@daniel156161/prism 0.2.100 → 0.2.105

package/node_modules/@earendil-works/pi-ai/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"","sourcesContent":["import type { TelemetryContext } from \"@earendil-works/pi-telemetry\";\nimport 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 { PiMessagesOptions } from \"./api/pi-messages.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\t| \"pi-messages\";\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| \"radius\"\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| \"baseten\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"cloudflare-workers-ai\"\n\t| \"cloudflare-ai-gateway\"\n\t| \"qwen-token-plan\"\n\t| \"qwen-token-plan-cn\"\n\t| \"qwen-token-plan-individual\"\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>;\nexport type FetchFunction = typeof globalThis.fetch;\nexport type SessionAffinityFormat = \"openai\" | \"openai-nosession\" | \"openrouter\";\n\nexport interface ProviderResponse {\n\tstatus: number;\n\theaders: Record<string, string>;\n}\n\n/** Authentication, HTTP transport, and lifecycle callbacks shared by provider requests. */\nexport interface ProviderRequestOptions<TModel = Model<Api>> {\n\tsignal?: AbortSignal;\n\t/** Explicit parent context for telemetry produced by this logical request. */\n\ttelemetryContext?: TelemetryContext;\n\tapiKey?: string;\n\t/**\n\t * Optional fetch implementation for provider HTTP requests.\n\t * Defaults to `globalThis.fetch`. Provider adapters that cannot inject a custom implementation may reject it.\n\t * This does not affect WebSocket transports.\n\t */\n\tfetch?: FetchFunction;\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\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: TModel) => 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: TModel) => 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 * 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}\n\nexport interface StreamOptions extends ProviderRequestOptions<Model<Api>> {\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\ttemperature?: number;\n\t/**\n\t * Arbitrary sampling parameters merged into the request body as-is, after the named request\n\t * fields, so keys here override them. Lets custom OpenAI-compatible servers (llama.cpp, vLLM,\n\t * SGLang, ...) receive parameters pi does not model, e.g. `top_p`, `top_k`, `min_p`,\n\t * `repetition_penalty`. Merged over `Model.samplingParams` per key. Only applied by\n\t * OpenAI-compatible adapters (completions, responses, Azure responses); other APIs ignore it.\n\t */\n\tsamplingParams?: Record<string, unknown>;\n\tmaxTokens?: number;\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 * 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 * 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}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\nexport interface DeferredFetchOptions extends ProviderRequestOptions<Model<Api>> {\n\t/**\n\t * Maximum provider long-poll duration in milliseconds.\n\t * Defaults to 0, which performs one status check.\n\t */\n\twait?: number;\n}\n\n/** Request options for best-effort deferred-response cancellation. */\nexport type DeferredCancelOptions = ProviderRequestOptions<Model<Api>>;\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\t\"pi-messages\": PiMessagesOptions;\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 `stream` and `streamSimple`; capable modules may also\n * export deferred-response methods. 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\tfetchDeferred?(\n\t\tmodel: Model<Api>,\n\t\thandle: DeferredHandle,\n\t\toptions?: DeferredFetchOptions,\n\t): AssistantMessageEventStream;\n\tcancelDeferred?(model: Model<Api>, handle: DeferredHandle, options?: DeferredCancelOptions): Promise<void>;\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 extends ProviderRequestOptions<ImagesModel<ImagesApi>> {\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/** Ask a capable provider to return a durable handle and continue the request asynchronously. */\n\tdeferred?: boolean | { window?: \"15m\" | \"1h\" | \"24h\" };\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\t/** OpenAI Responses namespace for calls to dynamically loaded or namespaced tools. */\n\tnamespace?: string;\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 = \"pending\" | \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\" | \"deferred\";\n\nexport type JsonValue = string | number | boolean | null | JsonValue[] | { [key: string]: JsonValue };\n\nexport interface DeferredHandle {\n\tprovider: string;\n\tmodelId: string;\n\tapi: string;\n\t/** Provider token, such as a response id or batch id plus row id. */\n\tid: string;\n\texpiresAt?: number;\n\tpollAfterMs?: number;\n\t/** Provider conversion data required to reconstruct the final assistant message. */\n\tdata?: JsonValue;\n}\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\tdeferred?: DeferredHandle;\n\terrorMessage?: string;\n\trawStopReason?: string;\n\t/**\n\t * Provider indication of whether the model explicitly ended its turn.\n\t * Preserved for debugging and does not currently affect agent control flow.\n\t */\n\tendTurn?: boolean;\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\t/** Usage from the tool execution itself, if available. Not part of main LLM context accounting. */\n\tusage?: Usage;\n\t/**\n\t * Names from `Context.tools` that became available after this result.\n\t * Providers with native deferred tool loading use this as the load point;\n\t * other providers ignore it and use `Context.tools` normally.\n\t */\n\taddedToolNames?: string[];\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\n/** OpenAI grammar variants for constrained sampling. */\nexport type GrammarFormat = \"openai_lark\" | \"openai_regex\";\n\nexport type GrammarVariants = Partial<Record<GrammarFormat, string>>;\n\n/**\n * Optional provider-side constrained sampling configs for a tool.\n *\n * The `json_schema` value roughly maps to the concept of `strict` in APIs which is\n * implemented as json-schema constrained sampling by APIs. Grammar variants let\n * callers provide provider-specific encodings of the same intended language.\n */\nexport type ConstrainedSamplingConfig =\n\t| {\n\t\t\ttype: \"json_schema\";\n\t\t\tstrict: \"prefer\" | \"require\";\n\t  }\n\t| {\n\t\t\ttype: \"grammar\";\n\t\t\tvariants: GrammarVariants;\n\t  };\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n\tconstrainedSampling?: false | ConstrainedSamplingConfig;\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| {\n\t\t\ttype: \"done\";\n\t\t\treason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\" | \"deferred\">;\n\t\t\tmessage: AssistantMessage;\n\t  }\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/** Whether streamed responses include `finish_reason`. When false, pi infers `stop` or `toolUse` when the stream ends. Default: true. */\n\tsupportsFinishReason?: 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, \"baseten\" uses configurable chat_template_args 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| \"baseten\"\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 `{ \"$var\": \"thinking.effort\" }` for pi-controlled thinking values. */\n\tchatTemplateKwargs?: Record<string, ChatTemplateKwargValue>;\n\t/** Arguments to send as `chat_template_args` when `thinkingFormat` is `baseten`. Use `{ \"$var\": \"thinking.enabled\" }` or `{ \"$var\": \"thinking.effort\" }` for pi-controlled thinking values. */\n\tchatTemplateArgs?: 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 top-level `thinking_token_budget` to cap reasoning tokens (vLLM). Reasoning and the answer share `max_tokens` on these endpoints, so without a budget a reasoning-heavy turn can consume the whole response and emit no answer. Default: false. */\n\tsupportsThinkingTokenBudget?: boolean;\n\t/** Whether the provider supports OpenAI custom tools with Lark/regex grammar formats. When false, grammar-constrained tools fall back to normal function tools. Default: false; the generated model catalog enables it for capable models. */\n\tsupportsOpenAIGrammarTools?: 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, or tool-result text content. */\n\tcacheControlFormat?: \"anthropic\";\n\t/** Whether to send session-affinity data from `options.sessionId`. Default: false. */\n\tsendSessionAffinityHeaders?: boolean;\n\t/** Provider-specific deferred tool serialization mode. */\n\tdeferredToolsMode?: \"kimi\";\n\t/** Session-affinity header format: `openai` sends `session_id`, `x-client-request-id`, and `x-session-affinity`; `openai-nosession` sends `x-client-request-id` and `x-session-affinity`; `openrouter` sends `x-session-id`. Does not affect the `prompt_cache_key` body param, which is governed by cache retention. Default: auto-detected. */\n\tsessionAffinityFormat?: SessionAffinityFormat;\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/** Session-affinity header format: `openai` sends `session_id` and `x-client-request-id`; `openai-nosession` sends `x-client-request-id`; `openrouter` sends `x-session-id`. Does not affect the `prompt_cache_key` body param, which is governed by cache retention. Default: auto-detected. */\n\tsessionAffinityFormat?: SessionAffinityFormat;\n\t/** Whether the provider supports `prompt_cache_retention: \"24h\"`. Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n\t/** Whether the provider supports strict JSON-schema function tools. Defaults are API-specific; generated OpenAI models enable it explicitly. */\n\tsupportsStrictMode?: boolean;\n\t/** Whether to emit OpenAI custom tools with Lark/regex grammar formats. When false, grammar-constrained tools fall back to normal function tools. Default: false; the generated model catalog enables it for capable models. */\n\tsupportsOpenAIGrammarTools?: boolean;\n\t/** Whether the model supports message-anchored `additional_tools` input items. Default: false. */\n\tsupportsAdditionalTools?: boolean;\n\t/** Whether the model supports client-executed tool search for deferred tools. Default: false. */\n\tsupportsToolSearch?: boolean;\n\t/** Whether the model accepts `prompt_cache_options` (OpenAI GPT-5.6+ explicit prompt caching). Older OpenAI models reject the parameter. Default: false. */\n\tsupportsExplicitPromptCacheMode?: 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\t/** Whether the provider supports Anthropic strict tool schemas. Default: false; generated Anthropic models enable it explicitly. */\n\tsupportsStrictTools?: boolean;\n\t/**\n\t * Whether the provider supports deferred tools loaded by `tool_reference`\n\t * blocks in tool results. Default: true for first-party Anthropic models\n\t * except Haiku and models older than Claude 4.5; false for other providers.\n\t */\n\tsupportsToolReferences?: boolean;\n}\n\n/** Compatibility settings for Amazon Bedrock models. */\nexport interface BedrockCompat {\n\t/** Whether the model supports Bedrock strict tool schemas. Default: false. */\n\tsupportsStrictMode?: 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\t/** Default sampling parameters for this model. See {@link StreamOptions.samplingParams}; per-request keys override these. */\n\tsamplingParams?: Record<string, unknown>;\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\" | \"azure-openai-responses\" | \"openai-codex-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: TApi extends \"bedrock-converse-stream\"\n\t\t\t\t\t? BedrockCompat\n\t\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 { TelemetryContext } from \"@earendil-works/pi-telemetry\";\nimport 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 { PiMessagesOptions } from \"./api/pi-messages.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\t| \"pi-messages\";\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| \"radius\"\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| \"baseten\"\n\t| \"opencode\"\n\t| \"opencode-go\"\n\t| \"kimi-coding\"\n\t| \"cloudflare-workers-ai\"\n\t| \"cloudflare-ai-gateway\"\n\t| \"qwen-token-plan\"\n\t| \"qwen-token-plan-cn\"\n\t| \"qwen-token-plan-individual\"\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 ToolChoice = \"auto\" | \"none\";\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\" | \"thinking.budget\";\n\t\t\tomitWhenOff?: boolean;\n\t  };\n\n/** Top-level request field used to cap reasoning tokens on OpenAI-compatible servers. */\nexport type ThinkingTokenBudgetField = \"thinking_token_budget\" | \"thinking_budget\" | \"thinking_budget_tokens\";\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>;\nexport type FetchFunction = typeof globalThis.fetch;\nexport type SessionAffinityFormat = \"openai\" | \"openai-nosession\" | \"openrouter\";\n\nexport interface ProviderResponse {\n\tstatus: number;\n\theaders: Record<string, string>;\n}\n\n/** Authentication, HTTP transport, and lifecycle callbacks shared by provider requests. */\nexport interface ProviderRequestOptions<TModel = Model<Api>> {\n\tsignal?: AbortSignal;\n\t/** Explicit parent context for telemetry produced by this logical request. */\n\ttelemetryContext?: TelemetryContext;\n\tapiKey?: string;\n\t/**\n\t * Optional fetch implementation for provider HTTP requests.\n\t * Defaults to `globalThis.fetch`. Provider adapters that cannot inject a custom implementation may reject it.\n\t * This does not affect WebSocket transports.\n\t */\n\tfetch?: FetchFunction;\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\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: TModel) => 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: TModel) => 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 * 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}\n\nexport interface StreamOptions extends ProviderRequestOptions<Model<Api>> {\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\ttemperature?: number;\n\t/**\n\t * Arbitrary sampling parameters merged into the request body as-is, after the named request\n\t * fields, so keys here override them. Lets custom OpenAI-compatible servers (llama.cpp, vLLM,\n\t * SGLang, ...) receive parameters pi does not model, e.g. `top_p`, `top_k`, `min_p`,\n\t * `repetition_penalty`. Merged over `Model.samplingParams` per key. Only applied by\n\t * OpenAI-compatible adapters (completions, responses, Azure responses); other APIs ignore it.\n\t */\n\tsamplingParams?: Record<string, unknown>;\n\tmaxTokens?: number;\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 * 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 * 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}\n\nexport type ProviderStreamOptions = StreamOptions & Record<string, unknown>;\n\nexport interface DeferredFetchOptions extends ProviderRequestOptions<Model<Api>> {\n\t/**\n\t * Maximum provider long-poll duration in milliseconds.\n\t * Defaults to 0, which performs one status check.\n\t */\n\twait?: number;\n}\n\n/** Request options for best-effort deferred-response cancellation. */\nexport type DeferredCancelOptions = ProviderRequestOptions<Model<Api>>;\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\t\"pi-messages\": PiMessagesOptions;\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 `stream` and `streamSimple`; capable modules may also\n * export deferred-response methods. 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\tfetchDeferred?(\n\t\tmodel: Model<Api>,\n\t\thandle: DeferredHandle,\n\t\toptions?: DeferredFetchOptions,\n\t): AssistantMessageEventStream;\n\tcancelDeferred?(model: Model<Api>, handle: DeferredHandle, options?: DeferredCancelOptions): Promise<void>;\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 extends ProviderRequestOptions<ImagesModel<ImagesApi>> {\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\nexport interface AnthropicAllowedFallbackModel {\n\tprovider: ProviderId;\n\tmodel: string;\n\tcost: ModelCost;\n}\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\t/** Provider-neutral tool selection for simple requests. When omitted, adapters use provider-specific behavior. */\n\ttoolChoice?: ToolChoice;\n\treasoning?: ThinkingLevel;\n\t/** Ask a capable provider to return a durable handle and continue the request asynchronously. */\n\tdeferred?: boolean | { window?: \"15m\" | \"1h\" | \"24h\" };\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; // Provider-specific opaque or serialized reasoning replay data\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\t/** OpenAI Responses namespace for calls to dynamically loaded or namespaced tools. */\n\tnamespace?: string;\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 = \"pending\" | \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\" | \"deferred\";\n\nexport type JsonValue = string | number | boolean | null | JsonValue[] | { [key: string]: JsonValue };\n\nexport interface DeferredHandle {\n\tprovider: string;\n\tmodelId: string;\n\tapi: string;\n\t/** Provider token, such as a response id or batch id plus row id. */\n\tid: string;\n\texpiresAt?: number;\n\tpollAfterMs?: number;\n\t/** Provider conversion data required to reconstruct the final assistant message. */\n\tdata?: JsonValue;\n}\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\tdeferred?: DeferredHandle;\n\terrorMessage?: string;\n\trawStopReason?: string;\n\t/**\n\t * Provider indication of whether the model explicitly ended its turn.\n\t * Preserved for debugging and does not currently affect agent control flow.\n\t */\n\tendTurn?: boolean;\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\t/** Usage from the tool execution itself, if available. Not part of main LLM context accounting. */\n\tusage?: Usage;\n\t/**\n\t * Names from `Context.tools` that became available after this result.\n\t * Providers with native deferred tool loading use this as the load point;\n\t * other providers ignore it and use `Context.tools` normally.\n\t */\n\taddedToolNames?: string[];\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\n/** OpenAI grammar variants for constrained sampling. */\nexport type GrammarFormat = \"openai_lark\" | \"openai_regex\";\n\nexport type GrammarVariants = Partial<Record<GrammarFormat, string>>;\n\n/**\n * Optional provider-side constrained sampling configs for a tool.\n *\n * The `json_schema` value roughly maps to the concept of `strict` in APIs which is\n * implemented as json-schema constrained sampling by APIs. Grammar variants let\n * callers provide provider-specific encodings of the same intended language.\n */\nexport type ConstrainedSamplingConfig =\n\t| {\n\t\t\ttype: \"json_schema\";\n\t\t\tstrict: \"prefer\" | \"require\";\n\t  }\n\t| {\n\t\t\ttype: \"grammar\";\n\t\t\tvariants: GrammarVariants;\n\t  };\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n\tconstrainedSampling?: false | ConstrainedSamplingConfig;\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| {\n\t\t\ttype: \"done\";\n\t\t\treason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\" | \"deferred\">;\n\t\t\tmessage: AssistantMessage;\n\t  }\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/** Whether streamed responses include `finish_reason`. When false, pi infers `stop` or `toolUse` when the stream ends. Default: true. */\n\tsupportsFinishReason?: 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, \"baseten\" uses configurable chat_template_args 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| \"baseten\"\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 `{ \"$var\": \"thinking.enabled\" }`, `{ \"$var\": \"thinking.effort\" }`, or `{ \"$var\": \"thinking.budget\" }` for pi-controlled thinking values. */\n\tchatTemplateKwargs?: Record<string, ChatTemplateKwargValue>;\n\t/** Arguments to send as `chat_template_args` when `thinkingFormat` is `baseten`. Use `{ \"$var\": \"thinking.enabled\" }`, `{ \"$var\": \"thinking.effort\" }`, or `{ \"$var\": \"thinking.budget\" }` for pi-controlled thinking values. */\n\tchatTemplateArgs?: 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/**\n\t * Top-level request field used to cap reasoning tokens from `thinkingBudgets`.\n\t * Reasoning and the answer share `max_tokens` on these endpoints, so without a budget a\n\t * reasoning-heavy turn can consume the whole response and emit no answer.\n\t * `\"thinking_token_budget\"` is vLLM, `\"thinking_budget\"` is Qwen/DashScope/SGLang,\n\t * `\"thinking_budget_tokens\"` is llama.cpp. Off by default; not set on the generated catalog.\n\t */\n\tthinkingTokenBudgetField?: ThinkingTokenBudgetField;\n\t/** Alias for `thinkingTokenBudgetField: \"thinking_token_budget\"` (vLLM). Prefer `thinkingTokenBudgetField`. Default: false. */\n\tsupportsThinkingTokenBudget?: boolean;\n\t/** Whether the provider supports OpenAI custom tools with Lark/regex grammar formats. When false, grammar-constrained tools fall back to normal function tools. Default: false; the generated model catalog enables it for capable models. */\n\tsupportsOpenAIGrammarTools?: 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, or tool-result text content. */\n\tcacheControlFormat?: \"anthropic\";\n\t/** Whether to send session-affinity data from `options.sessionId`. Default: false. */\n\tsendSessionAffinityHeaders?: boolean;\n\t/** Provider-specific deferred tool serialization mode. */\n\tdeferredToolsMode?: \"kimi\";\n\t/** Session-affinity header format: `openai` sends `session_id`, `x-client-request-id`, and `x-session-affinity`; `openai-nosession` sends `x-client-request-id` and `x-session-affinity`; `openrouter` sends `x-session-id`. Does not affect the `prompt_cache_key` body param, which is governed by cache retention. Default: auto-detected. */\n\tsessionAffinityFormat?: SessionAffinityFormat;\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/** Session-affinity header format: `openai` sends `session_id` and `x-client-request-id`; `openai-nosession` sends `x-client-request-id`; `openrouter` sends `x-session-id`. Does not affect the `prompt_cache_key` body param, which is governed by cache retention. Default: auto-detected. */\n\tsessionAffinityFormat?: SessionAffinityFormat;\n\t/** Whether the provider supports `prompt_cache_retention: \"24h\"`. Default: true. */\n\tsupportsLongCacheRetention?: boolean;\n\t/** Whether the provider supports strict JSON-schema function tools. Defaults are API-specific; generated OpenAI models enable it explicitly. */\n\tsupportsStrictMode?: boolean;\n\t/** Whether to emit OpenAI custom tools with Lark/regex grammar formats. When false, grammar-constrained tools fall back to normal function tools. Default: false; the generated model catalog enables it for capable models. */\n\tsupportsOpenAIGrammarTools?: boolean;\n\t/** Whether the model supports message-anchored `additional_tools` input items. Default: false. */\n\tsupportsAdditionalTools?: boolean;\n\t/** Whether the model supports client-executed tool search for deferred tools. Default: false. */\n\tsupportsToolSearch?: boolean;\n\t/** Whether the model accepts `prompt_cache_options` (OpenAI GPT-5.6+ explicit prompt caching). Older OpenAI models reject the parameter. Default: false. */\n\tsupportsExplicitPromptCacheMode?: 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\t/** Whether the provider supports Anthropic strict tool schemas. Default: false; generated Anthropic models enable it explicitly. */\n\tsupportsStrictTools?: boolean;\n\t/**\n\t * Models Anthropic accepts in `fallbacks` for server-side refusal fallback,\n\t * with local pricing metadata for returned fallback responses. When absent or\n\t * empty, callers must omit `fallbacks`; Anthropic rejects the field for models\n\t * with no permitted fallback targets.\n\t */\n\tallowedFallbackModels?: AnthropicAllowedFallbackModel[];\n\t/**\n\t * Whether the provider supports deferred tools loaded by `tool_reference`\n\t * blocks in tool results. Default: true for first-party Anthropic models\n\t * except Haiku and models older than Claude 4.5; false for other providers.\n\t */\n\tsupportsToolReferences?: boolean;\n}\n\n/** Compatibility settings for Amazon Bedrock models. */\nexport interface BedrockCompat {\n\t/** Whether the model supports Bedrock strict tool schemas. Default: false. */\n\tsupportsStrictMode?: 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\t/** Default sampling parameters for this model. See {@link StreamOptions.samplingParams}; per-request keys override these. */\n\tsamplingParams?: Record<string, unknown>;\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\" | \"azure-openai-responses\" | \"openai-codex-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: TApi extends \"bedrock-converse-stream\"\n\t\t\t\t\t? BedrockCompat\n\t\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/node_modules/@earendil-works/pi-ai/dist/utils/sleep.d.ts

@@ -0,0 +1,2 @@
+export declare function sleep(ms: number, signal: AbortSignal): Promise<void>;
+//# sourceMappingURL=sleep.d.ts.map

package/node_modules/@earendil-works/pi-ai/dist/utils/sleep.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sleep.d.ts","sourceRoot":"","sources":["../../src/utils/sleep.ts"],"names":[],"mappings":"AAAA,wBAAgB,KAAK,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CAapE","sourcesContent":["export function sleep(ms: number, signal: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tsignal.throwIfAborted();\n\t\tconst onAbort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(signal.reason);\n\t\t};\n\t\tconst timeout = setTimeout(() => {\n\t\t\tsignal.removeEventListener(\"abort\", onAbort);\n\t\t\tresolve();\n\t\t}, ms);\n\t\tsignal.addEventListener(\"abort\", onAbort, { once: true });\n\t});\n}\n"]}

package/node_modules/@earendil-works/pi-ai/dist/utils/sleep.js

@@ -0,0 +1,15 @@
+export function sleep(ms, signal) {
+    return new Promise((resolve, reject) => {
+        signal.throwIfAborted();
+        const onAbort = () => {
+            clearTimeout(timeout);
+            reject(signal.reason);
+        };
+        const timeout = setTimeout(() => {
+            signal.removeEventListener("abort", onAbort);
+            resolve();
+        }, ms);
+        signal.addEventListener("abort", onAbort, { once: true });
+    });
+}
+//# sourceMappingURL=sleep.js.map

package/node_modules/@earendil-works/pi-ai/dist/utils/sleep.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sleep.js","sourceRoot":"","sources":["../../src/utils/sleep.ts"],"names":[],"mappings":"AAAA,MAAM,UAAU,KAAK,CAAC,EAAU,EAAE,MAAmB,EAAiB;IACrE,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,CAAC;QACvC,MAAM,CAAC,cAAc,EAAE,CAAC;QACxB,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;YACrB,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAAA,CACtB,CAAC;QACF,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC;YAChC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC7C,OAAO,EAAE,CAAC;QAAA,CACV,EAAE,EAAE,CAAC,CAAC;QACP,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IAAA,CAC1D,CAAC,CAAC;AAAA,CACH","sourcesContent":["export function sleep(ms: number, signal: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tsignal.throwIfAborted();\n\t\tconst onAbort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(signal.reason);\n\t\t};\n\t\tconst timeout = setTimeout(() => {\n\t\t\tsignal.removeEventListener(\"abort\", onAbort);\n\t\t\tresolve();\n\t\t}, ms);\n\t\tsignal.addEventListener(\"abort\", onAbort, { once: true });\n\t});\n}\n"]}

package/node_modules/@earendil-works/pi-ai/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@earendil-works/pi-ai",
-	"version": "0.84.2",
+	"version": "0.84.4",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -62,9 +62,8 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "0.91.1",
 		"@aws-sdk/client-bedrock-runtime": "3.1048.0",
-		"@earendil-works/pi-telemetry": "^0.84.2",
+		"@earendil-works/pi-telemetry": "^0.84.4",
 		"@google/genai": "1.52.0",
-		"@opentelemetry/api": "1.9.0",
 		"@smithy/node-http-handler": "4.7.3",
 		"http-proxy-agent": "7.0.2",
 		"https-proxy-agent": "7.0.6",
@@ -93,7 +92,7 @@
 		"node": ">=22.19.0"
 	},
 	"devDependencies": {
-		"@types/node": "24.12.4",
+		"@types/node": "22.19.19",
 		"canvas": "3.2.3",
 		"vitest": "4.1.9"
 	}

package/node_modules/@earendil-works/pi-client/CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## [0.84.4] - 2026-08-28
+
+## [0.84.3] - 2026-08-24
+
 ## [0.84.2] - 2026-08-14
 
 ## [0.84.1] - 2026-08-07

package/node_modules/@earendil-works/pi-client/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@earendil-works/pi-client",
-	"version": "0.84.2",
+	"version": "0.84.4",
 	"description": "Transport-neutral client for remote pi sessions over framed CBOR bytes",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -47,7 +47,7 @@
 		"node": ">=22.19.0"
 	},
 	"dependencies": {
-		"@earendil-works/pi-protocol": "^0.84.2"
+		"@earendil-works/pi-protocol": "^0.84.4"
 	},
 	"devDependencies": {
 		"shx": "0.4.0",

package/node_modules/@earendil-works/pi-coding-agent/CHANGELOG.md

@@ -1,5 +1,130 @@
 # Changelog
 
+## [0.84.4] - 2026-08-28
+
+### New Features
+
+- **Terminal capability overrides** — Override detected terminal hyperlink, image, and truecolor support. See [Capability Overrides](docs/terminal-setup.md#capability-overrides).
+- **Extension UI prompt events** — Integrations can distinguish active agent work from time spent waiting for `ctx.ui` prompts. See [Extension UI prompt events](docs/extensions.md#ui_prompt_start--ui_prompt_end).
+- **RPC queue clearing** — Retrieve and clear queued steering and follow-up messages with `clear_queue`. See [RPC `clear_queue`](docs/rpc.md#clear_queue).
+- **Fullscreen selection copy controls** — Disable automatic selection copying in fullscreen mode and use Ctrl+X to copy the active selection. See [UI & Display](docs/settings.md#ui--display).
+- **DeepSeek V4 Flash Vision (experimental)** — Use the vision-capable model through the built-in DeepSeek provider. See [API Keys](docs/providers.md#api-keys).
+
+### Added
+
+- Added `ui_prompt_start` and `ui_prompt_end` extension events so host integrations can distinguish active agent work from waiting on user-facing `ctx.ui` prompts ([#8355](https://github.com/earendil-works/pi/pull/8355) by [@cristinaponcela](https://github.com/cristinaponcela)).
+- Added `detectSupportedImageMimeTypeFromFile()` to the public library exports ([#8600](https://github.com/earendil-works/pi/pull/8600) by [@xl0](https://github.com/xl0)).
+- Added inherited experimental vision-capable `deepseek-v4-flash-vision-exp` model support.
+- Added transcript usage notices for compaction and branch summaries when cache miss notices are enabled.
+- Added RPC `clear_queue` to retrieve and remove queued steering and follow-up messages ([#8432](https://github.com/earendil-works/pi/issues/8432)).
+- Added environment variables and advanced settings for overriding auto-detected terminal hyperlink, image, and truecolor capabilities ([#8665](https://github.com/earendil-works/pi/issues/8665)).
+- Added `fullscreenCopyOnSelect` to disable automatic fullscreen selection copy; when disabled, `Ctrl+X` copies the active text selection before falling back to the last assistant message, while `/tree` still copies the selected message ([#7720](https://github.com/earendil-works/pi/issues/7720)).
+
+### Fixed
+
+- Fixed toggling thinking visibility clearing partial output from running Bash tools ([#8611](https://github.com/earendil-works/pi/issues/8611)).
+- Fixed Windows shell aborts crashing Pi when `taskkill.exe` is unavailable on `PATH` ([#6596](https://github.com/earendil-works/pi/issues/6596)).
+- Fixed resumed sessions corrupting the next appended entry when their JSONL file lacks a trailing newline ([#8345](https://github.com/earendil-works/pi/issues/8345)).
+- Fixed extension messages sent with `triggerTurn: false` while the agent is running being inserted between a tool call and its result, which made providers that validate message order reject the replayed history. They are now appended once the turn's tool results are in ([#8537](https://github.com/earendil-works/pi/issues/8537)).
+- Fixed compaction and branch summaries forcing `toolChoice: "none"` ([#8649](https://github.com/earendil-works/pi/issues/8649), [#8638](https://github.com/earendil-works/pi/issues/8638)).
+- Fixed large tool results crossing the auto-compaction threshold being sent to the provider before compaction. Pi now compacts between tool execution and the next assistant response in the same run, and restores interactive progress when that run resumes ([#6879](https://github.com/earendil-works/pi/issues/6879)).
+- Fixed Google Vertex requests failing with `HttpsProxyAgent is not a constructor` when the bundled Node.js runtime uses an HTTP(S) proxy ([#8610](https://github.com/earendil-works/pi/issues/8610)).
+- Fixed saving a default model from a non-empty model scope so it remains available in that scope.
+- Fixed inherited `@` file autocomplete ranking to prefer direct and shallower matches over similarly ranked nested paths ([#8669](https://github.com/earendil-works/pi/pull/8669)).
+- Fixed inherited OpenAI-compatible streams serializing thinking signatures repeatedly during streaming ([#8671](https://github.com/earendil-works/pi/pull/8671)).
+- Fixed inherited main-screen rendering crashing when image-heavy output exceeded V8's string length limit ([#8028](https://github.com/earendil-works/pi/issues/8028)).
+- Fixed inherited fullscreen double-click word selection splitting paths and kebab-case tokens on `/` and `-` ([#8676](https://github.com/earendil-works/pi/pull/8676)).
+- Fixed inherited Cloudflare AI Gateway catalogs omitting supported `workers-ai/*` passthrough models.
+- Fixed inherited OpenAI-compatible reasoning replay to merge consecutive streamed text and summary `reasoning_details` deltas.
+- Fixed inherited OpenRouter reasoning controls so reasoning-mandatory models do not receive `effort: "none"` ([#8614](https://github.com/earendil-works/pi/pull/8614) by [@davidbrai](https://github.com/davidbrai)).
+- Fixed inherited OpenAI-compatible Chat Completions ignoring an explicitly requested `toolChoice` when no tools are defined.
+- Fixed inherited fragmented Mistral tool calls splitting when continuation chunks omit the tool-call ID ([#8387](https://github.com/earendil-works/pi/issues/8387)).
+
+## [0.84.3] - 2026-08-24
+
+### New Features
+
+- **PowerShell tool** — Use optional native PowerShell command execution on Windows. See [PowerShell Tool](docs/windows.md#powershell-tool).
+- **Safer managed updates** — Stage, verify, and atomically activate updates for installer-managed installations. See [Install and Manage](docs/packages.md#install-and-manage).
+- **Model and thinking controls** — Select thinking levels with `/thinking`, search defaults, keep selections session-scoped, and persist them explicitly with Ctrl+S. See [Models and Thinking](docs/keybindings.md#models-and-thinking).
+
+### Breaking Changes
+
+- Renamed the inherited `GoogleThinkingLevel` type to `GoogleApiThinkingLevel` and added `ResolvedGoogleThinkingLevel` for normalized adapter levels.
+
+### Added
+
+- Added an optional `powershell` tool for Windows, configurable through `defaultTools` and the SDK. See [PowerShell Tool](docs/windows.md#powershell-tool).
+- Added a `/thinking` selector and searchable default choices to the model and thinking selectors; Ctrl+S saves the selected model as the global default. See [Models and Thinking](docs/keybindings.md#models-and-thinking).
+- Added optional routing session IDs to exported compaction summary helpers so callers can preserve provider routing without enabling prompt cache writes.
+- Added 
+
+
+
+usage notices for compaction and branch summaries when cache miss notices are enabled.
+- Added `session_compact_failed` extension events so compaction failures and aborts expose their reason, retry state, source, and error message to handlers ([#8175](https://github.com/earendil-works/pi/issues/8175)).
+- Added inherited provider-neutral `toolChoice` support to simple stream requests.
+- Added inherited automatic Anthropic server-side refusal fallback for supported first-party models, including returned-model usage pricing ([#8017](https://github.com/earendil-works/pi/issues/8017)).
+- Added inherited configurable OpenAI-compatible thinking-token budget fields for vLLM, Qwen/SGLang, and llama.cpp servers. See [OpenAI Compatibility](docs/models.md#openai-compatibility) ([#8275](https://github.com/earendil-works/pi/pull/8275) by [@bnsd55](https://github.com/bnsd55)).
+- Added inherited China-specific ZAI Coding Plan models, including GLM-4.6V vision support and API-equivalent usage cost estimates ([#8220](https://github.com/earendil-works/pi/issues/8220)).
+- Added inherited `deepseek-v4-pro-0813` support to the Qwen Token Plan Individual catalog ([#8194](https://github.com/earendil-works/pi/issues/8194)).
+
+### Changed
+
+- Changed experimental installer-managed installations so `pi update` stages, verifies, and atomically activates the selected release in place. See [Install and Manage](docs/packages.md#install-and-manage).
+- Changed inherited built-in xAI models to use the Responses API with encrypted reasoning replay and made Grok 4.6 the default xAI model ([#8124](https://github.com/earendil-works/pi/pull/8124) by [@Jaaneek](https://github.com/Jaaneek)).
+- Changed inherited Anthropic, Azure OpenAI, Google, Mistral, and OpenAI adapters to send Pi's default `User-Agent` unless overridden ([#8305](https://github.com/earendil-works/pi/issues/8305)).
+- Changed Windows and WSL keybinding defaults to avoid terminal-reserved shortcuts for image paste, model cycling, editor undo, fullscreen transcript navigation and search, and message queueing ([#8372](https://github.com/earendil-works/pi/issues/8372)).
+- Changed Bun release archives to ship the native clipboard binary only inside the wrapper package, removing a duplicate platform package from each archive.
+- Changed package resource glob expansion to use Node.js's built-in implementation with deterministic visible-path matching, reducing the installed runtime dependency tree.
+- Changed the bundled Node.js runtime to load jiti only when importing an extension and Babel only when uncached source needs transformation, reducing CLI startup time and bundle size.
+- Changed syntax highlighting to initialize only twenty common languages eagerly and defer the remaining grammars until after the initial TUI render, reducing CLI startup time.
+- Changed the Node.js CLI and RPC entrypoints to load a bundled runtime, reducing startup filesystem reads while keeping the public library and legacy module paths on the modular runtime for normal dependency identity.
+- Changed session sharing to render clickable terminal links, display only the canonical Radius artifact URL, and include the current system prompt and active tool definitions in Radius session shares.
+
+### Fixed
+
+- Fixed failed extension factories leaving event subscriptions, provider registrations, and default flag state active ([#8424](https://github.com/earendil-works/pi/pull/8424) by [@acmerfight](https://github.com/acmerfight)).
+- Fixed `models.json` typings omitting the documented OpenAI-compatible `compat.supportsFinishReason` provider and model override ([#8487](https://github.com/earendil-works/pi/pull/8487) by [@petrroll](https://github.com/petrroll)).
+- Fixed `/model` and `/thinking` selections being persisted globally unless explicitly saved with Ctrl+S ([#5263](https://github.com/earendil-works/pi/issues/5263)).
+- Fixed JSON and RPC `toolcall_start` events omitting the tool call id and name ([#7953](https://github.com/earendil-works/pi/pull/7953) by [@christianklotz](https://github.com/christianklotz)).
+- Fixed extensions failing to load when the Node.js CLI runs as a single-executable application ([#8237](https://github.com/earendil-works/pi/issues/8237)).
+- Fixed nested Markdown skills inside `.agents/skills/` grouping directories not being discovered.
+- Fixed compaction and branch summarization requests exposing tools to providers.
+- Fixed single-object `edit` tool inputs failing validation by accepting them as one-edit arrays in both coding-agent and harness edit tools ([#7835](https://github.com/earendil-works/pi/issues/7835)).
+- Fixed root Markdown files such as `README.md` and `AGENTS.md` in skill directories being reported as broken skills unless they declare valid skill frontmatter ([#7805](https://github.com/earendil-works/pi/issues/7805)).
+- Fixed the default Cerebras model referencing an unavailable Z.AI model.
+- Fixed inherited OpenAI-compatible Chat Completions reasoning replay to preserve and resend assistant-level `reasoning_details` verbatim and in order ([#7994](https://github.com/earendil-works/pi/issues/7994)).
+- Fixed inherited Anthropic server-side fallback responses being priced with the requested model instead of the returned fallback model ([#8285](https://github.com/earendil-works/pi/issues/8285)).
+- Fixed inherited GitHub Copilot login triggering model-policy rate limits by limiting policy updates, retrying model discovery once, and honoring server retry delays ([#7850](https://github.com/earendil-works/pi/issues/7850)).
+- Fixed inherited Amazon Bedrock dropping and failing to replay opaque redacted reasoning from non-Anthropic models ([#8314](https://github.com/earendil-works/pi/pull/8314) by [@seiji](https://github.com/seiji)).
+- Fixed inherited Z.AI Coding Plan models deriving incomplete reasoning-effort metadata, including missing GLM-5.3 low, high, and max levels ([#8336](https://github.com/earendil-works/pi/issues/8336)).
+- Fixed inherited DeepSeek V4 Flash on OpenCode and OpenCode Go omitting its supported low thinking level ([#8181](https://github.com/earendil-works/pi/pull/8181) by [@tianshuang](https://github.com/tianshuang)).
+- Fixed inherited Azure OpenAI Responses ignoring `toolChoice` in provider-specific stream requests.
+- Fixed inherited Amazon Bedrock response hooks receiving only a synthesized request id instead of the raw response headers ([#8234](https://github.com/earendil-works/pi/issues/8234)).
+- Fixed inherited Kimi usage reporting so top-level `cached_tokens` count as cache reads instead of normal input tokens ([#8075](https://github.com/earendil-works/pi/issues/8075)).
+- Fixed inherited Google custom models ignoring `thinkingLevelMap`, which dropped extended thinking controls ([#8135](https://github.com/earendil-works/pi/issues/8135)).
+- Fixed writes to `auth.json` and `models-store.json` overriding administrator-managed file permissions and ACLs ([#7779](https://github.com/earendil-works/pi/issues/7779)).
+- Fixed UTF-8 BOM markers preventing frontmatter and user configuration files from loading ([#8337](https://github.com/earendil-works/pi/issues/8337)).
+- Fixed invalid settings files being easy to miss during interactive startup by rendering warnings with the file path inside the TUI ([#7829](https://github.com/earendil-works/pi/issues/7829)).
+- Fixed the subagent example repeatedly prompting before running project-local agents in trusted repositories ([#8261](https://github.com/earendil-works/pi/issues/8261)).
+- Added `session_compact_failed` extension events so compaction failures and aborts expose their reason, retry state, source, and error message to handlers ([#8175](https://github.com/earendil-works/pi/issues/8175)).
+- Fixed truncated compaction and branch summaries being persisted when generation reaches its output token limit ([#7048](https://github.com/earendil-works/pi/issues/7048)).
+- Fixed npm package update checks treating older registry versions as available updates, preventing `pi update` from downgrading already-newer installed packages ([#8226](https://github.com/earendil-works/pi/issues/8226)).
+- Fixed built-in llama.cpp models disappearing from `/model` when `/llama` refreshed a configured server under `PI_OFFLINE`, and included idle-slept `sleeping` router models plus autoloadable unloaded presets in the selectable catalog ([#8558](https://github.com/earendil-works/pi/pull/8558) by [@cristinaponcela](https://github.com/cristinaponcela)).
+- Fixed `pi.registerFlag()` accepting default values that do not match the declared flag type ([#8064](https://github.com/earendil-works/pi/issues/8064)).
+- Fixed Z.AI Coding Plan defaults referencing the removed GLM-5.1 model ([#8096](https://github.com/earendil-works/pi/issues/8096)).
+- Fixed repeated ambiguous truncated-response recovery being mislabeled as context overflow ([#8130](https://github.com/earendil-works/pi/issues/8130)).
+- Fixed duplicate fullscreen right-click paste in VS Code-based terminals on Windows ([#8186](https://github.com/earendil-works/pi/issues/8186)).
+- Fixed inherited padded text exceeding narrow terminal widths ([#8252](https://github.com/earendil-works/pi/issues/8252)).
+- Fixed inherited wrapped Markdown table links leaking color into borders and neighboring cells, including tables inside blockquotes ([#8335](https://github.com/earendil-works/pi/issues/8335)).
+- Fixed llama.cpp login guidance to direct users to `/llama` before `/model` when no local models are loaded ([#8203](https://github.com/earendil-works/pi/issues/8203)).
+- Fixed hung pi.dev model catalog requests consuming the entire refresh deadline without retrying ([#8198](https://github.com/earendil-works/pi/issues/8198)).
+- Fixed inherited Xiaomi model catalogs listing shut-down MiMo V2 models in `/model` and `--list-models` ([#8187](https://github.com/earendil-works/pi/issues/8187)).
+- Fixed branch summary entries recording the navigation destination in `fromId` instead of the pre-navigation source leaf.
+- Fixed threshold auto-compaction being skipped when providers omit streaming usage data ([#8328](https://github.com/earendil-works/pi/issues/8328)).
+- Fixed dash-prefixed prompts being parsed as options by supporting `--` as an end-of-options delimiter ([#7269](https://github.com/earendil-works/pi/issues/7269)).
+
 ## [0.84.2] - 2026-08-14
 
 ### New Features
@@ -30,6 +155,8 @@
 
 ### Fixed
 
+- Fixed root Markdown files such as `README.md` and `AGENTS.md` in skill directories being reported as broken skills unless they declare valid skill frontmatter ([#7805](https://github.com/earendil-works/pi/issues/7805)).
+- Fixed single-object `edit` tool inputs failing validation by accepting them as one-edit arrays in both coding-agent and harness edit tools ([#7835](https://github.com/earendil-works/pi/issues/7835)).
 - Fixed managed-tool downloads delaying TUI startup and hiding diagnostics in fullscreen mode by mounting the TUI first and showing download progress and warnings inside it.
 - Fixed opening a model selector immediately after startup cancelling and restarting the in-progress model catalog refresh.
 - Fixed inherited GitHub Copilot login triggering API rate limits while enabling model policies by limiting concurrent policy updates ([#6187](https://github.com/earendil-works/pi/issues/6187)).

package/node_modules/@earendil-works/pi-coding-agent/README.md

@@ -96,7 +96,7 @@ Then just talk to pi. By default, pi gives the model four tools: `read`, `write`
 
 ## Providers & Models
 
-For each built-in provider, pi maintains a list of tool-capable models. Configured provider catalogs refresh automatically; run `pi update --models` to force an immediate refresh. Authenticate via subscription (`/login`) or API key, then select any model from that provider via `/model` (or Ctrl+L).
+For each built-in provider, pi maintains a list of tool-capable models. Configured provider catalogs refresh automatically; run `pi update --models` to force an immediate refresh. Authenticate via subscription (`/login`) or API key, then select any model from that provider via `/model` (or Ctrl+L). Press Ctrl+S in the model picker to save the highlighted model as the startup default.
 
 **Subscriptions:**
 - Anthropic Claude Pro/Max
@@ -178,9 +178,10 @@ Type `/` in the editor to trigger commands. [Extensions](#extensions) can regist
 |---------|-------------|
 | `/login`, `/logout` | Manage provider credentials |
 | [`/llama`](docs/llama-cpp.md) | Download, load, and unload llama.cpp router models |
-| `/model` | Switch models |
+| `/model` | Switch models; Ctrl+S in the picker saves the startup default |
+| `/thinking` | Switch thinking level; Ctrl+S in the picker saves the startup default |
 | `/scoped-models` | Enable/disable models for Ctrl+P cycling |
-| `/settings` | Thinking level, theme, message delivery, transport |
+| `/settings` | Theme, message delivery, transport, and other preferences |
 | `/resume` | Pick from previous sessions |
 | `/new` | Start a new session |
 | `/name <name>` | Set session display name |
@@ -216,7 +217,7 @@ See `/hotkeys` for the full list. Customize via `~/.pi/agent/keybindings.json`.
 | Shift+Tab | Cycle thinking level |
 | Ctrl+O | Collapse/expand tool output |
 | Ctrl+T | Collapse/expand thinking blocks |
-| Ctrl+X | Copy the last assistant message |
+| Ctrl+X | Copy the last assistant message; with fullscreen copy-on-select disabled, copy the active text selection |
 
 ### Message Queue
 
@@ -514,7 +515,7 @@ Read the [blog post](https://mariozechner.at/posts/2025-11-30-pi-coding-agent/)
 ## CLI Reference
 
 ```bash
-pi [options] [@files...] [messages...]
+pi [options] [--] [@files...] [messages...]
 ```
 
 ### Package Commands
@@ -584,7 +585,7 @@ cat README.md | pi -p "Summarize this text"
 | `--no-builtin-tools`, `-nbt` | Disable built-in tools by default but keep extension/custom tools enabled |
 | `--no-tools`, `-nt` | Disable all tools by default |
 
-Available built-in tools: `read`, `bash`, `edit`, `write`, `grep`, `find`, `ls`
+Available built-in tools: `read`, `bash`, `powershell` (Windows), `edit`, `write`, `grep`, `find`, `ls`
 
 ### Resource Options
 
@@ -613,6 +614,7 @@ Combine `--no-*` with explicit flags to load exactly what you need, ignoring set
 | `--verbose` | Force verbose startup |
 | `-a`, `--approve` | Trust project-local files for this run |
 | `-na`, `--no-approve` | Ignore project-local files for this run |
+| `--` | Stop option parsing; remaining arguments are prompts or `@file` inputs |
 | `-h`, `--help` | Show help |
 | `-v`, `--version` | Show version |
 
@@ -635,6 +637,9 @@ pi "List all .ts files in src/"
 # Non-interactive
 pi -p "Summarize this codebase"
 
+# Prompt beginning with a dash
+pi -p -- "- Summarize these points"
+
 # Non-interactive with piped stdin
 cat README.md | pi -p "Summarize this text"
 
@@ -678,7 +683,7 @@ pi --thinking high "Solve this complex problem"
 | `PI_CACHE_RETENTION` | Set to `long` for extended prompt cache (Anthropic: 1h, OpenAI: 24h) |
 | `VISUAL`, `EDITOR` | Fallback external editor for Ctrl+G when `externalEditor` is unset; defaults to Notepad on Windows and `nano` elsewhere |
 
-Commands run by the LLM-callable bash tool also receive current session metadata:
+Commands run by the LLM-callable `bash` and `powershell` tools also receive current session metadata:
 
 | Variable | Description |
 |----------|-------------|
@@ -688,7 +693,7 @@ Commands run by the LLM-callable bash tool also receive current session metadata
 | `PI_MODEL` | Currently selected model ID |
 | `PI_REASONING_LEVEL` | Current effective reasoning level |
 
-These values are resolved when each command starts. See [Environment Variables](docs/environment-variables.md#bash-tool-session-environment) for semantics, examples, and custom-tool opt-out.
+These values are resolved when each command starts. See [Environment Variables](docs/environment-variables.md#shell-tool-session-environment) for semantics, examples, and custom-tool opt-out.
 
 ---