@mariozechner/pi-ai 0.44.0 → 0.45.0

package/dist/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,0BAA0B,CAAC;AACjE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAC3D,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,kCAAkC,CAAC;AAC/E,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,uCAAuC,CAAC;AACzF,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,mCAAmC,CAAC;AAClF,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,iCAAiC,CAAC;AAC9E,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AAE3E,YAAY,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AAE3E,MAAM,MAAM,GAAG,GACZ,oBAAoB,GACpB,kBAAkB,GAClB,wBAAwB,GACxB,oBAAoB,GACpB,sBAAsB,GACtB,mBAAmB,GACnB,eAAe,CAAC;AAEnB,MAAM,WAAW,aAAa;IAC7B,oBAAoB,EAAE,gBAAgB,CAAC;IACvC,oBAAoB,EAAE,wBAAwB,CAAC;IAC/C,kBAAkB,EAAE,sBAAsB,CAAC;IAC3C,wBAAwB,EAAE,2BAA2B,CAAC;IACtD,sBAAsB,EAAE,aAAa,CAAC;IACtC,mBAAmB,EAAE,sBAAsB,CAAC;IAC5C,eAAe,EAAE,mBAAmB,CAAC;CACrC;AAWD,MAAM,MAAM,aAAa,CAAC,IAAI,SAAS,GAAG,IAAI,aAAa,CAAC,IAAI,CAAC,CAAC;AAElE,MAAM,MAAM,aAAa,GACtB,WAAW,GACX,QAAQ,GACR,mBAAmB,GACnB,oBAAoB,GACpB,eAAe,GACf,QAAQ,GACR,cAAc,GACd,gBAAgB,GAChB,KAAK,GACL,MAAM,GACN,UAAU,GACV,YAAY,GACZ,KAAK,GACL,SAAS,GACT,UAAU,CAAC;AACd,MAAM,MAAM,QAAQ,GAAG,aAAa,GAAG,MAAM,CAAC;AAE9C,MAAM,MAAM,aAAa,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAE5E,yEAAyE;AACzE,MAAM,WAAW,eAAe;IAC/B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAGD,MAAM,WAAW,aAAa;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;;;;OAIG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;CACnB;AAGD,MAAM,WAAW,mBAAoB,SAAQ,aAAa;IACzD,SAAS,CAAC,EAAE,aAAa,CAAC;IAC1B,4EAA4E;IAC5E,eAAe,CAAC,EAAE,eAAe,CAAC;CAClC;AAGD,MAAM,MAAM,cAAc,CAAC,IAAI,SAAS,GAAG,IAAI,CAC9C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,EAAE,aAAa,CAAC,IAAI,CAAC,KACxB,2BAA2B,CAAC;AAEjC,MAAM,WAAW,WAAW;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,eAAe;IAC/B,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY;IAC5B,IAAI,EAAE,OAAO,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,QAAQ;IACxB,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC/B,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,KAAK;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE;QACL,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;QACnB,KAAK,EAAE,MAAM,CAAC;KACd,CAAC;CACF;AAED,MAAM,MAAM,UAAU,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,WAAW;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,GAAG,CAAC,WAAW,GAAG,YAAY,CAAC,EAAE,CAAC;IACjD,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,gBAAgB;IAChC,IAAI,EAAE,WAAW,CAAC;IAClB,OAAO,EAAE,CAAC,WAAW,GAAG,eAAe,GAAG,QAAQ,CAAC,EAAE,CAAC;IACtD,GAAG,EAAE,GAAG,CAAC;IACT,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,KAAK,CAAC;IACb,UAAU,EAAE,UAAU,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,iBAAiB,CAAC,QAAQ,GAAG,GAAG;IAChD,IAAI,EAAE,YAAY,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,CAAC,WAAW,GAAG,YAAY,CAAC,EAAE,CAAC;IACxC,OAAO,CAAC,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,OAAO,GAAG,WAAW,GAAG,gBAAgB,GAAG,iBAAiB,CAAC;AAEzE,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAEjD,MAAM,WAAW,IAAI,CAAC,WAAW,SAAS,OAAO,GAAG,OAAO;IAC1D,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,WAAW,CAAC;CACxB;AAED,MAAM,WAAW,OAAO;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,OAAO,EAAE,CAAC;IACpB,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;CACf;AAED,MAAM,MAAM,qBAAqB,GAC9B;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC5C;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACvE;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC3E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC3E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,QAAQ,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC7F;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS,CAAC,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACvG;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,SAAS,GAAG,OAAO,CAAC,CAAC;IAAC,KAAK,EAAE,gBAAgB,CAAA;CAAE,CAAC;AAEhG;;;GAGG;AACH,MAAM,WAAW,YAAY;IAC5B,wFAAwF;IACxF,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,yGAAyG;IACzG,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,yFAAyF;IACzF,uBAAuB,CAAC,EAAE,OAAO,CAAC;IAClC,qIAAqI;IACrI,wBAAwB,CAAC,EAAE,OAAO,CAAC;IACnC,0EAA0E;IAC1E,cAAc,CAAC,EAAE,uBAAuB,GAAG,YAAY,CAAC;IACxD,sFAAsF;IACtF,sBAAsB,CAAC,EAAE,OAAO,CAAC;IACjC,2HAA2H;IAC3H,gCAAgC,CAAC,EAAE,OAAO,CAAC;IAC3C,4HAA4H;IAC5H,sBAAsB,CAAC,EAAE,OAAO,CAAC;IACjC,kIAAkI;IAClI,sBAAsB,CAAC,EAAE,OAAO,CAAC;CACjC;AAGD,MAAM,WAAW,KAAK,CAAC,IAAI,SAAS,GAAG;IACtC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,IAAI,CAAC;IACV,QAAQ,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,OAAO,CAAC;IACnB,KAAK,EAAE,CAAC,MAAM,GAAG,OAAO,CAAC,EAAE,CAAC;IAC5B,IAAI,EAAE;QACL,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;KACnB,CAAC;IACF,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,kGAAkG;IAClG,MAAM,CAAC,EAAE,IAAI,SAAS,oBAAoB,GAAG,YAAY,GAAG,KAAK,CAAC;CAClE","sourcesContent":["import type { AnthropicOptions } from \"./providers/anthropic.js\";\nimport type { GoogleOptions } from \"./providers/google.js\";\nimport type { GoogleGeminiCliOptions } from \"./providers/google-gemini-cli.js\";\nimport type { GoogleVertexOptions } from \"./providers/google-vertex.js\";\nimport type { OpenAICodexResponsesOptions } from \"./providers/openai-codex-responses.js\";\nimport type { OpenAICompletionsOptions } from \"./providers/openai-completions.js\";\nimport type { OpenAIResponsesOptions } from \"./providers/openai-responses.js\";\nimport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type Api =\n\t| \"openai-completions\"\n\t| \"openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\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\"google-generative-ai\": GoogleOptions;\n\t\"google-gemini-cli\": GoogleGeminiCliOptions;\n\t\"google-vertex\": GoogleVertexOptions;\n}\n\n// Compile-time exhaustiveness check - this will fail if ApiOptionsMap doesn't have all KnownApi keys\ntype _CheckExhaustive = ApiOptionsMap extends Record<Api, StreamOptions>\n\t? Record<Api, StreamOptions> extends ApiOptionsMap\n\t\t? true\n\t\t: [\"ApiOptionsMap is missing some KnownApi values\", Exclude<Api, keyof ApiOptionsMap>]\n\t: [\"ApiOptionsMap doesn't extend Record<KnownApi, StreamOptions>\"];\nconst _exhaustive: _CheckExhaustive = true;\n\n// Helper type to get options for a specific API\nexport type OptionsForApi<TApi extends Api> = ApiOptionsMap[TApi];\n\nexport type KnownProvider =\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"openai-codex\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"zai\"\n\t| \"mistral\"\n\t| \"opencode\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\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 interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\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}\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n}\n\n// Generic StreamFunction with typed options\nexport type StreamFunction<TApi extends Api> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions: OptionsForApi<TApi>,\n) => AssistantMessageEventStream;\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, the message ID\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"@sinclair/typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for openai-completions API.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Whether tool call IDs must be normalized to Mistral format (exactly 9 alphanumeric chars). Default: auto-detected from URL. */\n\trequiresMistralToolIds?: boolean;\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: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/** Compatibility overrides for openai-completions API. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\" ? OpenAICompat : never;\n}\n"]}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,+BAA+B,CAAC;AACpE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,0BAA0B,CAAC;AACjE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAC3D,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,kCAAkC,CAAC;AAC/E,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,8BAA8B,CAAC;AACxE,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,uCAAuC,CAAC;AACzF,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,mCAAmC,CAAC;AAClF,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,iCAAiC,CAAC;AAC9E,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AAE3E,YAAY,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AAE3E,MAAM,MAAM,GAAG,GACZ,oBAAoB,GACpB,kBAAkB,GAClB,wBAAwB,GACxB,oBAAoB,GACpB,yBAAyB,GACzB,sBAAsB,GACtB,mBAAmB,GACnB,eAAe,CAAC;AAEnB,MAAM,WAAW,aAAa;IAC7B,oBAAoB,EAAE,gBAAgB,CAAC;IACvC,yBAAyB,EAAE,cAAc,CAAC;IAC1C,oBAAoB,EAAE,wBAAwB,CAAC;IAC/C,kBAAkB,EAAE,sBAAsB,CAAC;IAC3C,wBAAwB,EAAE,2BAA2B,CAAC;IACtD,sBAAsB,EAAE,aAAa,CAAC;IACtC,mBAAmB,EAAE,sBAAsB,CAAC;IAC5C,eAAe,EAAE,mBAAmB,CAAC;CACrC;AAWD,MAAM,MAAM,aAAa,CAAC,IAAI,SAAS,GAAG,IAAI,aAAa,CAAC,IAAI,CAAC,CAAC;AAElE,MAAM,MAAM,aAAa,GACtB,gBAAgB,GAChB,WAAW,GACX,QAAQ,GACR,mBAAmB,GACnB,oBAAoB,GACpB,eAAe,GACf,QAAQ,GACR,cAAc,GACd,gBAAgB,GAChB,KAAK,GACL,MAAM,GACN,UAAU,GACV,YAAY,GACZ,KAAK,GACL,SAAS,GACT,SAAS,GACT,UAAU,CAAC;AACd,MAAM,MAAM,QAAQ,GAAG,aAAa,GAAG,MAAM,CAAC;AAE9C,MAAM,MAAM,aAAa,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAE5E,yEAAyE;AACzE,MAAM,WAAW,eAAe;IAC/B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAGD,MAAM,WAAW,aAAa;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;;;;OAIG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC;CACnB;AAGD,MAAM,WAAW,mBAAoB,SAAQ,aAAa;IACzD,SAAS,CAAC,EAAE,aAAa,CAAC;IAC1B,4EAA4E;IAC5E,eAAe,CAAC,EAAE,eAAe,CAAC;CAClC;AAGD,MAAM,MAAM,cAAc,CAAC,IAAI,SAAS,GAAG,IAAI,CAC9C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,EAAE,aAAa,CAAC,IAAI,CAAC,KACxB,2BAA2B,CAAC;AAEjC,MAAM,WAAW,WAAW;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,eAAe;IAC/B,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY;IAC5B,IAAI,EAAE,OAAO,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,QAAQ;IACxB,IAAI,EAAE,UAAU,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC/B,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,MAAM,WAAW,KAAK;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE;QACL,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;QACnB,KAAK,EAAE,MAAM,CAAC;KACd,CAAC;CACF;AAED,MAAM,MAAM,UAAU,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,WAAW;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,GAAG,CAAC,WAAW,GAAG,YAAY,CAAC,EAAE,CAAC;IACjD,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,gBAAgB;IAChC,IAAI,EAAE,WAAW,CAAC;IAClB,OAAO,EAAE,CAAC,WAAW,GAAG,eAAe,GAAG,QAAQ,CAAC,EAAE,CAAC;IACtD,GAAG,EAAE,GAAG,CAAC;IACT,QAAQ,EAAE,QAAQ,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,KAAK,CAAC;IACb,UAAU,EAAE,UAAU,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,iBAAiB,CAAC,QAAQ,GAAG,GAAG;IAChD,IAAI,EAAE,YAAY,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,CAAC,WAAW,GAAG,YAAY,CAAC,EAAE,CAAC;IACxC,OAAO,CAAC,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,OAAO,GAAG,WAAW,GAAG,gBAAgB,GAAG,iBAAiB,CAAC;AAEzE,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAEjD,MAAM,WAAW,IAAI,CAAC,WAAW,SAAS,OAAO,GAAG,OAAO;IAC1D,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,WAAW,CAAC;CACxB;AAED,MAAM,WAAW,OAAO;IACvB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,OAAO,EAAE,CAAC;IACpB,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;CACf;AAED,MAAM,MAAM,qBAAqB,GAC9B;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC5C;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACvE;IAAE,IAAI,EAAE,YAAY,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACtF;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC3E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC3E;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC1F;IAAE,IAAI,EAAE,cAAc,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,QAAQ,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GAC7F;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,SAAS,CAAC,CAAC;IAAC,OAAO,EAAE,gBAAgB,CAAA;CAAE,GACvG;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,SAAS,GAAG,OAAO,CAAC,CAAC;IAAC,KAAK,EAAE,gBAAgB,CAAA;CAAE,CAAC;AAEhG;;;GAGG;AACH,MAAM,WAAW,YAAY;IAC5B,wFAAwF;IACxF,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,yGAAyG;IACzG,qBAAqB,CAAC,EAAE,OAAO,CAAC;IAChC,yFAAyF;IACzF,uBAAuB,CAAC,EAAE,OAAO,CAAC;IAClC,qIAAqI;IACrI,wBAAwB,CAAC,EAAE,OAAO,CAAC;IACnC,0EAA0E;IAC1E,cAAc,CAAC,EAAE,uBAAuB,GAAG,YAAY,CAAC;IACxD,sFAAsF;IACtF,sBAAsB,CAAC,EAAE,OAAO,CAAC;IACjC,2HAA2H;IAC3H,gCAAgC,CAAC,EAAE,OAAO,CAAC;IAC3C,4HAA4H;IAC5H,sBAAsB,CAAC,EAAE,OAAO,CAAC;IACjC,kIAAkI;IAClI,sBAAsB,CAAC,EAAE,OAAO,CAAC;CACjC;AAGD,MAAM,WAAW,KAAK,CAAC,IAAI,SAAS,GAAG;IACtC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,IAAI,CAAC;IACV,QAAQ,EAAE,QAAQ,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,OAAO,CAAC;IACnB,KAAK,EAAE,CAAC,MAAM,GAAG,OAAO,CAAC,EAAE,CAAC;IAC5B,IAAI,EAAE;QACL,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;KACnB,CAAC;IACF,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,kGAAkG;IAClG,MAAM,CAAC,EAAE,IAAI,SAAS,oBAAoB,GAAG,YAAY,GAAG,KAAK,CAAC;CAClE","sourcesContent":["import type { BedrockOptions } from \"./providers/amazon-bedrock.js\";\nimport type { AnthropicOptions } from \"./providers/anthropic.js\";\nimport type { GoogleOptions } from \"./providers/google.js\";\nimport type { GoogleGeminiCliOptions } from \"./providers/google-gemini-cli.js\";\nimport type { GoogleVertexOptions } from \"./providers/google-vertex.js\";\nimport type { OpenAICodexResponsesOptions } from \"./providers/openai-codex-responses.js\";\nimport type { OpenAICompletionsOptions } from \"./providers/openai-completions.js\";\nimport type { OpenAIResponsesOptions } from \"./providers/openai-responses.js\";\nimport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type Api =\n\t| \"openai-completions\"\n\t| \"openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\n\nexport interface ApiOptionsMap {\n\t\"anthropic-messages\": AnthropicOptions;\n\t\"bedrock-converse-stream\": BedrockOptions;\n\t\"openai-completions\": OpenAICompletionsOptions;\n\t\"openai-responses\": OpenAIResponsesOptions;\n\t\"openai-codex-responses\": OpenAICodexResponsesOptions;\n\t\"google-generative-ai\": GoogleOptions;\n\t\"google-gemini-cli\": GoogleGeminiCliOptions;\n\t\"google-vertex\": GoogleVertexOptions;\n}\n\n// Compile-time exhaustiveness check - this will fail if ApiOptionsMap doesn't have all KnownApi keys\ntype _CheckExhaustive = ApiOptionsMap extends Record<Api, StreamOptions>\n\t? Record<Api, StreamOptions> extends ApiOptionsMap\n\t\t? true\n\t\t: [\"ApiOptionsMap is missing some KnownApi values\", Exclude<Api, keyof ApiOptionsMap>]\n\t: [\"ApiOptionsMap doesn't extend Record<KnownApi, StreamOptions>\"];\nconst _exhaustive: _CheckExhaustive = true;\n\n// Helper type to get options for a specific API\nexport type OptionsForApi<TApi extends Api> = ApiOptionsMap[TApi];\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"openai-codex\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"zai\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"opencode\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\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 interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\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}\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n}\n\n// Generic StreamFunction with typed options\nexport type StreamFunction<TApi extends Api> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions: OptionsForApi<TApi>,\n) => AssistantMessageEventStream;\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, the message ID\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"@sinclair/typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for openai-completions API.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Whether tool call IDs must be normalized to Mistral format (exactly 9 alphanumeric chars). Default: auto-detected from URL. */\n\trequiresMistralToolIds?: boolean;\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: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/** Compatibility overrides for openai-completions API. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\" ? OpenAICompat : never;\n}\n"]}

package/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAoCA,MAAM,WAAW,GAAqB,IAAI,CAAC","sourcesContent":["import type { AnthropicOptions } from \"./providers/anthropic.js\";\nimport type { GoogleOptions } from \"./providers/google.js\";\nimport type { GoogleGeminiCliOptions } from \"./providers/google-gemini-cli.js\";\nimport type { GoogleVertexOptions } from \"./providers/google-vertex.js\";\nimport type { OpenAICodexResponsesOptions } from \"./providers/openai-codex-responses.js\";\nimport type { OpenAICompletionsOptions } from \"./providers/openai-completions.js\";\nimport type { OpenAIResponsesOptions } from \"./providers/openai-responses.js\";\nimport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type Api =\n\t| \"openai-completions\"\n\t| \"openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\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\"google-generative-ai\": GoogleOptions;\n\t\"google-gemini-cli\": GoogleGeminiCliOptions;\n\t\"google-vertex\": GoogleVertexOptions;\n}\n\n// Compile-time exhaustiveness check - this will fail if ApiOptionsMap doesn't have all KnownApi keys\ntype _CheckExhaustive = ApiOptionsMap extends Record<Api, StreamOptions>\n\t? Record<Api, StreamOptions> extends ApiOptionsMap\n\t\t? true\n\t\t: [\"ApiOptionsMap is missing some KnownApi values\", Exclude<Api, keyof ApiOptionsMap>]\n\t: [\"ApiOptionsMap doesn't extend Record<KnownApi, StreamOptions>\"];\nconst _exhaustive: _CheckExhaustive = true;\n\n// Helper type to get options for a specific API\nexport type OptionsForApi<TApi extends Api> = ApiOptionsMap[TApi];\n\nexport type KnownProvider =\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"openai-codex\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"zai\"\n\t| \"mistral\"\n\t| \"opencode\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\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 interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\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}\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n}\n\n// Generic StreamFunction with typed options\nexport type StreamFunction<TApi extends Api> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions: OptionsForApi<TApi>,\n) => AssistantMessageEventStream;\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, the message ID\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"@sinclair/typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for openai-completions API.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Whether tool call IDs must be normalized to Mistral format (exactly 9 alphanumeric chars). Default: auto-detected from URL. */\n\trequiresMistralToolIds?: boolean;\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: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/** Compatibility overrides for openai-completions API. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\" ? OpenAICompat : never;\n}\n"]}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAuCA,MAAM,WAAW,GAAqB,IAAI,CAAC","sourcesContent":["import type { BedrockOptions } from \"./providers/amazon-bedrock.js\";\nimport type { AnthropicOptions } from \"./providers/anthropic.js\";\nimport type { GoogleOptions } from \"./providers/google.js\";\nimport type { GoogleGeminiCliOptions } from \"./providers/google-gemini-cli.js\";\nimport type { GoogleVertexOptions } from \"./providers/google-vertex.js\";\nimport type { OpenAICodexResponsesOptions } from \"./providers/openai-codex-responses.js\";\nimport type { OpenAICompletionsOptions } from \"./providers/openai-completions.js\";\nimport type { OpenAIResponsesOptions } from \"./providers/openai-responses.js\";\nimport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type { AssistantMessageEventStream } from \"./utils/event-stream.js\";\n\nexport type Api =\n\t| \"openai-completions\"\n\t| \"openai-responses\"\n\t| \"openai-codex-responses\"\n\t| \"anthropic-messages\"\n\t| \"bedrock-converse-stream\"\n\t| \"google-generative-ai\"\n\t| \"google-gemini-cli\"\n\t| \"google-vertex\";\n\nexport interface ApiOptionsMap {\n\t\"anthropic-messages\": AnthropicOptions;\n\t\"bedrock-converse-stream\": BedrockOptions;\n\t\"openai-completions\": OpenAICompletionsOptions;\n\t\"openai-responses\": OpenAIResponsesOptions;\n\t\"openai-codex-responses\": OpenAICodexResponsesOptions;\n\t\"google-generative-ai\": GoogleOptions;\n\t\"google-gemini-cli\": GoogleGeminiCliOptions;\n\t\"google-vertex\": GoogleVertexOptions;\n}\n\n// Compile-time exhaustiveness check - this will fail if ApiOptionsMap doesn't have all KnownApi keys\ntype _CheckExhaustive = ApiOptionsMap extends Record<Api, StreamOptions>\n\t? Record<Api, StreamOptions> extends ApiOptionsMap\n\t\t? true\n\t\t: [\"ApiOptionsMap is missing some KnownApi values\", Exclude<Api, keyof ApiOptionsMap>]\n\t: [\"ApiOptionsMap doesn't extend Record<KnownApi, StreamOptions>\"];\nconst _exhaustive: _CheckExhaustive = true;\n\n// Helper type to get options for a specific API\nexport type OptionsForApi<TApi extends Api> = ApiOptionsMap[TApi];\n\nexport type KnownProvider =\n\t| \"amazon-bedrock\"\n\t| \"anthropic\"\n\t| \"google\"\n\t| \"google-gemini-cli\"\n\t| \"google-antigravity\"\n\t| \"google-vertex\"\n\t| \"openai\"\n\t| \"openai-codex\"\n\t| \"github-copilot\"\n\t| \"xai\"\n\t| \"groq\"\n\t| \"cerebras\"\n\t| \"openrouter\"\n\t| \"zai\"\n\t| \"mistral\"\n\t| \"minimax\"\n\t| \"opencode\";\nexport type Provider = KnownProvider | string;\n\nexport type ThinkingLevel = \"minimal\" | \"low\" | \"medium\" | \"high\" | \"xhigh\";\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 interface StreamOptions {\n\ttemperature?: number;\n\tmaxTokens?: number;\n\tsignal?: AbortSignal;\n\tapiKey?: string;\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}\n\n// Unified options with reasoning passed to streamSimple() and completeSimple()\nexport interface SimpleStreamOptions extends StreamOptions {\n\treasoning?: ThinkingLevel;\n\t/** Custom token budgets for thinking levels (token-based providers only) */\n\tthinkingBudgets?: ThinkingBudgets;\n}\n\n// Generic StreamFunction with typed options\nexport type StreamFunction<TApi extends Api> = (\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions: OptionsForApi<TApi>,\n) => AssistantMessageEventStream;\n\nexport interface TextContent {\n\ttype: \"text\";\n\ttext: string;\n\ttextSignature?: string; // e.g., for OpenAI responses, the message ID\n}\n\nexport interface ThinkingContent {\n\ttype: \"thinking\";\n\tthinking: string;\n\tthinkingSignature?: string; // e.g., for OpenAI responses, the reasoning item ID\n}\n\nexport interface ImageContent {\n\ttype: \"image\";\n\tdata: string; // base64 encoded image data\n\tmimeType: string; // e.g., \"image/jpeg\", \"image/png\"\n}\n\nexport interface ToolCall {\n\ttype: \"toolCall\";\n\tid: string;\n\tname: string;\n\targuments: Record<string, any>;\n\tthoughtSignature?: string; // Google-specific: opaque signature for reusing thought context\n}\n\nexport interface Usage {\n\tinput: number;\n\toutput: number;\n\tcacheRead: number;\n\tcacheWrite: number;\n\ttotalTokens: number;\n\tcost: {\n\t\tinput: number;\n\t\toutput: number;\n\t\tcacheRead: number;\n\t\tcacheWrite: number;\n\t\ttotal: number;\n\t};\n}\n\nexport type StopReason = \"stop\" | \"length\" | \"toolUse\" | \"error\" | \"aborted\";\n\nexport interface UserMessage {\n\trole: \"user\";\n\tcontent: string | (TextContent | ImageContent)[];\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface AssistantMessage {\n\trole: \"assistant\";\n\tcontent: (TextContent | ThinkingContent | ToolCall)[];\n\tapi: Api;\n\tprovider: Provider;\n\tmodel: string;\n\tusage: Usage;\n\tstopReason: StopReason;\n\terrorMessage?: string;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport interface ToolResultMessage<TDetails = any> {\n\trole: \"toolResult\";\n\ttoolCallId: string;\n\ttoolName: string;\n\tcontent: (TextContent | ImageContent)[]; // Supports text and images\n\tdetails?: TDetails;\n\tisError: boolean;\n\ttimestamp: number; // Unix timestamp in milliseconds\n}\n\nexport type Message = UserMessage | AssistantMessage | ToolResultMessage;\n\nimport type { TSchema } from \"@sinclair/typebox\";\n\nexport interface Tool<TParameters extends TSchema = TSchema> {\n\tname: string;\n\tdescription: string;\n\tparameters: TParameters;\n}\n\nexport interface Context {\n\tsystemPrompt?: string;\n\tmessages: Message[];\n\ttools?: Tool[];\n}\n\nexport type AssistantMessageEvent =\n\t| { type: \"start\"; partial: AssistantMessage }\n\t| { type: \"text_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"text_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"text_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"thinking_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"thinking_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"thinking_end\"; contentIndex: number; content: string; partial: AssistantMessage }\n\t| { type: \"toolcall_start\"; contentIndex: number; partial: AssistantMessage }\n\t| { type: \"toolcall_delta\"; contentIndex: number; delta: string; partial: AssistantMessage }\n\t| { type: \"toolcall_end\"; contentIndex: number; toolCall: ToolCall; partial: AssistantMessage }\n\t| { type: \"done\"; reason: Extract<StopReason, \"stop\" | \"length\" | \"toolUse\">; message: AssistantMessage }\n\t| { type: \"error\"; reason: Extract<StopReason, \"aborted\" | \"error\">; error: AssistantMessage };\n\n/**\n * Compatibility settings for openai-completions API.\n * Use this to override URL-based auto-detection for custom providers.\n */\nexport interface OpenAICompat {\n\t/** Whether the provider supports the `store` field. Default: auto-detected from URL. */\n\tsupportsStore?: boolean;\n\t/** Whether the provider supports the `developer` role (vs `system`). Default: auto-detected from URL. */\n\tsupportsDeveloperRole?: boolean;\n\t/** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */\n\tsupportsReasoningEffort?: boolean;\n\t/** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */\n\tsupportsUsageInStreaming?: boolean;\n\t/** Which field to use for max tokens. Default: auto-detected from URL. */\n\tmaxTokensField?: \"max_completion_tokens\" | \"max_tokens\";\n\t/** Whether tool results require the `name` field. Default: auto-detected from URL. */\n\trequiresToolResultName?: boolean;\n\t/** Whether a user message after tool results requires an assistant message in between. Default: auto-detected from URL. */\n\trequiresAssistantAfterToolResult?: boolean;\n\t/** Whether thinking blocks must be converted to text blocks with <thinking> delimiters. Default: auto-detected from URL. */\n\trequiresThinkingAsText?: boolean;\n\t/** Whether tool call IDs must be normalized to Mistral format (exactly 9 alphanumeric chars). Default: auto-detected from URL. */\n\trequiresMistralToolIds?: boolean;\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: Provider;\n\tbaseUrl: string;\n\treasoning: boolean;\n\tinput: (\"text\" | \"image\")[];\n\tcost: {\n\t\tinput: number; // $/million tokens\n\t\toutput: number; // $/million tokens\n\t\tcacheRead: number; // $/million tokens\n\t\tcacheWrite: number; // $/million tokens\n\t};\n\tcontextWindow: number;\n\tmaxTokens: number;\n\theaders?: Record<string, string>;\n\t/** Compatibility overrides for openai-completions API. If not set, auto-detected from baseUrl. */\n\tcompat?: TApi extends \"openai-completions\" ? OpenAICompat : never;\n}\n"]}

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

@@ -1 +1 @@
-{"version":3,"file":"overflow.d.ts","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAuCpD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2CG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,gBAAgB,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,OAAO,CAwB5F;AAED;;GAEG;AACH,wBAAgB,mBAAmB,IAAI,MAAM,EAAE,CAE9C","sourcesContent":["import type { AssistantMessage } from \"../types.js\";\n\n/**\n * Regex patterns to detect context overflow errors from different providers.\n *\n * These patterns match error messages returned when the input exceeds\n * the model's context window.\n *\n * Provider-specific patterns (with example error messages):\n *\n * - Anthropic: \"prompt is too long: 213462 tokens > 200000 maximum\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - Google: \"The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)\"\n * - xAI: \"This model's maximum prompt length is 131072 but the request contains 537812 tokens\"\n * - Groq: \"Please reduce the length of the messages or completion\"\n * - OpenRouter: \"This endpoint's maximum context length is X tokens. However, you requested about Y tokens\"\n * - llama.cpp: \"the request exceeds the available context size, try increasing it\"\n * - LM Studio: \"tokens to keep from the initial prompt is greater than the context length\"\n * - GitHub Copilot: \"prompt token count of X exceeds the limit of Y\"\n * - Cerebras: Returns \"400 status code (no body)\" - handled separately below\n * - Mistral: Returns \"400 status code (no body)\" - handled separately below\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Ollama: Silently truncates input - not detectable via error message\n */\nconst OVERFLOW_PATTERNS = [\n\t/prompt is too long/i, // Anthropic\n\t/exceeds the context window/i, // OpenAI (Completions & Responses API)\n\t/input token count.*exceeds the maximum/i, // Google (Gemini)\n\t/maximum prompt length is \\d+/i, // xAI (Grok)\n\t/reduce the length of the messages/i, // Groq\n\t/maximum context length is \\d+ tokens/i, // OpenRouter (all backends)\n\t/exceeds the limit of \\d+/i, // GitHub Copilot\n\t/exceeds the available context size/i, // llama.cpp server\n\t/greater than the context length/i, // LM Studio\n\t/context[_ ]length[_ ]exceeded/i, // Generic fallback\n\t/too many tokens/i, // Generic fallback\n\t/token limit exceeded/i, // Generic fallback\n];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles two cases:\n * 1. Error-based overflow: Most providers return stopReason \"error\" with a\n *    specific error message pattern.\n * 2. Silent overflow: Some providers accept overflow requests and return\n *    successfully. For these, we check if usage.input exceeds the context window.\n *\n * ## Reliability by Provider\n *\n * **Reliable detection (returns error with detectable message):**\n * - Anthropic: \"prompt is too long: X tokens > Y maximum\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\"\n * - Google Gemini: \"input token count exceeds the maximum\"\n * - xAI (Grok): \"maximum prompt length is X but request contains Y\"\n * - Groq: \"reduce the length of the messages\"\n * - Cerebras: 400/413/429 status code (no body)\n * - Mistral: 400/413/429 status code (no body)\n * - OpenRouter (all backends): \"maximum context length is X tokens\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n *\n * **Unreliable detection:**\n * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),\n *   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.\n * - Ollama: Silently truncates input without error. Cannot be detected via this function.\n *   The response will have usage.input < expected, but we don't know the expected value.\n *\n * ## Custom Providers\n *\n * If you've added custom models via settings.json, this function may not detect\n * overflow errors from those providers. To add support:\n *\n * 1. Send a request that exceeds the model's context window\n * 2. Check the errorMessage in the response\n * 3. Create a regex pattern that matches the error\n * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or\n *    check the errorMessage yourself before calling this function\n *\n * @param message - The assistant message to check\n * @param contextWindow - Optional context window size for detecting silent overflow (z.ai)\n * @returns true if the message indicates a context overflow\n */\nexport function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean {\n\t// Case 1: Check error message patterns\n\tif (message.stopReason === \"error\" && message.errorMessage) {\n\t\t// Check known patterns\n\t\tif (OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\n\t\t// Cerebras and Mistral return 400/413/429 with no body - check for status code pattern\n\t\t// 429 can indicate token-based rate limiting which correlates with context overflow\n\t\tif (/^4(00|13|29)\\s*(status code)?\\s*\\(no body\\)/i.test(message.errorMessage)) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 2: Silent overflow (z.ai style) - successful but usage exceeds context\n\tif (contextWindow && message.stopReason === \"stop\") {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens > contextWindow) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n"]}
+{"version":3,"file":"overflow.d.ts","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AA0CpD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2CG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,gBAAgB,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,OAAO,CAwB5F;AAED;;GAEG;AACH,wBAAgB,mBAAmB,IAAI,MAAM,EAAE,CAE9C","sourcesContent":["import type { AssistantMessage } from \"../types.js\";\n\n/**\n * Regex patterns to detect context overflow errors from different providers.\n *\n * These patterns match error messages returned when the input exceeds\n * the model's context window.\n *\n * Provider-specific patterns (with example error messages):\n *\n * - Anthropic: \"prompt is too long: 213462 tokens > 200000 maximum\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - Google: \"The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)\"\n * - xAI: \"This model's maximum prompt length is 131072 but the request contains 537812 tokens\"\n * - Groq: \"Please reduce the length of the messages or completion\"\n * - OpenRouter: \"This endpoint's maximum context length is X tokens. However, you requested about Y tokens\"\n * - llama.cpp: \"the request exceeds the available context size, try increasing it\"\n * - LM Studio: \"tokens to keep from the initial prompt is greater than the context length\"\n * - GitHub Copilot: \"prompt token count of X exceeds the limit of Y\"\n * - MiniMax: \"invalid params, context window exceeds limit\"\n * - Cerebras: Returns \"400 status code (no body)\" - handled separately below\n * - Mistral: Returns \"400 status code (no body)\" - handled separately below\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Ollama: Silently truncates input - not detectable via error message\n */\nconst OVERFLOW_PATTERNS = [\n\t/prompt is too long/i, // Anthropic\n\t/input is too long for requested model/i, // Amazon Bedrock\n\t/exceeds the context window/i, // OpenAI (Completions & Responses API)\n\t/input token count.*exceeds the maximum/i, // Google (Gemini)\n\t/maximum prompt length is \\d+/i, // xAI (Grok)\n\t/reduce the length of the messages/i, // Groq\n\t/maximum context length is \\d+ tokens/i, // OpenRouter (all backends)\n\t/exceeds the limit of \\d+/i, // GitHub Copilot\n\t/exceeds the available context size/i, // llama.cpp server\n\t/greater than the context length/i, // LM Studio\n\t/context window exceeds limit/i, // MiniMax\n\t/context[_ ]length[_ ]exceeded/i, // Generic fallback\n\t/too many tokens/i, // Generic fallback\n\t/token limit exceeded/i, // Generic fallback\n];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles two cases:\n * 1. Error-based overflow: Most providers return stopReason \"error\" with a\n *    specific error message pattern.\n * 2. Silent overflow: Some providers accept overflow requests and return\n *    successfully. For these, we check if usage.input exceeds the context window.\n *\n * ## Reliability by Provider\n *\n * **Reliable detection (returns error with detectable message):**\n * - Anthropic: \"prompt is too long: X tokens > Y maximum\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\"\n * - Google Gemini: \"input token count exceeds the maximum\"\n * - xAI (Grok): \"maximum prompt length is X but request contains Y\"\n * - Groq: \"reduce the length of the messages\"\n * - Cerebras: 400/413/429 status code (no body)\n * - Mistral: 400/413/429 status code (no body)\n * - OpenRouter (all backends): \"maximum context length is X tokens\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n *\n * **Unreliable detection:**\n * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),\n *   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.\n * - Ollama: Silently truncates input without error. Cannot be detected via this function.\n *   The response will have usage.input < expected, but we don't know the expected value.\n *\n * ## Custom Providers\n *\n * If you've added custom models via settings.json, this function may not detect\n * overflow errors from those providers. To add support:\n *\n * 1. Send a request that exceeds the model's context window\n * 2. Check the errorMessage in the response\n * 3. Create a regex pattern that matches the error\n * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or\n *    check the errorMessage yourself before calling this function\n *\n * @param message - The assistant message to check\n * @param contextWindow - Optional context window size for detecting silent overflow (z.ai)\n * @returns true if the message indicates a context overflow\n */\nexport function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean {\n\t// Case 1: Check error message patterns\n\tif (message.stopReason === \"error\" && message.errorMessage) {\n\t\t// Check known patterns\n\t\tif (OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\n\t\t// Cerebras and Mistral return 400/413/429 with no body - check for status code pattern\n\t\t// 429 can indicate token-based rate limiting which correlates with context overflow\n\t\tif (/^4(00|13|29)\\s*(status code)?\\s*\\(no body\\)/i.test(message.errorMessage)) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 2: Silent overflow (z.ai style) - successful but usage exceeds context\n\tif (contextWindow && message.stopReason === \"stop\") {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens > contextWindow) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n"]}

package/dist/utils/overflow.js

@@ -15,6 +15,7 @@
  * - llama.cpp: "the request exceeds the available context size, try increasing it"
  * - LM Studio: "tokens to keep from the initial prompt is greater than the context length"
  * - GitHub Copilot: "prompt token count of X exceeds the limit of Y"
+ * - MiniMax: "invalid params, context window exceeds limit"
  * - Cerebras: Returns "400 status code (no body)" - handled separately below
  * - Mistral: Returns "400 status code (no body)" - handled separately below
  * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow
@@ -22,6 +23,7 @@
  */
 const OVERFLOW_PATTERNS = [
     /prompt is too long/i, // Anthropic
+    /input is too long for requested model/i, // Amazon Bedrock
     /exceeds the context window/i, // OpenAI (Completions & Responses API)
     /input token count.*exceeds the maximum/i, // Google (Gemini)
     /maximum prompt length is \d+/i, // xAI (Grok)
@@ -30,6 +32,7 @@ const OVERFLOW_PATTERNS = [
     /exceeds the limit of \d+/i, // GitHub Copilot
     /exceeds the available context size/i, // llama.cpp server
     /greater than the context length/i, // LM Studio
+    /context window exceeds limit/i, // MiniMax
     /context[_ ]length[_ ]exceeded/i, // Generic fallback
     /too many tokens/i, // Generic fallback
     /token limit exceeded/i, // Generic fallback

package/dist/utils/overflow.js.map

@@ -1 +1 @@
-{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,iBAAiB,GAAG;IACzB,qBAAqB,EAAE,YAAY;IACnC,6BAA6B,EAAE,uCAAuC;IACtE,yCAAyC,EAAE,kBAAkB;IAC7D,+BAA+B,EAAE,aAAa;IAC9C,oCAAoC,EAAE,OAAO;IAC7C,uCAAuC,EAAE,4BAA4B;IACrE,2BAA2B,EAAE,iBAAiB;IAC9C,qCAAqC,EAAE,mBAAmB;IAC1D,kCAAkC,EAAE,YAAY;IAChD,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;CAC5C,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2CG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB,EAAW;IAC7F,uCAAuC;IACvC,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QAC5D,uBAAuB;QACvB,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,EAAE,CAAC;YAClE,OAAO,IAAI,CAAC;QACb,CAAC;QAED,uFAAuF;QACvF,oFAAoF;QACpF,IAAI,8CAA8C,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;YAC/E,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,8EAA8E;IAC9E,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,MAAM,EAAE,CAAC;QACpD,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,GAAG,aAAa,EAAE,CAAC;YACjC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AAAA,CACb;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,GAAa;IAC/C,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAAA,CAC9B","sourcesContent":["import type { AssistantMessage } from \"../types.js\";\n\n/**\n * Regex patterns to detect context overflow errors from different providers.\n *\n * These patterns match error messages returned when the input exceeds\n * the model's context window.\n *\n * Provider-specific patterns (with example error messages):\n *\n * - Anthropic: \"prompt is too long: 213462 tokens > 200000 maximum\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - Google: \"The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)\"\n * - xAI: \"This model's maximum prompt length is 131072 but the request contains 537812 tokens\"\n * - Groq: \"Please reduce the length of the messages or completion\"\n * - OpenRouter: \"This endpoint's maximum context length is X tokens. However, you requested about Y tokens\"\n * - llama.cpp: \"the request exceeds the available context size, try increasing it\"\n * - LM Studio: \"tokens to keep from the initial prompt is greater than the context length\"\n * - GitHub Copilot: \"prompt token count of X exceeds the limit of Y\"\n * - Cerebras: Returns \"400 status code (no body)\" - handled separately below\n * - Mistral: Returns \"400 status code (no body)\" - handled separately below\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Ollama: Silently truncates input - not detectable via error message\n */\nconst OVERFLOW_PATTERNS = [\n\t/prompt is too long/i, // Anthropic\n\t/exceeds the context window/i, // OpenAI (Completions & Responses API)\n\t/input token count.*exceeds the maximum/i, // Google (Gemini)\n\t/maximum prompt length is \\d+/i, // xAI (Grok)\n\t/reduce the length of the messages/i, // Groq\n\t/maximum context length is \\d+ tokens/i, // OpenRouter (all backends)\n\t/exceeds the limit of \\d+/i, // GitHub Copilot\n\t/exceeds the available context size/i, // llama.cpp server\n\t/greater than the context length/i, // LM Studio\n\t/context[_ ]length[_ ]exceeded/i, // Generic fallback\n\t/too many tokens/i, // Generic fallback\n\t/token limit exceeded/i, // Generic fallback\n];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles two cases:\n * 1. Error-based overflow: Most providers return stopReason \"error\" with a\n *    specific error message pattern.\n * 2. Silent overflow: Some providers accept overflow requests and return\n *    successfully. For these, we check if usage.input exceeds the context window.\n *\n * ## Reliability by Provider\n *\n * **Reliable detection (returns error with detectable message):**\n * - Anthropic: \"prompt is too long: X tokens > Y maximum\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\"\n * - Google Gemini: \"input token count exceeds the maximum\"\n * - xAI (Grok): \"maximum prompt length is X but request contains Y\"\n * - Groq: \"reduce the length of the messages\"\n * - Cerebras: 400/413/429 status code (no body)\n * - Mistral: 400/413/429 status code (no body)\n * - OpenRouter (all backends): \"maximum context length is X tokens\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n *\n * **Unreliable detection:**\n * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),\n *   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.\n * - Ollama: Silently truncates input without error. Cannot be detected via this function.\n *   The response will have usage.input < expected, but we don't know the expected value.\n *\n * ## Custom Providers\n *\n * If you've added custom models via settings.json, this function may not detect\n * overflow errors from those providers. To add support:\n *\n * 1. Send a request that exceeds the model's context window\n * 2. Check the errorMessage in the response\n * 3. Create a regex pattern that matches the error\n * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or\n *    check the errorMessage yourself before calling this function\n *\n * @param message - The assistant message to check\n * @param contextWindow - Optional context window size for detecting silent overflow (z.ai)\n * @returns true if the message indicates a context overflow\n */\nexport function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean {\n\t// Case 1: Check error message patterns\n\tif (message.stopReason === \"error\" && message.errorMessage) {\n\t\t// Check known patterns\n\t\tif (OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\n\t\t// Cerebras and Mistral return 400/413/429 with no body - check for status code pattern\n\t\t// 429 can indicate token-based rate limiting which correlates with context overflow\n\t\tif (/^4(00|13|29)\\s*(status code)?\\s*\\(no body\\)/i.test(message.errorMessage)) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 2: Silent overflow (z.ai style) - successful but usage exceeds context\n\tif (contextWindow && message.stopReason === \"stop\") {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens > contextWindow) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n"]}
+{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,MAAM,iBAAiB,GAAG;IACzB,qBAAqB,EAAE,YAAY;IACnC,wCAAwC,EAAE,iBAAiB;IAC3D,6BAA6B,EAAE,uCAAuC;IACtE,yCAAyC,EAAE,kBAAkB;IAC7D,+BAA+B,EAAE,aAAa;IAC9C,oCAAoC,EAAE,OAAO;IAC7C,uCAAuC,EAAE,4BAA4B;IACrE,2BAA2B,EAAE,iBAAiB;IAC9C,qCAAqC,EAAE,mBAAmB;IAC1D,kCAAkC,EAAE,YAAY;IAChD,+BAA+B,EAAE,UAAU;IAC3C,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;CAC5C,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2CG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB,EAAW;IAC7F,uCAAuC;IACvC,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QAC5D,uBAAuB;QACvB,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,EAAE,CAAC;YAClE,OAAO,IAAI,CAAC;QACb,CAAC;QAED,uFAAuF;QACvF,oFAAoF;QACpF,IAAI,8CAA8C,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;YAC/E,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,8EAA8E;IAC9E,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,MAAM,EAAE,CAAC;QACpD,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,GAAG,aAAa,EAAE,CAAC;YACjC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AAAA,CACb;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,GAAa;IAC/C,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAAA,CAC9B","sourcesContent":["import type { AssistantMessage } from \"../types.js\";\n\n/**\n * Regex patterns to detect context overflow errors from different providers.\n *\n * These patterns match error messages returned when the input exceeds\n * the model's context window.\n *\n * Provider-specific patterns (with example error messages):\n *\n * - Anthropic: \"prompt is too long: 213462 tokens > 200000 maximum\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - Google: \"The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)\"\n * - xAI: \"This model's maximum prompt length is 131072 but the request contains 537812 tokens\"\n * - Groq: \"Please reduce the length of the messages or completion\"\n * - OpenRouter: \"This endpoint's maximum context length is X tokens. However, you requested about Y tokens\"\n * - llama.cpp: \"the request exceeds the available context size, try increasing it\"\n * - LM Studio: \"tokens to keep from the initial prompt is greater than the context length\"\n * - GitHub Copilot: \"prompt token count of X exceeds the limit of Y\"\n * - MiniMax: \"invalid params, context window exceeds limit\"\n * - Cerebras: Returns \"400 status code (no body)\" - handled separately below\n * - Mistral: Returns \"400 status code (no body)\" - handled separately below\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Ollama: Silently truncates input - not detectable via error message\n */\nconst OVERFLOW_PATTERNS = [\n\t/prompt is too long/i, // Anthropic\n\t/input is too long for requested model/i, // Amazon Bedrock\n\t/exceeds the context window/i, // OpenAI (Completions & Responses API)\n\t/input token count.*exceeds the maximum/i, // Google (Gemini)\n\t/maximum prompt length is \\d+/i, // xAI (Grok)\n\t/reduce the length of the messages/i, // Groq\n\t/maximum context length is \\d+ tokens/i, // OpenRouter (all backends)\n\t/exceeds the limit of \\d+/i, // GitHub Copilot\n\t/exceeds the available context size/i, // llama.cpp server\n\t/greater than the context length/i, // LM Studio\n\t/context window exceeds limit/i, // MiniMax\n\t/context[_ ]length[_ ]exceeded/i, // Generic fallback\n\t/too many tokens/i, // Generic fallback\n\t/token limit exceeded/i, // Generic fallback\n];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles two cases:\n * 1. Error-based overflow: Most providers return stopReason \"error\" with a\n *    specific error message pattern.\n * 2. Silent overflow: Some providers accept overflow requests and return\n *    successfully. For these, we check if usage.input exceeds the context window.\n *\n * ## Reliability by Provider\n *\n * **Reliable detection (returns error with detectable message):**\n * - Anthropic: \"prompt is too long: X tokens > Y maximum\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\"\n * - Google Gemini: \"input token count exceeds the maximum\"\n * - xAI (Grok): \"maximum prompt length is X but request contains Y\"\n * - Groq: \"reduce the length of the messages\"\n * - Cerebras: 400/413/429 status code (no body)\n * - Mistral: 400/413/429 status code (no body)\n * - OpenRouter (all backends): \"maximum context length is X tokens\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n *\n * **Unreliable detection:**\n * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),\n *   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.\n * - Ollama: Silently truncates input without error. Cannot be detected via this function.\n *   The response will have usage.input < expected, but we don't know the expected value.\n *\n * ## Custom Providers\n *\n * If you've added custom models via settings.json, this function may not detect\n * overflow errors from those providers. To add support:\n *\n * 1. Send a request that exceeds the model's context window\n * 2. Check the errorMessage in the response\n * 3. Create a regex pattern that matches the error\n * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or\n *    check the errorMessage yourself before calling this function\n *\n * @param message - The assistant message to check\n * @param contextWindow - Optional context window size for detecting silent overflow (z.ai)\n * @returns true if the message indicates a context overflow\n */\nexport function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean {\n\t// Case 1: Check error message patterns\n\tif (message.stopReason === \"error\" && message.errorMessage) {\n\t\t// Check known patterns\n\t\tif (OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\n\t\t// Cerebras and Mistral return 400/413/429 with no body - check for status code pattern\n\t\t// 429 can indicate token-based rate limiting which correlates with context overflow\n\t\tif (/^4(00|13|29)\\s*(status code)?\\s*\\(no body\\)/i.test(message.errorMessage)) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 2: Silent overflow (z.ai style) - successful but usage exceeds context\n\tif (contextWindow && message.stopReason === \"stop\") {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens > contextWindow) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n"]}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@mariozechner/pi-ai",
-	"version": "0.44.0",
+	"version": "0.45.0",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -23,6 +23,7 @@
 	},
 	"dependencies": {
 		"@anthropic-ai/sdk": "0.71.2",
+		"@aws-sdk/client-bedrock-runtime": "^3.966.0",
 		"@google/genai": "1.34.0",
 		"@mistralai/mistralai": "1.10.0",
 		"@sinclair/typebox": "^0.34.41",
@@ -39,6 +40,7 @@
 		"openai",
 		"anthropic",
 		"gemini",
+		"bedrock",
 		"unified",
 		"api"
 	],