token);\n\t\tif (urlFromToken) return urlFromToken;\n\t}\n\t// Fallback for enterprise or if token parsing fails\n\tif (enterpriseDomain) return `https://copilot-api.${enterpriseDomain}`;\n\treturn \"https://api.individual.githubcopilot.com\";\n}\n\nfunction asRecord(value: unknown): Record<string, unknown> | undefined {\n\treturn value && typeof value === \"object\" ? (value as Record<string, unknown>) : undefined;\n}\n\nfunction isSelectableCopilotModel(item: Record<string, unknown>): boolean {\n\tconst policy = asRecord(item.policy);\n\tconst capabilities = asRecord(item.capabilities);\n\tconst supports = asRecord(capabilities?.supports);\n\treturn item.model_picker_enabled === true && policy?.state !== \"disabled\" && supports?.tool_calls !== false;\n}\n\nfunction parseAvailableCopilotModelIds(raw: unknown): string[] {\n\tconst data = asRecord(raw)?.data;\n\tif (!Array.isArray(data)) {\n\t\tthrow new Error(\"Invalid Copilot models response\");\n\t}\n\n\tconst ids: string[] = [];\n\tfor (const rawItem of data) {\n\t\tconst item = asRecord(rawItem);\n\t\tconst id = item?.id;\n\t\tif (typeof id === \"string\" && item && isSelectableCopilotModel(item)) {\n\t\t\tids.push(id);\n\t\t}\n\t}\n\treturn ids;\n}\n\nasync function fetchAvailableGitHubCopilotModelIds(copilotToken: string, enterpriseDomain?: string): Promise<string[]> {\n\tconst baseUrl = getGitHubCopilotBaseUrl(copilotToken, enterpriseDomain);\n\tconst raw = await fetchJson(`${baseUrl}/models`, {\n\t\theaders: {\n\t\t\tAccept: \"application/json\",\n\t\t\tAuthorization: `Bearer ${copilotToken}`,\n\t\t\t...COPILOT_HEADERS,\n\t\t\t\"X-GitHub-Api-Version\": COPILOT_API_VERSION,\n\t\t},\n\t\tsignal: AbortSignal.timeout(5000),\n\t});\n\treturn parseAvailableCopilotModelIds(raw);\n}\n\nasync function fetchJson(url: string, init: RequestInit): Promise<unknown> {\n\tconst response = await fetch(url, init);\n\tif (!response.ok) {\n\t\tconst text = await response.text();\n\t\tthrow new Error(`${response.status} ${response.statusText}: ${text}`);\n\t}\n\treturn response.json();\n}\n\nasync function startDeviceFlow(domain: string): Promise<DeviceCodeResponse> {\n\tconst urls = getUrls(domain);\n\tconst data = await fetchJson(urls.deviceCodeUrl, {\n\t\tmethod: \"POST\",\n\t\theaders: {\n\t\t\tAccept: \"application/json\",\n\t\t\t\"Content-Type\": \"application/x-www-form-urlencoded\",\n\t\t\t\"User-Agent\": \"GitHubCopilotChat/0.35.0\",\n\t\t},\n\t\tbody: new URLSearchParams({\n\t\t\tclient_id: CLIENT_ID,\n\t\t\tscope: \"read:user\",\n\t\t}),\n\t});\n\n\tif (!data || typeof data !== \"object\") {\n\t\tthrow new Error(\"Invalid device code response\");\n\t}\n\n\tconst deviceCode = (data as Record<string, unknown>).device_code;\n\tconst userCode = (data as Record<string, unknown>).user_code;\n\tconst verificationUri = (data as Record<string, unknown>).verification_uri;\n\tconst interval = (data as Record<string, unknown>).interval;\n\tconst expiresIn = (data as Record<string, unknown>).expires_in;\n\n\tif (\n\t\ttypeof deviceCode !== \"string\" ||\n\t\ttypeof userCode !== \"string\" ||\n\t\ttypeof verificationUri !== \"string\" ||\n\t\t(interval !== undefined && typeof interval !== \"number\") ||\n\t\ttypeof expiresIn !== \"number\"\n\t) {\n\t\tthrow new Error(\"Invalid device code response fields\");\n\t}\n\n\t// The verification URI is opened in the user's browser and to prevent `open` from\n\t// opening an executable or similar, we force it to be a URL.\n\tlet parsedUri: URL;\n\ttry {\n\t\tparsedUri = new URL(verificationUri);\n\t} catch {\n\t\tthrow new Error(\"Untrusted verification_uri in device code response\");\n\t}\n\tif (parsedUri.protocol !== \"https:\" && parsedUri.protocol !== \"http:\") {\n\t\tthrow new Error(\"Untrusted verification_uri in device code response\");\n\t}\n\n\treturn {\n\t\tdevice_code: deviceCode,\n\t\tuser_code: userCode,\n\t\tverification_uri: parsedUri.href,\n\t\tinterval,\n\t\texpires_in: expiresIn,\n\t};\n}\n\nasync function pollForGitHubAccessToken(\n\tdomain: string,\n\tdevice: DeviceCodeResponse,\n\tsignal?: AbortSignal,\n): Promise<string> {\n\tconst urls = getUrls(domain);\n\treturn pollOAuthDeviceCodeFlow<string>({\n\t\tintervalSeconds: device.interval,\n\t\texpiresInSeconds: device.expires_in,\n\t\twaitBeforeFirstPoll: true,\n\t\tsignal,\n\t\tpoll: async () => {\n\t\t\tconst raw = await fetchJson(urls.accessTokenUrl, {\n\t\t\t\tmethod: \"POST\",\n\t\t\t\theaders: {\n\t\t\t\t\tAccept: \"application/json\",\n\t\t\t\t\t\"Content-Type\": \"application/x-www-form-urlencoded\",\n\t\t\t\t\t\"User-Agent\": \"GitHubCopilotChat/0.35.0\",\n\t\t\t\t},\n\t\t\t\tbody: new URLSearchParams({\n\t\t\t\t\tclient_id: CLIENT_ID,\n\t\t\t\t\tdevice_code: device.device_code,\n\t\t\t\t\tgrant_type: \"urn:ietf:params:oauth:grant-type:device_code\",\n\t\t\t\t}),\n\t\t\t});\n\n\t\t\tif (raw && typeof raw === \"object\" && typeof (raw as DeviceTokenSuccessResponse).access_token === \"string\") {\n\t\t\t\treturn { status: \"complete\", value: (raw as DeviceTokenSuccessResponse).access_token };\n\t\t\t}\n\n\t\t\tif (raw && typeof raw === \"object\" && typeof (raw as DeviceTokenErrorResponse).error === \"string\") {\n\t\t\t\tconst { error, error_description: description, interval } = raw as DeviceTokenErrorResponse;\n\t\t\t\tif (error === \"authorization_pending\") {\n\t\t\t\t\treturn { status: \"pending\" };\n\t\t\t\t}\n\n\t\t\t\tif (error === \"slow_down\") {\n\t\t\t\t\treturn { status: \"slow_down\", intervalSeconds: typeof interval === \"number\" ? interval : undefined };\n\t\t\t\t}\n\n\t\t\t\tconst descriptionSuffix = description ? `: ${description}` : \"\";\n\t\t\t\treturn { status: \"failed\", message: `Device flow failed: ${error}${descriptionSuffix}` };\n\t\t\t}\n\n\t\t\treturn { status: \"failed\", message: \"Invalid device token response\" };\n\t\t},\n\t});\n}\n\nasync function refreshGitHubCopilotAccessToken(\n\trefreshToken: string,\n\tenterpriseDomain?: string,\n): Promise<OAuthCredentials> {\n\tconst domain = enterpriseDomain || \"github.com\";\n\tconst urls = getUrls(domain);\n\n\tconst raw = await fetchJson(urls.copilotTokenUrl, {\n\t\theaders: {\n\t\t\tAccept: \"application/json\",\n\t\t\tAuthorization: `Bearer ${refreshToken}`,\n\t\t\t...COPILOT_HEADERS,\n\t\t},\n\t});\n\n\tif (!raw || typeof raw !== \"object\") {\n\t\tthrow new Error(\"Invalid Copilot token response\");\n\t}\n\n\tconst token = (raw as Record<string, unknown>).token;\n\tconst expiresAt = (raw as Record<string, unknown>).expires_at;\n\n\tif (typeof token !== \"string\" || typeof expiresAt !== \"number\") {\n\t\tthrow new Error(\"Invalid Copilot token response fields\");\n\t}\n\n\treturn {\n\t\trefresh: refreshToken,\n\t\taccess: token,\n\t\texpires: expiresAt * 1000 - 5 * 60 * 1000,\n\t\tenterpriseUrl: enterpriseDomain,\n\t};\n}\n\n/**\n * Refresh GitHub Copilot token\n */\nexport async function refreshGitHubCopilotToken(\n\trefreshToken: string,\n\tenterpriseDomain?: string,\n): Promise<OAuthCredentials> {\n\tconst credentials = await refreshGitHubCopilotAccessToken(refreshToken, enterpriseDomain);\n\treturn {\n\t\t...credentials,\n\t\tavailableModelIds: await fetchAvailableGitHubCopilotModelIds(credentials.access, enterpriseDomain),\n\t};\n}\n\n/**\n * Enable a model for the user's GitHub Copilot account.\n * This is required for some models (like Claude, Grok) before they can be used.\n */\nasync function enableGitHubCopilotModel(token: string, modelId: string, enterpriseDomain?: string): Promise<boolean> {\n\tconst baseUrl = getGitHubCopilotBaseUrl(token, enterpriseDomain);\n\tconst url = `${baseUrl}/models/${modelId}/policy`;\n\n\ttry {\n\t\tconst response = await fetch(url, {\n\t\t\tmethod: \"POST\",\n\t\t\theaders: {\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t\tAuthorization: `Bearer ${token}`,\n\t\t\t\t...COPILOT_HEADERS,\n\t\t\t\t\"openai-intent\": \"chat-policy\",\n\t\t\t\t\"x-interaction-type\": \"chat-policy\",\n\t\t\t},\n\t\t\tbody: JSON.stringify({ state: \"enabled\" }),\n\t\t});\n\t\treturn response.ok;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\n/**\n * Enable all known GitHub Copilot models that may require policy acceptance.\n * Called after successful login to ensure all models are available.\n */\nasync function enableAllGitHubCopilotModels(\n\ttoken: string,\n\tenterpriseDomain?: string,\n\tonProgress?: (model: string, success: boolean) => void,\n): Promise<void> {\n\tconst models = Object.values(GITHUB_COPILOT_MODELS);\n\tawait Promise.all(\n\t\tmodels.map(async (model) => {\n\t\t\tconst success = await enableGitHubCopilotModel(token, model.id, enterpriseDomain);\n\t\t\tonProgress?.(model.id, success);\n\t\t}),\n\t);\n}\n\n/**\n * Login with GitHub Copilot OAuth (device code flow)\n *\n * @param options.onDeviceCode - Callback with URL and user code\n * @param options.onPrompt - Callback to prompt user for input\n * @param options.onProgress - Optional progress callback\n * @param options.signal - Optional AbortSignal for cancellation\n */\nexport async function loginGitHubCopilot(options: {\n\tonDeviceCode: (info: OAuthDeviceCodeInfo) => void;\n\tonPrompt: (prompt: { message: string; placeholder?: string; allowEmpty?: boolean }) => Promise<string>;\n\tonProgress?: (message: string) => void;\n\tsignal?: AbortSignal;\n}): Promise<OAuthCredentials> {\n\tconst input = await options.onPrompt({\n\t\tmessage: \"GitHub Enterprise URL/domain (blank for github.com)\",\n\t\tplaceholder: \"company.ghe.com\",\n\t\tallowEmpty: true,\n\t});\n\n\tif (options.signal?.aborted) {\n\t\tthrow new Error(\"Login cancelled\");\n\t}\n\n\tconst trimmed = input.trim();\n\tconst enterpriseDomain = normalizeDomain(input);\n\tif (trimmed && !enterpriseDomain) {\n\t\tthrow new Error(\"Invalid GitHub Enterprise URL/domain\");\n\t}\n\tconst domain = enterpriseDomain || \"github.com\";\n\n\tconst device = await startDeviceFlow(domain);\n\toptions.onDeviceCode({\n\t\tuserCode: device.user_code,\n\t\tverificationUri: device.verification_uri,\n\t\tintervalSeconds: device.interval,\n\t\texpiresInSeconds: device.expires_in,\n\t});\n\n\tconst githubAccessToken = await pollForGitHubAccessToken(domain, device, options.signal);\n\tconst credentials = await refreshGitHubCopilotAccessToken(githubAccessToken, enterpriseDomain ?? undefined);\n\n\t// Enable all models after successful login\n\toptions.onProgress?.(\"Enabling models...\");\n\tawait enableAllGitHubCopilotModels(credentials.access, enterpriseDomain ?? undefined);\n\n\t// Fetch availability after policy enable so newly enabled models are included,\n\t// while unavailable models are still filtered out.\n\treturn {\n\t\t...credentials,\n\t\tavailableModelIds: await fetchAvailableGitHubCopilotModelIds(credentials.access, enterpriseDomain ?? undefined),\n\t};\n}\n\nfunction copilotEnterpriseDomain(credential: OAuthCredential): string | undefined {\n\tconst enterpriseUrl = credential.enterpriseUrl;\n\tif (typeof enterpriseUrl !== \"string\" || !enterpriseUrl) return undefined;\n\treturn normalizeDomain(enterpriseUrl) ?? undefined;\n}\n\nexport const githubCopilotOAuth: OAuthAuth = {\n\tname: \"GitHub Copilot\",\n\n\tasync login(callbacks) {\n\t\tconst credentials = await loginGitHubCopilot({\n\t\t\tonDeviceCode: (info) => callbacks.notify({ type: \"device_code\", ...info }),\n\t\t\tonPrompt: (prompt) =>\n\t\t\t\tcallbacks.prompt({ type: \"text\", message: prompt.message, placeholder: prompt.placeholder }),\n\t\t\tonProgress: (message) => callbacks.notify({ type: \"progress\", message }),\n\t\t\tsignal: callbacks.signal,\n\t\t});\n\t\treturn { ...credentials, type: \"oauth\" };\n\t},\n\n\tasync refresh(credential) {\n\t\treturn {\n\t\t\t...(await refreshGitHubCopilotToken(credential.refresh, copilotEnterpriseDomain(credential))),\n\t\t\ttype: \"oauth\",\n\t\t};\n\t},\n\n\t/** Per-credential baseUrl from the token's proxy endpoint replaces the old `modifyModels` rewriting. */\n\tasync toAuth(credential) {\n\t\treturn {\n\t\t\tapiKey: credential.access,\n\t\t\tbaseUrl: getGitHubCopilotBaseUrl(credential.access, copilotEnterpriseDomain(credential)),\n\t\t};\n\t},\n};\n\nexport const githubCopilotOAuthProvider: OAuthProviderInterface = {\n\tid: \"github-copilot\",\n\tname: \"GitHub Copilot\",\n\n\tasync login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials> {\n\t\treturn loginGitHubCopilot({\n\t\t\tonDeviceCode: callbacks.onDeviceCode,\n\t\t\tonPrompt: callbacks.onPrompt,\n\t\t\tonProgress: callbacks.onProgress,\n\t\t\tsignal: callbacks.signal,\n\t\t});\n\t},\n\n\tasync refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials> {\n\t\tconst creds = credentials as CopilotCredentials;\n\t\treturn refreshGitHubCopilotToken(creds.refresh, creds.enterpriseUrl);\n\t},\n\n\tgetApiKey(credentials: OAuthCredentials): string {\n\t\treturn credentials.access;\n\t},\n\n\tmodifyModels(models: Model<Api>[], credentials: OAuthCredentials): Model<Api>[] {\n\t\tconst creds = credentials as CopilotCredentials;\n\t\tconst domain = creds.enterpriseUrl ? (normalizeDomain(creds.enterpriseUrl) ?? undefined) : undefined;\n\t\tconst baseUrl = getGitHubCopilotBaseUrl(creds.access, domain);\n\t\t// Older stored Pi auth entries do not have account-specific model IDs yet;\n\t\t// keep their existing generated-catalog behavior until the next refresh/login.\n\t\tconst availableModelIds = \"availableModelIds\" in creds ? new Set(creds.availableModelIds) : undefined;\n\n\t\treturn models.flatMap((m) => {\n\t\t\tif (m.provider !== \"github-copilot\") return [m];\n\t\t\tif (availableModelIds && !availableModelIds.has(m.id)) return [];\n\t\t\treturn [{ ...m, baseUrl }];\n\t\t});\n\t},\n};\n"]}

package/dist/utils/oauth/github-copilot.js

@@ -156,6 +156,7 @@ async function pollForGitHubAccessToken(domain, device, signal) {
     return pollOAuthDeviceCodeFlow({
         intervalSeconds: device.interval,
         expiresInSeconds: device.expires_in,
+        waitBeforeFirstPoll: true,
         signal,
         poll: async () => {
             const raw = await fetchJson(urls.accessTokenUrl, {
@@ -175,12 +176,12 @@ async function pollForGitHubAccessToken(domain, device, signal) {
                 return { status: "complete", value: raw.access_token };
             }
             if (raw && typeof raw === "object" && typeof raw.error === "string") {
-                const { error, error_description: description } = raw;
+                const { error, error_description: description, interval } = raw;
                 if (error === "authorization_pending") {
                     return { status: "pending" };
                 }
                 if (error === "slow_down") {
-                    return { status: "slow_down" };
+                    return { status: "slow_down", intervalSeconds: typeof interval === "number" ? interval : undefined };
                 }
                 const descriptionSuffix = description ? `: ${description}` : "";
                 return { status: "failed", message: `Device flow failed: ${error}${descriptionSuffix}` };