@code-yeongyu/senpi-ai 2026.9.28 → 2026.9.29-2

package/dist/utils/overflow.js.map

@@ -1 +1 @@
-{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,MAAM,iBAAiB,GAAG;IACzB,6BAA6B,EAAE,qEAAqE;IACpG,qBAAqB,EAAE,2BAA2B;IAClD,oBAAoB,EAAE,kDAAkD;IACxE,wCAAwC,EAAE,iBAAiB;IAC3D,yDAAyD,EAAE,uCAAuC;IAClG,4FAA4F,EAAE,sCAAsC;IACpI,yCAAyC,EAAE,kBAAkB;IAC7D,+BAA+B,EAAE,aAAa;IAC9C,oCAAoC,EAAE,OAAO;IAC7C,uCAAuC,EAAE,6BAA6B;IACtE,kEAAkE,EAAE,sBAAsB;IAC1F,iFAAiF,EAAE,cAAc;IACjG,2BAA2B,EAAE,iBAAiB;IAC9C,qCAAqC,EAAE,mBAAmB;IAC1D,kCAAkC,EAAE,YAAY;IAChD,+BAA+B,EAAE,UAAU;IAC3C,6BAA6B,EAAE,kBAAkB;IACjD,sDAAsD,EAAE,UAAU;IAClE,+EAA+E,EAAE,aAAa;IAC9F,gCAAgC,EAAE,yDAAyD;IAC3F,oDAAoD,EAAE,iCAAiC;IACvF,kCAAkC,EAAE,8BAA8B;IAClE,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;IAC5C,+CAA+C,EAAE,iCAAiC;IAClF,0DAA0D,EAAE,4UAA4U;IACxY,qGAAqG,EAAE,+GAA+G;IACtN,0EAA0E,EAAE,+DAA+D;CAC3I,CAAC;AAEF;;;;;;;;;;GAUG;AACH,MAAM,qBAAqB,GAAG;IAC7B,2CAA2C,EAAE,oFAAoF;IACjI,aAAa,EAAE,wBAAwB;IACvC,oBAAoB,EAAE,yBAAyB;IAC/C,qCAAqC,EAAE,sDAAsD;IAC7F,UAAU,EAAE,iCAAiC;IAC7C,UAAU,EAAE,mCAAmC;IAC/C,iBAAiB,EAAE,qCAAqC;IACxD,wBAAwB,EAAE,iCAAiC;IAC3D,QAAQ,EAAE,kBAAkB;IAC5B,kBAAkB,EAAE,iCAAiC;IACrD,aAAa,EAAE,sCAAsC;CACrD,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,0BAA0B,GAAG,sBAAsB,CAAC;AAE1D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqDG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB;IAClF,uCAAuC;IACvC,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QAC5D,oFAAoF;QACpF,MAAM,aAAa,GAAG,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,CAAC;QACvF,IAAI,CAAC,aAAa,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,EAAE,CAAC;YACpF,OAAO,IAAI,CAAC;QACb,CAAC;QACD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;QAC5B,MAAM,gBAAgB,GACrB,CAAC,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;QAC5F,IACC,CAAC,aAAa;YACd,gBAAgB;YAChB,0BAA0B,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC;YACrD,CAAC,aAAa,KAAK,SAAS,IAAI,aAAa,IAAI,CAAC,IAAI,oBAAoB,CAAC,OAAO,CAAC,IAAI,aAAa,GAAG,GAAG,CAAC,EAC1G,CAAC;YACF,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,sFAAsF;IACtF,qFAAqF;IACrF,gEAAgE;IAChE,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpF,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,IAAI,aAAa,GAAG,IAAI,EAAE,CAAC;YACzC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AACd,CAAC;AACD;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,OAAyB,EAAE,gBAAwB;IACtF,OAAO,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,gBAAgB,GAAG,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,gBAAgB,CAAC;AAC3G,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB;IAClC,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAC/B,CAAC;AAED,SAAS,oBAAoB,CAAC,OAE7B;IACA,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;IAC5B,IAAI,CAAC,KAAK;QAAE,OAAO,CAAC,CAAC;IACrB,OAAO,CACN,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,IAAI,CAAC,CAAC,CAChH,CAAC;AACH,CAAC;AAED,MAAM,UAAU,gCAAgC,CAC/C,OAIC,EACD,eAAuB;IAEvB,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;AAC7C,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,8BAA8B,CAC7C,OAIC,EACD,aAAqB;IAErB,MAAM,MAAM,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;IAC7C,OAAO,CACN,OAAO,CAAC,UAAU,KAAK,OAAO;QAC9B,0BAA0B,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC;QAC3D,aAAa,GAAG,CAAC;QACjB,MAAM,GAAG,CAAC;QACV,MAAM,GAAG,aAAa,GAAG,GAAG,CAC5B,CAAC;AACH,CAAC;AAED,MAAM,UAAU,kCAAkC,CAAC,OAIlD;IACA,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;AAC7C,CAAC;AAED,MAAM,UAAU,yCAAyC,CAAC,OAAkD;IAC3G,OAAO,OAAO,EAAE,eAAe,KAAK,IAAI,CAAC;AAC1C,CAAC;AAED,MAAM,UAAU,iCAAiC,CAAC,SAAkB,EAAE,YAAoB;IACzF,IAAI,SAAS;QAAE,OAAO,KAAK,CAAC;IAC5B,OAAO,qBAAqB,CAAC,IAAI,CAAC,YAAY,IAAI,EAAE,CAAC,CAAC;AACvD,CAAC;AAED,MAAM,UAAU,gCAAgC,CAC/C,QAAW,EACX,QAA4B,EAC5B,MAA0B;IAE1B,IAAI,MAAM,KAAK,UAAU;QAAE,OAAO,QAAQ,CAAC;IAC3C,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,kBAAkB;QAAE,OAAO,QAAQ,CAAC;IAC9E,OAAO,EAAE,GAAG,QAAQ,EAAE,gBAAgB,EAAE,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC;AACxE,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\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 * - Anthropic: \"413 {\\\"error\\\":{\\\"type\\\":\\\"request_too_large\\\",\\\"message\\\":\\\"Request exceeds the maximum size\\\"}}\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - OpenAI: \"Your input exceeds the model's context window\" / \"exceeds this model's context window\"\n * - OpenAI/LiteLLM: \"Requested token count exceeds the model's maximum context length of 131072 tokens\"\n * - OpenAI-compatible: \"Input length (265330) exceeds model's maximum context length (262144).\"\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 * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (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 * - Kimi For Coding: \"Your request exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - Cerebras: \"400/413 status code (no body)\"\n * - Gateways: \"413 Request body too large\" / \"Request Entity Too Large\" / \"Payload Too Large\" (byte-size overflow)\n * - kiro-lb gateways: \"Request payload is 1095225 bytes, over the 1085435 byte limit Kiro accepts.\" / \"Request payload is N tokens, over the M token limit Kiro accepts.\" (HTTP 400 local payload guard)\n * - Kiro upstream via kiro-lb: \"Model context limit reached. Conversation size exceeds model capacity.\" (CONTENT_LENGTH_EXCEEDS_THRESHOLD token overflow)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Xiaomi MiMo: Truncates input to fill contextWindow exactly, then returns finish_reason \"length\"\n *   with output=0 (no room left to generate). Detected via stopReason \"length\" + zero output +\n *   input filling the context window.\n * - DashScope/Qwen: \"Range of input length should be [1, X]\" (HTTP 400 invalid_parameter_error)\n * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n * - pi-ai pre-flight guard (api/context-room.ts): \"Context window exhausted: the conversation is estimated at X of Y tokens, ...\" - raised before any provider call\n */\nconst OVERFLOW_PATTERNS = [\n\t/^Context window exhausted: /, // pi-ai pre-flight guard: no answer room left, provider never called\n\t/prompt is too long/i, // Anthropic token overflow\n\t/request_too_large/i, // Anthropic request byte-size overflow (HTTP 413)\n\t/input is too long for requested model/i, // Amazon Bedrock\n\t/exceeds (?:(?:the|this) )?(?:model'?s )?context window/i, // OpenAI (Completions & Responses API)\n\t/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\\d,]+ tokens?|\\s*\\([\\d,]+\\))/i, // OpenAI-compatible proxies (LiteLLM)\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 (most backends)\n\t/exceeds (?:the )?maximum allowed input length of [\\d,]+ tokens?/i, // OpenRouter/Poolside\n\t/input \\(\\d+ tokens\\) is longer than the model'?s context length \\(\\d+ tokens\\)/i, // Together AI\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/exceeded model token limit/i, // Kimi For Coding\n\t/too large for model with \\d+ maximum context length/i, // Mistral\n\t/prompt has [\\d,]+ tokens?, but the configured context size is [\\d,]+ tokens?/i, // DS4 server\n\t/model_context_window_exceeded/i, // z.ai non-standard finish_reason surfaced as error text\n\t/prompt too long; exceeded (?:max )?context length/i, // Ollama explicit overflow error\n\t/range of input length should be/i, // DashScope / Qwen Token Plan\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\t/^4(?:00|13)\\s*(?:status code)?\\s*\\(no body\\)/i, // Cerebras: 400/413 with no body\n\t/(?:request[ _])?(?:body|entity|payload)[_ ]too[_ ]large/i, // Gateway HTTP 413 byte-size rejections (\"Request body too large\", \"Request Entity Too Large\", \"body_too_large\", \"Payload Too Large\"). Substring-anchored by design (JSON bodies lack an adjacent status code); a non-context size rejection (e.g. an oversized image) can over-match, which costs one bounded shrink-retry, never a wedge.\n\t/Request payload is \\d+ (?:bytes, over the \\d+ byte|tokens, over the \\d+ token) limit Kiro accepts\\./, // kiro-lb local byte/token payload guard (HTTP 400; Anthropic uses invalid_request_error, OpenAI uses detail).\n\t/Model context limit reached\\. Conversation size exceeds model capacity\\./, // kiro-lb enhancement of Kiro CONTENT_LENGTH_EXCEEDS_THRESHOLD\n];\n\n/**\n * Patterns that indicate non-overflow errors (e.g. rate limiting, server errors).\n * Error messages matching any of these are excluded from overflow detection\n * even if they also match an OVERFLOW_PATTERN.\n *\n * Example: Bedrock formats throttling errors as \"ThrottlingException: Too many tokens,\n * please wait before trying again.\" which would match the /too many tokens/i overflow\n * pattern without this exclusion. Token-quota / TPM messages such as \"Too many tokens\n * per minute\" or \"This request exceeds the limit of 30000 tokens per minute\" similarly\n * match generic overflow fallbacks and must stay on the rate-limit path.\n */\nconst NON_OVERFLOW_PATTERNS = [\n\t/^(Throttling error|Service unavailable):/i, // AWS Bedrock non-overflow errors (human-readable prefixes from formatBedrockError)\n\t/rate limit/i, // Generic rate limiting\n\t/too many requests/i, // Generic HTTP 429 style\n\t/tokens per (?:min|minute|hour|day)/i, // Token-quota windows (TPM/TPH/TPD), not context size\n\t/\\bTPM\\b/i, // Tokens-per-minute abbreviation\n\t/\\bRPM\\b/i, // Requests-per-minute abbreviation\n\t/quota exceeded/i, // Provider quota, not context window\n\t/retry (?:after|in) \\d/i, // Retry-after rate-limit wording\n\t/^429\\b/, // HTTP 429 prefix\n\t/status code 429/i, // HTTP 429 mentioned mid-message\n\t/overloaded/i, // Provider capacity, not context size\n];\n\n/**\n * Cursor's api2.cursor.sh surfaces a context overflow as a bare gRPC\n * `resource_exhausted` end-stream — the same wording its backend uses for\n * quota and poisoned-conversation rejections. Token evidence disambiguates:\n * a request that already streamed or billed tokens overflowed mid-flight,\n * while a zero-token rejection is a quota/conversation failure that must stay\n * on the rate-limit path (the cursor client rotates the conversation id for\n * those and retry handling supplies the backoff).\n */\nconst RESOURCE_EXHAUSTED_PATTERN = /resource.?exhausted/i;\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles three 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 * 3. Length-stop overflow: Xiaomi MiMo can return \"length\" with zero output when\n *    the input fills 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\" or \"request_too_large\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\", \"exceeds the model's context window\", \"exceeds this model's context window\", \"exceeds the model's maximum context length of X tokens\", or \"exceeds model's maximum context length (X)\"\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 status code (no body)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - OpenRouter (most backends): \"maximum context length is X tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n * - Kimi For Coding: \"exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - DashScope/Qwen: \"Range of input length should be [1, X]\"\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 * - Xiaomi MiMo: Truncates input to fit contextWindow then returns stopReason \"length\" with\n *   output=0. Pass contextWindow param to detect via the \"filled context + zero output\" signal.\n * - Ollama: May truncate input silently for some setups, but may also return explicit\n *   overflow errors that match the patterns above. Silent truncation still cannot be\n *   detected here because we do not know the expected token count.\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// Skip messages matching known non-overflow patterns (e.g. throttling / rate-limit)\n\t\tconst isNonOverflow = NON_OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!));\n\t\tif (!isNonOverflow && OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\t\tconst usage = message.usage;\n\t\tconst hasTokenEvidence =\n\t\t\t(usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite) > 0;\n\t\tif (\n\t\t\t!isNonOverflow &&\n\t\t\thasTokenEvidence &&\n\t\t\tRESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage) &&\n\t\t\t(contextWindow === undefined || contextWindow <= 0 || cursorZeroTokenCount(message) >= contextWindow * 0.5)\n\t\t) {\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\t// Case 3: Length-stop overflow (Xiaomi MiMo style) - server truncates oversized input\n\t// to fit the context window, leaving no room for output. Returns stopReason \"length\"\n\t// with output=0 and input+cacheRead filling the context window.\n\tif (contextWindow && message.stopReason === \"length\" && message.usage.output === 0) {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens >= contextWindow * 0.99) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n/**\n * Check whether a length stop ended below the caller or model's intended output limit.\n * Such responses may be caused by context pressure or provider-side truncation, so callers\n * can make one bounded compact-and-retry attempt. `desiredMaxOutput` must be the original\n * limit before any context-based clamping.\n */\nexport function isRecoverableLength(message: AssistantMessage, desiredMaxOutput: number): boolean {\n\treturn message.stopReason === \"length\" && desiredMaxOutput > 0 && message.usage.output < desiredMaxOutput;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n\nfunction cursorZeroTokenCount(message: {\n\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n}): number {\n\tconst usage = message.usage;\n\tif (!usage) return 0;\n\treturn (\n\t\tusage.totalTokens || (usage.input ?? 0) + (usage.output ?? 0) + (usage.cacheRead ?? 0) + (usage.cacheWrite ?? 0)\n\t);\n}\n\nexport function isCursorPayloadResourceExhausted(\n\tmessage: {\n\t\tstopReason?: string;\n\t\terrorMessage?: string;\n\t\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n\t},\n\t_estimateTokens: number,\n): boolean {\n\tif (message.stopReason !== \"error\" || !/resource.?exhausted/i.test(message.errorMessage || \"\")) {\n\t\treturn false;\n\t}\n\treturn !(cursorZeroTokenCount(message) > 0);\n}\n\n/**\n * Detects Cursor's verified usage-pool exhaustion signature: a token-bearing\n * `resource_exhausted` error while the conversation is well below the model\n * context window.\n */\nexport function isCursorQuotaResourceExhausted(\n\tmessage: {\n\t\tstopReason?: string;\n\t\terrorMessage?: string;\n\t\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n\t},\n\tcontextWindow: number,\n): boolean {\n\tconst tokens = cursorZeroTokenCount(message);\n\treturn (\n\t\tmessage.stopReason === \"error\" &&\n\t\tRESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage || \"\") &&\n\t\tcontextWindow > 0 &&\n\t\ttokens > 0 &&\n\t\ttokens < contextWindow * 0.5\n\t);\n}\n\nexport function isCursorZeroTokenResourceExhausted(message: {\n\tstopReason?: string;\n\terrorMessage?: string;\n\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n}): boolean {\n\tif (message.stopReason !== \"error\" || !/resource.?exhausted/i.test(message.errorMessage || \"\")) {\n\t\treturn false;\n\t}\n\treturn !(cursorZeroTokenCount(message) > 0);\n}\n\nexport function shouldSkipProviderFallbackForCursorZeroRe(options: { sameModelRemint?: boolean } | undefined): boolean {\n\treturn options?.sameModelRemint === true;\n}\n\nexport function shouldRetryOverflowWithoutCompact(compacted: boolean, errorMessage: string): boolean {\n\tif (compacted) return false;\n\treturn /Nothing to compact/i.test(errorMessage || \"\");\n}\n\nexport function cursorOverflowCompactionSettings<T extends { keepRecentTokens?: number; restorationEnabled?: boolean }>(\n\tsettings: T,\n\tprovider: string | undefined,\n\treason: string | undefined,\n): T {\n\tif (reason !== \"overflow\") return settings;\n\tif (provider !== \"cursor\" && provider !== \"cursor-cli-oauth\") return settings;\n\treturn { ...settings, keepRecentTokens: 0, restorationEnabled: false };\n}\n"]}
+{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAuCG;AACH,MAAM,iBAAiB,GAAG;IACzB,6BAA6B,EAAE,qEAAqE;IACpG,yEAAyE,EAAE,2EAA2E;IACtJ,qBAAqB,EAAE,2BAA2B;IAClD,oBAAoB,EAAE,kDAAkD;IACxE,wCAAwC,EAAE,iBAAiB;IAC3D,yDAAyD,EAAE,uCAAuC;IAClG,4FAA4F,EAAE,sCAAsC;IACpI,yCAAyC,EAAE,kBAAkB;IAC7D,+BAA+B,EAAE,aAAa;IAC9C,oCAAoC,EAAE,OAAO;IAC7C,uCAAuC,EAAE,6BAA6B;IACtE,kEAAkE,EAAE,sBAAsB;IAC1F,iFAAiF,EAAE,cAAc;IACjG,4DAA4D,EAAE,iBAAiB;IAC/E,qCAAqC,EAAE,mBAAmB;IAC1D,kCAAkC,EAAE,YAAY;IAChD,+BAA+B,EAAE,UAAU;IAC3C,6BAA6B,EAAE,kBAAkB;IACjD,sDAAsD,EAAE,UAAU;IAClE,+EAA+E,EAAE,aAAa;IAC9F,gCAAgC,EAAE,yDAAyD;IAC3F,oDAAoD,EAAE,iCAAiC;IACvF,kCAAkC,EAAE,8BAA8B;IAClE,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;IAC5C,+CAA+C,EAAE,iCAAiC;IAClF,0DAA0D,EAAE,4UAA4U;IACxY,qGAAqG,EAAE,+GAA+G;IACtN,0EAA0E,EAAE,+DAA+D;CAC3I,CAAC;AAEF;;;;;;;;;;GAUG;AACH,MAAM,qBAAqB,GAAG;IAC7B,2CAA2C,EAAE,oFAAoF;IACjI,aAAa,EAAE,wBAAwB;IACvC,oBAAoB,EAAE,yBAAyB;IAC/C,qCAAqC,EAAE,sDAAsD;IAC7F,UAAU,EAAE,iCAAiC;IAC7C,UAAU,EAAE,mCAAmC;IAC/C,iBAAiB,EAAE,qCAAqC;IACxD,wBAAwB,EAAE,iCAAiC;IAC3D,QAAQ,EAAE,kBAAkB;IAC5B,kBAAkB,EAAE,iCAAiC;IACrD,aAAa,EAAE,sCAAsC;CACrD,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,0BAA0B,GAAG,sBAAsB,CAAC;AAE1D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqDG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB;IAClF,uCAAuC;IACvC,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QAC5D,oFAAoF;QACpF,MAAM,aAAa,GAAG,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,CAAC;QACvF,IAAI,CAAC,aAAa,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,EAAE,CAAC;YACpF,OAAO,IAAI,CAAC;QACb,CAAC;QACD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;QAC5B,MAAM,gBAAgB,GACrB,CAAC,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;QAC5F,IACC,CAAC,aAAa;YACd,gBAAgB;YAChB,0BAA0B,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC;YACrD,CAAC,aAAa,KAAK,SAAS,IAAI,aAAa,IAAI,CAAC,IAAI,oBAAoB,CAAC,OAAO,CAAC,IAAI,aAAa,GAAG,GAAG,CAAC,EAC1G,CAAC;YACF,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,sFAAsF;IACtF,qFAAqF;IACrF,gEAAgE;IAChE,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpF,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,IAAI,aAAa,GAAG,IAAI,EAAE,CAAC;YACzC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AACd,CAAC;AACD;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,OAAyB,EAAE,gBAAwB;IACtF,OAAO,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,gBAAgB,GAAG,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,gBAAgB,CAAC;AAC3G,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB;IAClC,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAC/B,CAAC;AAED,SAAS,oBAAoB,CAAC,OAE7B;IACA,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;IAC5B,IAAI,CAAC,KAAK;QAAE,OAAO,CAAC,CAAC;IACrB,OAAO,CACN,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,IAAI,CAAC,CAAC,CAChH,CAAC;AACH,CAAC;AAED,MAAM,UAAU,gCAAgC,CAC/C,OAIC,EACD,eAAuB;IAEvB,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;AAC7C,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,8BAA8B,CAC7C,OAIC,EACD,aAAqB;IAErB,MAAM,MAAM,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;IAC7C,OAAO,CACN,OAAO,CAAC,UAAU,KAAK,OAAO;QAC9B,0BAA0B,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC;QAC3D,aAAa,GAAG,CAAC;QACjB,MAAM,GAAG,CAAC;QACV,MAAM,GAAG,aAAa,GAAG,GAAG,CAC5B,CAAC;AACH,CAAC;AAED,MAAM,UAAU,kCAAkC,CAAC,OAIlD;IACA,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;AAC7C,CAAC;AAED,MAAM,UAAU,yCAAyC,CAAC,OAAkD;IAC3G,OAAO,OAAO,EAAE,eAAe,KAAK,IAAI,CAAC;AAC1C,CAAC;AAED,MAAM,UAAU,iCAAiC,CAAC,SAAkB,EAAE,YAAoB;IACzF,IAAI,SAAS;QAAE,OAAO,KAAK,CAAC;IAC5B,OAAO,qBAAqB,CAAC,IAAI,CAAC,YAAY,IAAI,EAAE,CAAC,CAAC;AACvD,CAAC;AAED,MAAM,UAAU,gCAAgC,CAC/C,QAAW,EACX,QAA4B,EAC5B,MAA0B;IAE1B,IAAI,MAAM,KAAK,UAAU;QAAE,OAAO,QAAQ,CAAC;IAC3C,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,kBAAkB;QAAE,OAAO,QAAQ,CAAC;IAC9E,OAAO,EAAE,GAAG,QAAQ,EAAE,gBAAgB,EAAE,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC;AACxE,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\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 * - Anthropic: \"413 {\\\"error\\\":{\\\"type\\\":\\\"request_too_large\\\",\\\"message\\\":\\\"Request exceeds the maximum size\\\"}}\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - OpenAI: \"Your input exceeds the model's context window\" / \"exceeds this model's context window\"\n * - OpenAI/LiteLLM: \"Requested token count exceeds the model's maximum context length of 131072 tokens\"\n * - OpenAI-compatible: \"Input length (265330) exceeds model's maximum context length (262144).\"\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 * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (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 * - Kimi For Coding: \"Your request exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - Cerebras: \"400/413 status code (no body)\"\n * - Gateways: \"413 Request body too large\" / \"Request Entity Too Large\" / \"Payload Too Large\" (byte-size overflow)\n * - kiro-lb gateways: \"Request payload is 1095225 bytes, over the 1085435 byte limit Kiro accepts.\" / \"Request payload is N tokens, over the M token limit Kiro accepts.\" (HTTP 400 local payload guard)\n * - Kiro upstream via kiro-lb: \"Model context limit reached. Conversation size exceeds model capacity.\" (CONTENT_LENGTH_EXCEEDS_THRESHOLD token overflow)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Xiaomi MiMo: Truncates input to fill contextWindow exactly, then returns finish_reason \"length\"\n *   with output=0 (no room left to generate). Detected via stopReason \"length\" + zero output +\n *   input filling the context window.\n * - DashScope/Qwen: \"Range of input length should be [1, X]\" (HTTP 400 invalid_parameter_error)\n * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n * - pi-ai pre-flight guard (api/context-room.ts): \"Context window exhausted: the conversation is estimated at X of Y tokens, ...\" - raised before any provider call\n * - anthropic-subscription cold-seed budget: \"The conversation is too long to resend (about X tokens, limit Y). Compacting it and retrying.\" - raised before the re-send is dispatched\n */\nconst OVERFLOW_PATTERNS = [\n\t/^Context window exhausted: /, // pi-ai pre-flight guard: no answer room left, provider never called\n\t/^The conversation is too long to resend \\(about \\d+ tokens, limit \\d+\\)/, // anthropic-subscription cold-seed budget: re-send refused before dispatch\n\t/prompt is too long/i, // Anthropic token overflow\n\t/request_too_large/i, // Anthropic request byte-size overflow (HTTP 413)\n\t/input is too long for requested model/i, // Amazon Bedrock\n\t/exceeds (?:(?:the|this) )?(?:model'?s )?context window/i, // OpenAI (Completions & Responses API)\n\t/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\\d,]+ tokens?|\\s*\\([\\d,]+\\))/i, // OpenAI-compatible proxies (LiteLLM)\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 (most backends)\n\t/exceeds (?:the )?maximum allowed input length of [\\d,]+ tokens?/i, // OpenRouter/Poolside\n\t/input \\(\\d+ tokens\\) is longer than the model'?s context length \\(\\d+ tokens\\)/i, // Together AI\n\t/model_max_prompt_tokens_exceeded|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/exceeded model token limit/i, // Kimi For Coding\n\t/too large for model with \\d+ maximum context length/i, // Mistral\n\t/prompt has [\\d,]+ tokens?, but the configured context size is [\\d,]+ tokens?/i, // DS4 server\n\t/model_context_window_exceeded/i, // z.ai non-standard finish_reason surfaced as error text\n\t/prompt too long; exceeded (?:max )?context length/i, // Ollama explicit overflow error\n\t/range of input length should be/i, // DashScope / Qwen Token Plan\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\t/^4(?:00|13)\\s*(?:status code)?\\s*\\(no body\\)/i, // Cerebras: 400/413 with no body\n\t/(?:request[ _])?(?:body|entity|payload)[_ ]too[_ ]large/i, // Gateway HTTP 413 byte-size rejections (\"Request body too large\", \"Request Entity Too Large\", \"body_too_large\", \"Payload Too Large\"). Substring-anchored by design (JSON bodies lack an adjacent status code); a non-context size rejection (e.g. an oversized image) can over-match, which costs one bounded shrink-retry, never a wedge.\n\t/Request payload is \\d+ (?:bytes, over the \\d+ byte|tokens, over the \\d+ token) limit Kiro accepts\\./, // kiro-lb local byte/token payload guard (HTTP 400; Anthropic uses invalid_request_error, OpenAI uses detail).\n\t/Model context limit reached\\. Conversation size exceeds model capacity\\./, // kiro-lb enhancement of Kiro CONTENT_LENGTH_EXCEEDS_THRESHOLD\n];\n\n/**\n * Patterns that indicate non-overflow errors (e.g. rate limiting, server errors).\n * Error messages matching any of these are excluded from overflow detection\n * even if they also match an OVERFLOW_PATTERN.\n *\n * Example: Bedrock formats throttling errors as \"ThrottlingException: Too many tokens,\n * please wait before trying again.\" which would match the /too many tokens/i overflow\n * pattern without this exclusion. Token-quota / TPM messages such as \"Too many tokens\n * per minute\" or \"This request exceeds the limit of 30000 tokens per minute\" similarly\n * match generic overflow fallbacks and must stay on the rate-limit path.\n */\nconst NON_OVERFLOW_PATTERNS = [\n\t/^(Throttling error|Service unavailable):/i, // AWS Bedrock non-overflow errors (human-readable prefixes from formatBedrockError)\n\t/rate limit/i, // Generic rate limiting\n\t/too many requests/i, // Generic HTTP 429 style\n\t/tokens per (?:min|minute|hour|day)/i, // Token-quota windows (TPM/TPH/TPD), not context size\n\t/\\bTPM\\b/i, // Tokens-per-minute abbreviation\n\t/\\bRPM\\b/i, // Requests-per-minute abbreviation\n\t/quota exceeded/i, // Provider quota, not context window\n\t/retry (?:after|in) \\d/i, // Retry-after rate-limit wording\n\t/^429\\b/, // HTTP 429 prefix\n\t/status code 429/i, // HTTP 429 mentioned mid-message\n\t/overloaded/i, // Provider capacity, not context size\n];\n\n/**\n * Cursor's api2.cursor.sh surfaces a context overflow as a bare gRPC\n * `resource_exhausted` end-stream — the same wording its backend uses for\n * quota and poisoned-conversation rejections. Token evidence disambiguates:\n * a request that already streamed or billed tokens overflowed mid-flight,\n * while a zero-token rejection is a quota/conversation failure that must stay\n * on the rate-limit path (the cursor client rotates the conversation id for\n * those and retry handling supplies the backoff).\n */\nconst RESOURCE_EXHAUSTED_PATTERN = /resource.?exhausted/i;\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles three 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 * 3. Length-stop overflow: Xiaomi MiMo can return \"length\" with zero output when\n *    the input fills 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\" or \"request_too_large\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\", \"exceeds the model's context window\", \"exceeds this model's context window\", \"exceeds the model's maximum context length of X tokens\", or \"exceeds model's maximum context length (X)\"\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 status code (no body)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - OpenRouter (most backends): \"maximum context length is X tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n * - Kimi For Coding: \"exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - DashScope/Qwen: \"Range of input length should be [1, X]\"\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 * - Xiaomi MiMo: Truncates input to fit contextWindow then returns stopReason \"length\" with\n *   output=0. Pass contextWindow param to detect via the \"filled context + zero output\" signal.\n * - Ollama: May truncate input silently for some setups, but may also return explicit\n *   overflow errors that match the patterns above. Silent truncation still cannot be\n *   detected here because we do not know the expected token count.\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// Skip messages matching known non-overflow patterns (e.g. throttling / rate-limit)\n\t\tconst isNonOverflow = NON_OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!));\n\t\tif (!isNonOverflow && OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\t\tconst usage = message.usage;\n\t\tconst hasTokenEvidence =\n\t\t\t(usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite) > 0;\n\t\tif (\n\t\t\t!isNonOverflow &&\n\t\t\thasTokenEvidence &&\n\t\t\tRESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage) &&\n\t\t\t(contextWindow === undefined || contextWindow <= 0 || cursorZeroTokenCount(message) >= contextWindow * 0.5)\n\t\t) {\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\t// Case 3: Length-stop overflow (Xiaomi MiMo style) - server truncates oversized input\n\t// to fit the context window, leaving no room for output. Returns stopReason \"length\"\n\t// with output=0 and input+cacheRead filling the context window.\n\tif (contextWindow && message.stopReason === \"length\" && message.usage.output === 0) {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens >= contextWindow * 0.99) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n/**\n * Check whether a length stop ended below the caller or model's intended output limit.\n * Such responses may be caused by context pressure or provider-side truncation, so callers\n * can make one bounded compact-and-retry attempt. `desiredMaxOutput` must be the original\n * limit before any context-based clamping.\n */\nexport function isRecoverableLength(message: AssistantMessage, desiredMaxOutput: number): boolean {\n\treturn message.stopReason === \"length\" && desiredMaxOutput > 0 && message.usage.output < desiredMaxOutput;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n\nfunction cursorZeroTokenCount(message: {\n\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n}): number {\n\tconst usage = message.usage;\n\tif (!usage) return 0;\n\treturn (\n\t\tusage.totalTokens || (usage.input ?? 0) + (usage.output ?? 0) + (usage.cacheRead ?? 0) + (usage.cacheWrite ?? 0)\n\t);\n}\n\nexport function isCursorPayloadResourceExhausted(\n\tmessage: {\n\t\tstopReason?: string;\n\t\terrorMessage?: string;\n\t\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n\t},\n\t_estimateTokens: number,\n): boolean {\n\tif (message.stopReason !== \"error\" || !/resource.?exhausted/i.test(message.errorMessage || \"\")) {\n\t\treturn false;\n\t}\n\treturn !(cursorZeroTokenCount(message) > 0);\n}\n\n/**\n * Detects Cursor's verified usage-pool exhaustion signature: a token-bearing\n * `resource_exhausted` error while the conversation is well below the model\n * context window.\n */\nexport function isCursorQuotaResourceExhausted(\n\tmessage: {\n\t\tstopReason?: string;\n\t\terrorMessage?: string;\n\t\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n\t},\n\tcontextWindow: number,\n): boolean {\n\tconst tokens = cursorZeroTokenCount(message);\n\treturn (\n\t\tmessage.stopReason === \"error\" &&\n\t\tRESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage || \"\") &&\n\t\tcontextWindow > 0 &&\n\t\ttokens > 0 &&\n\t\ttokens < contextWindow * 0.5\n\t);\n}\n\nexport function isCursorZeroTokenResourceExhausted(message: {\n\tstopReason?: string;\n\terrorMessage?: string;\n\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n}): boolean {\n\tif (message.stopReason !== \"error\" || !/resource.?exhausted/i.test(message.errorMessage || \"\")) {\n\t\treturn false;\n\t}\n\treturn !(cursorZeroTokenCount(message) > 0);\n}\n\nexport function shouldSkipProviderFallbackForCursorZeroRe(options: { sameModelRemint?: boolean } | undefined): boolean {\n\treturn options?.sameModelRemint === true;\n}\n\nexport function shouldRetryOverflowWithoutCompact(compacted: boolean, errorMessage: string): boolean {\n\tif (compacted) return false;\n\treturn /Nothing to compact/i.test(errorMessage || \"\");\n}\n\nexport function cursorOverflowCompactionSettings<T extends { keepRecentTokens?: number; restorationEnabled?: boolean }>(\n\tsettings: T,\n\tprovider: string | undefined,\n\treason: string | undefined,\n): T {\n\tif (reason !== \"overflow\") return settings;\n\tif (provider !== \"cursor\" && provider !== \"cursor-cli-oauth\") return settings;\n\treturn { ...settings, keepRecentTokens: 0, restorationEnabled: false };\n}\n"]}

package/dist/utils/process-singleton.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Returns the one value this process keeps under `key`, creating it on first use.
+ *
+ * Module-level state that every copy of a module must share lives here instead of in a
+ * module-local binding: the release bundle emits some providers as self-contained chunks
+ * that carry their own copy of the modules they import (#2334), so a module-local `Map`
+ * or `Set` would silently split into one instance per copy.
+ */
+export declare function processSingleton<T extends object>(key: string, create: () => T): T;
+//# sourceMappingURL=process-singleton.d.ts.map

package/dist/utils/process-singleton.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"process-singleton.d.ts","sourceRoot":"","sources":["../../src/utils/process-singleton.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,wBAAgB,gBAAgB,CAAC,CAAC,SAAS,MAAM,EAAE,GAAG,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC,CAOlF"}

package/dist/utils/process-singleton.js

@@ -0,0 +1,18 @@
+/**
+ * Returns the one value this process keeps under `key`, creating it on first use.
+ *
+ * Module-level state that every copy of a module must share lives here instead of in a
+ * module-local binding: the release bundle emits some providers as self-contained chunks
+ * that carry their own copy of the modules they import (#2334), so a module-local `Map`
+ * or `Set` would silently split into one instance per copy.
+ */
+export function processSingleton(key, create) {
+    const slot = Symbol.for(key);
+    const existing = Reflect.get(globalThis, slot);
+    if (existing !== undefined)
+        return existing;
+    const created = create();
+    Reflect.set(globalThis, slot, created);
+    return created;
+}
+//# sourceMappingURL=process-singleton.js.map

package/dist/utils/process-singleton.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"process-singleton.js","sourceRoot":"","sources":["../../src/utils/process-singleton.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,MAAM,UAAU,gBAAgB,CAAmB,GAAW,EAAE,MAAe;IAC9E,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAC7B,MAAM,QAAQ,GAAkB,OAAO,CAAC,GAAG,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC;IAC9D,IAAI,QAAQ,KAAK,SAAS;QAAE,OAAO,QAAQ,CAAC;IAC5C,MAAM,OAAO,GAAG,MAAM,EAAE,CAAC;IACzB,OAAO,CAAC,GAAG,CAAC,UAAU,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;IACvC,OAAO,OAAO,CAAC;AAChB,CAAC","sourcesContent":["/**\n * Returns the one value this process keeps under `key`, creating it on first use.\n *\n * Module-level state that every copy of a module must share lives here instead of in a\n * module-local binding: the release bundle emits some providers as self-contained chunks\n * that carry their own copy of the modules they import (#2334), so a module-local `Map`\n * or `Set` would silently split into one instance per copy.\n */\nexport function processSingleton<T extends object>(key: string, create: () => T): T {\n\tconst slot = Symbol.for(key);\n\tconst existing: T | undefined = Reflect.get(globalThis, slot);\n\tif (existing !== undefined) return existing;\n\tconst created = create();\n\tReflect.set(globalThis, slot, created);\n\treturn created;\n}\n"]}

package/dist/utils/prompt-cache-ttl.d.ts

@@ -24,7 +24,9 @@ export type PromptCacheLifetime = {
 };
 export declare function isAnthropicApiBaseUrl(baseUrl: string): boolean;
 export declare function getAnthropicCompat(model: Model<"anthropic-messages">): Required<Omit<AnthropicMessagesCompat, "forceAdaptiveThinking" | "supportsMidConvoEffort" | "sessionAffinityFormat">> & Pick<AnthropicMessagesCompat, "sessionAffinityFormat">;
-export type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "vllmPriority" | "cacheControlFormat" | "toolCallFormat" | "deferredToolsMode" | "toolSchemaFlavor" | "supportsPromptCacheKey" | "chatTemplateArgs" | "supportsThinkingTokenBudget" | "thinkingTokenBudgetField" | "veniceParameters"> & {
+export type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "vllmPriority" | "cacheControlFormat" | "toolCallFormat" | "deferredToolsMode" | "toolSchemaFlavor" | "supportsPromptCacheKey" | "chatTemplateArgs" | "supportsThinkingTokenBudget" | "thinkingTokenBudgetField" | "veniceParameters" | "supportsForcedToolChoice"> & {
+    /** Declared forced tool_choice support; absent means supported. */
+    supportsForcedToolChoice?: OpenAICompletionsCompat["supportsForcedToolChoice"];
     cacheControlFormat?: OpenAICompletionsCompat["cacheControlFormat"];
     supportsPromptCacheKey?: OpenAICompletionsCompat["supportsPromptCacheKey"];
     toolCallFormat?: OpenAICompletionsCompat["toolCallFormat"];

package/dist/utils/prompt-cache-ttl.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"prompt-cache-ttl.d.ts","sourceRoot":"","sources":["../../src/utils/prompt-cache-ttl.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,uBAAuB,EACvB,GAAG,EAEH,KAAK,EACL,uBAAuB,EACvB,WAAW,EACX,MAAM,aAAa,CAAC;AAGrB,eAAO,MAAM,8BAA8B,MAAM,CAAC;AAClD,eAAO,MAAM,6BAA6B,OAAO,CAAC;AAClD;;;GAGG;AACH,eAAO,MAAM,wCAAwC,OAAO,CAAC;AAE7D;;;;;;;GAOG;AACH,MAAM,MAAM,mBAAmB,GAC5B;IAAE,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC;IAAC,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAA;CAAE,GACrD;IAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAA;CAAE,GAChC;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAiB7B,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAM9D;AAsBD,wBAAgB,kBAAkB,CACjC,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAChC,QAAQ,CACV,IAAI,CAAC,uBAAuB,EAAE,uBAAuB,GAAG,wBAAwB,GAAG,uBAAuB,CAAC,CAC3G,GACA,IAAI,CAAC,uBAAuB,EAAE,uBAAuB,CAAC,CAkCtD;AAED,MAAM,MAAM,+BAA+B,GAAG,IAAI,CACjD,QAAQ,CAAC,uBAAuB,CAAC,EAC/B,cAAc,GACd,oBAAoB,GACpB,gBAAgB,GAChB,mBAAmB,GACnB,kBAAkB,GAClB,wBAAwB,GACxB,kBAAkB,GAClB,6BAA6B,GAC7B,0BAA0B,GAC1B,kBAAkB,CACpB,GAAG;IACH,kBAAkB,CAAC,EAAE,uBAAuB,CAAC,oBAAoB,CAAC,CAAC;IACnE,sBAAsB,CAAC,EAAE,uBAAuB,CAAC,wBAAwB,CAAC,CAAC;IAC3E,cAAc,CAAC,EAAE,uBAAuB,CAAC,gBAAgB,CAAC,CAAC;IAC3D,iBAAiB,CAAC,EAAE,uBAAuB,CAAC,mBAAmB,CAAC,CAAC;IACjE,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;IAC/D,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;IAC/D,2BAA2B,CAAC,EAAE,uBAAuB,CAAC,6BAA6B,CAAC,CAAC;IACrF,wBAAwB,CAAC,EAAE,uBAAuB,CAAC,0BAA0B,CAAC,CAAC;IAC/E,8EAA8E;IAC9E,YAAY,CAAC,EAAE,uBAAuB,CAAC,cAAc,CAAC,CAAC;IACvD,4EAA4E;IAC5E,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;CAC/D,CAAC;AA+GF;;;GAGG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAAG,+BAA+B,CAwC9G;AAED,wBAAgB,8BAA8B,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAM5F;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,GAAG,OAAO,CAKxF;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,OAAO,CAmBzG;AAuED;;;GAGG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,mBAAmB,CAgDpG;AAED;;;GAGG;AACH,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAGrG"}
+{"version":3,"file":"prompt-cache-ttl.d.ts","sourceRoot":"","sources":["../../src/utils/prompt-cache-ttl.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,uBAAuB,EACvB,GAAG,EAEH,KAAK,EACL,uBAAuB,EACvB,WAAW,EACX,MAAM,aAAa,CAAC;AAGrB,eAAO,MAAM,8BAA8B,MAAM,CAAC;AAClD,eAAO,MAAM,6BAA6B,OAAO,CAAC;AAClD;;;GAGG;AACH,eAAO,MAAM,wCAAwC,OAAO,CAAC;AAE7D;;;;;;;GAOG;AACH,MAAM,MAAM,mBAAmB,GAC5B;IAAE,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC;IAAC,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAA;CAAE,GACrD;IAAE,QAAQ,CAAC,IAAI,EAAE,aAAa,CAAA;CAAE,GAChC;IAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAiB7B,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAM9D;AAuBD,wBAAgB,kBAAkB,CACjC,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAChC,QAAQ,CACV,IAAI,CAAC,uBAAuB,EAAE,uBAAuB,GAAG,wBAAwB,GAAG,uBAAuB,CAAC,CAC3G,GACA,IAAI,CAAC,uBAAuB,EAAE,uBAAuB,CAAC,CAkCtD;AAED,MAAM,MAAM,+BAA+B,GAAG,IAAI,CACjD,QAAQ,CAAC,uBAAuB,CAAC,EAC/B,cAAc,GACd,oBAAoB,GACpB,gBAAgB,GAChB,mBAAmB,GACnB,kBAAkB,GAClB,wBAAwB,GACxB,kBAAkB,GAClB,6BAA6B,GAC7B,0BAA0B,GAC1B,kBAAkB,GAClB,0BAA0B,CAC5B,GAAG;IACH,mEAAmE;IACnE,wBAAwB,CAAC,EAAE,uBAAuB,CAAC,0BAA0B,CAAC,CAAC;IAC/E,kBAAkB,CAAC,EAAE,uBAAuB,CAAC,oBAAoB,CAAC,CAAC;IACnE,sBAAsB,CAAC,EAAE,uBAAuB,CAAC,wBAAwB,CAAC,CAAC;IAC3E,cAAc,CAAC,EAAE,uBAAuB,CAAC,gBAAgB,CAAC,CAAC;IAC3D,iBAAiB,CAAC,EAAE,uBAAuB,CAAC,mBAAmB,CAAC,CAAC;IACjE,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;IAC/D,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;IAC/D,2BAA2B,CAAC,EAAE,uBAAuB,CAAC,6BAA6B,CAAC,CAAC;IACrF,wBAAwB,CAAC,EAAE,uBAAuB,CAAC,0BAA0B,CAAC,CAAC;IAC/E,8EAA8E;IAC9E,YAAY,CAAC,EAAE,uBAAuB,CAAC,cAAc,CAAC,CAAC;IACvD,4EAA4E;IAC5E,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;CAC/D,CAAC;AA+GF;;;GAGG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAAG,+BAA+B,CAyC9G;AAED,wBAAgB,8BAA8B,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAM5F;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,GAAG,OAAO,CAKxF;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,OAAO,CAmBzG;AAuED;;;GAGG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,mBAAmB,CAgDpG;AAED;;;GAGG;AACH,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAGrG"}

package/dist/utils/prompt-cache-ttl.js

@@ -29,9 +29,10 @@ export function isAnthropicApiBaseUrl(baseUrl) {
 }
 /**
  * Families that reject forced tool use (`tool_choice` `any` / `tool` return 400): every Fable and
- * Mythos release, and Claude Opus 5.5 (`claude-opus-5-5`; gateways may spell it `claude-opus-5.5`).
+ * Mythos release, Claude Opus 5.5 and Claude Sonnet 5.5 (`claude-opus-5-5`, `claude-sonnet-5-5`;
+ * gateways may spell them with a dot).
  */
-const FORCED_TOOL_CHOICE_REJECTING_MODEL_ID = /^claude-(?:(?:fable|mythos)(?:-|$)|opus-5[.-]5(?:[.-]|$))/i;
+const FORCED_TOOL_CHOICE_REJECTING_MODEL_ID = /^claude-(?:(?:fable|mythos)(?:-|$)|(?:opus|sonnet)-5[.-]5(?:[.-]|$))/i;
 /**
  * Default for `supportsToolReferences`: first-party Anthropic models except
  * Haiku (rejects client-side tool_reference blocks) and models that predate
@@ -210,6 +211,7 @@ export function getOpenAICompletionsCompat(model) {
         sessionAffinityFormat: model.compat.sessionAffinityFormat ?? detected.sessionAffinityFormat,
         supportsPromptCacheKey: model.compat.supportsPromptCacheKey ?? detected.supportsPromptCacheKey,
         supportsMaxOutputTokens: model.compat.supportsMaxOutputTokens ?? detected.supportsMaxOutputTokens,
+        supportsForcedToolChoice: model.compat.supportsForcedToolChoice,
         vllmPriority: model.compat.vllmPriority ?? detected.vllmPriority,
         supportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,
     };

package/dist/utils/prompt-cache-ttl.js.map

@@ -1 +1 @@
-{"version":3,"file":"prompt-cache-ttl.js","sourceRoot":"","sources":["../../src/utils/prompt-cache-ttl.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AAExD,MAAM,CAAC,MAAM,8BAA8B,GAAG,GAAG,CAAC;AAClD,MAAM,CAAC,MAAM,6BAA6B,GAAG,IAAI,CAAC;AAClD;;;GAGG;AACH,MAAM,CAAC,MAAM,wCAAwC,GAAG,IAAI,CAAC;AAe7D,MAAM,eAAe,GAAwB,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC;AAC9D,MAAM,wBAAwB,GAAwB,EAAE,IAAI,EAAE,aAAa,EAAE,CAAC;AAE9E,SAAS,GAAG,CAAC,UAAkB;IAC9B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC;AACpC,CAAC;AAED,SAAS,UAAU,CAAC,OAAe;IAClC,IAAI,CAAC;QACJ,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;IAChD,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AACF,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,OAAe;IACpD,IAAI,CAAC;QACJ,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,QAAQ,KAAK,mBAAmB,CAAC;IAC1D,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AACF,CAAC;AAED;;;GAGG;AACH,MAAM,qCAAqC,GAAG,4DAA4D,CAAC;AAE3G;;;;GAIG;AACH,SAAS,6BAA6B,CAAC,KAAkC;IACxE,IAAI,KAAK,CAAC,QAAQ,KAAK,WAAW,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC;QAAE,OAAO,KAAK,CAAC;IAC/E,MAAM,OAAO,GAAG,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,uDAAuD,CAAC,CAAC;IACxF,IAAI,CAAC,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3B,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;IACjC,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3E,OAAO,KAAK,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC;AACjD,CAAC;AAED,MAAM,UAAU,kBAAkB,CACjC,KAAkC;IAKlC,uEAAuE;IACvE,MAAM,WAAW,GAAG,KAAK,CAAC,QAAQ,KAAK,WAAW,CAAC;IACnD,MAAM,8BAA8B,GACnC,KAAK,CAAC,QAAQ,KAAK,uBAAuB,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC;IACnF,MAAM,QAAQ,GAAG,KAAK,CAAC,QAAQ,KAAK,QAAQ,IAAI,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,oBAAoB,CAAC,CAAC;IAChG,8EAA8E;IAC9E,uDAAuD;IACvD,MAAM,YAAY,GAAG,KAAK,CAAC,QAAQ,KAAK,YAAY,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;IAChG,OAAO;QACN,+BAA+B,EAAE,KAAK,CAAC,MAAM,EAAE,+BAA+B,IAAI,CAAC,WAAW;QAC9F,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,CAAC,WAAW;QACpF,0BAA0B,EACzB,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,CAAC,CAAC,CAAC,WAAW,IAAI,8BAA8B,IAAI,YAAY,CAAC;QAC9G,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC;QACvG,2BAA2B,EAAE,KAAK,CAAC,MAAM,EAAE,2BAA2B,IAAI,CAAC,WAAW;QACtF,wBAAwB,EAAE,KAAK,CAAC,MAAM,EAAE,wBAAwB,IAAI,CAAC,QAAQ;QAC7E,mBAAmB,EAAE,KAAK,CAAC,MAAM,EAAE,mBAAmB,IAAI,IAAI;QAC9D,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,IAAI;QAC5D,wBAAwB,EACvB,KAAK,CAAC,MAAM,EAAE,wBAAwB,IAAI,CAAC,qCAAqC,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAChG,mBAAmB,EAAE,KAAK,CAAC,MAAM,EAAE,mBAAmB,IAAI,KAAK;QAC/D,sBAAsB,EACrB,KAAK,CAAC,MAAM,EAAE,sBAAsB,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,mBAAmB,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,MAAM,CAAC;QACzG,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,EAAE;QAChE,mBAAmB,EAAE,KAAK,CAAC,MAAM,EAAE,mBAAmB,IAAI,KAAK;QAC/D,sBAAsB,EAAE,KAAK,CAAC,MAAM,EAAE,sBAAsB,IAAI,6BAA6B,CAAC,KAAK,CAAC;QACpG,sEAAsE;QACtE,0EAA0E;QAC1E,0EAA0E;QAC1E,yEAAyE;QACzE,WAAW;QACX,iBAAiB,EAAE,KAAK,CAAC,MAAM,EAAE,iBAAiB,IAAI,qBAAqB,CAAC,KAAK,CAAC,OAAO,CAAC;KAC1F,CAAC;AACH,CAAC;AA6BD;;;;GAIG;AACH,SAAS,6BAA6B,CAAC,KAAkC;IACxE,MAAM,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAChC,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;IAE9B,MAAM,KAAK,GACV,QAAQ,KAAK,KAAK;QAClB,QAAQ,KAAK,eAAe;QAC5B,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC;QAC5B,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;IACtC,MAAM,UAAU,GACf,QAAQ,KAAK,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;IACxG,MAAM,UAAU,GAAG,QAAQ,KAAK,YAAY,IAAI,QAAQ,KAAK,eAAe,IAAI,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;IAClH,MAAM,YAAY,GAAG,QAAQ,KAAK,YAAY,IAAI,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;IACpF,MAAM,qBAAqB,GAAG,QAAQ,KAAK,uBAAuB,IAAI,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IAC7G,MAAM,qBAAqB,GAAG,QAAQ,KAAK,uBAAuB,IAAI,OAAO,CAAC,QAAQ,CAAC,2BAA2B,CAAC,CAAC;IACpH,MAAM,QAAQ,GAAG,QAAQ,KAAK,QAAQ,IAAI,OAAO,CAAC,QAAQ,CAAC,0BAA0B,CAAC,CAAC;IACvF,MAAM,SAAS,GAAG,QAAQ,KAAK,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;IAClF,MAAM,UAAU,GAAG,QAAQ,KAAK,UAAU,IAAI,OAAO,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;IAE7F,MAAM,aAAa,GAClB,QAAQ;QACR,QAAQ,KAAK,UAAU;QACvB,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC;QAC/B,QAAQ,KAAK,KAAK;QAClB,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC;QAC5B,UAAU;QACV,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC;QAC7B,UAAU;QACV,KAAK;QACL,UAAU;QACV,QAAQ,KAAK,UAAU;QACvB,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC;QAC/B,qBAAqB;QACrB,qBAAqB;QACrB,SAAS,CAAC;IAEX,MAAM,YAAY,GACjB,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC;QAC7B,UAAU;QACV,UAAU;QACV,qBAAqB;QACrB,UAAU;QACV,QAAQ;QACR,SAAS;QACT,KAAK,CAAC;IAEP,MAAM,MAAM,GAAG,QAAQ,KAAK,KAAK,IAAI,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IAClE,MAAM,8BAA8B,GACnC,YAAY,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,YAAY,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC;IACvF,MAAM,8BAA8B,GAAG,CAAC,YAAY,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IAC1E,MAAM,mBAAmB,GAAG,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC;IACpF,MAAM,8BAA8B,GAAG,8BAA8B,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CACrF,mBAAmB,CAAC,UAAU,CAAC,MAAM,CAAC,CACtC,CAAC;IACF,MAAM,kBAAkB,GAAG,QAAQ,KAAK,YAAY,IAAI,8BAA8B,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IAEjH,OAAO;QACN,aAAa,EAAE,CAAC,aAAa;QAC7B,qBAAqB,EAAE,8BAA8B,IAAI,CAAC,CAAC,aAAa,IAAI,CAAC,YAAY,CAAC;QAC1F,uBAAuB,EACtB,CAAC,MAAM,IAAI,CAAC,KAAK,IAAI,CAAC,UAAU,IAAI,CAAC,UAAU,IAAI,CAAC,qBAAqB,IAAI,CAAC,QAAQ,IAAI,CAAC,SAAS;QACrG,wBAAwB,EAAE,IAAI;QAC9B,oBAAoB,EAAE,IAAI;QAC1B,cAAc,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,uBAAuB;QACrE,sBAAsB,EAAE,KAAK;QAC7B,gCAAgC,EAAE,KAAK;QACvC,sBAAsB,EAAE,KAAK;QAC7B,2CAA2C,EAAE,UAAU;QACvD,cAAc,EAAE,UAAU;YACzB,CAAC,CAAC,UAAU;YACZ,CAAC,CAAC,KAAK;gBACN,CAAC,CAAC,KAAK;gBACP,CAAC,CAAC,UAAU;oBACX,CAAC,CAAC,UAAU;oBACZ,CAAC,CAAC,SAAS;wBACV,CAAC,CAAC,UAAU;wBACZ,CAAC,CAAC,YAAY;4BACb,CAAC,CAAC,YAAY;4BACd,CAAC,CAAC,QAAQ;QACf,iBAAiB,EAAE,EAAE;QACrB,oBAAoB,EAAE,EAAE;QACxB,kBAAkB,EAAE,EAAE;QACtB,aAAa,EAAE,KAAK;QACpB,kBAAkB,EAAE,CAAC,UAAU,IAAI,CAAC,UAAU,IAAI,CAAC,qBAAqB,IAAI,CAAC,QAAQ;QACrF,gBAAgB,EAAE,UAAU,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,SAAS;QAC1D,wBAAwB,EAAE,IAAI;QAC9B,cAAc,EAAE,SAAS;QACzB,0BAA0B,EAAE,KAAK;QACjC,kBAAkB;QAClB,0BAA0B,EAAE,YAAY;QACxC,iBAAiB,EAAE,SAAS;QAC5B,qBAAqB,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ;QAC7D,sBAAsB,EAAE,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC;QACxE,uBAAuB,EAAE,IAAI;QAC7B,0BAA0B,EAAE,CAAC,CAC5B,UAAU;YACV,qBAAqB;YACrB,qBAAqB;YACrB,QAAQ;YACR,SAAS,CACT;KACD,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,0BAA0B,CAAC,KAAkC;IAC5E,MAAM,QAAQ,GAAG,6BAA6B,CAAC,KAAK,CAAC,CAAC;IACtD,IAAI,CAAC,KAAK,CAAC,MAAM;QAAE,OAAO,QAAQ,CAAC;IAEnC,OAAO;QACN,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,qBAAqB,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,qBAAqB;QAC3F,uBAAuB,EAAE,KAAK,CAAC,MAAM,CAAC,uBAAuB,IAAI,QAAQ,CAAC,uBAAuB;QACjG,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,oBAAoB,EAAE,KAAK,CAAC,MAAM,CAAC,oBAAoB,IAAI,QAAQ,CAAC,oBAAoB;QACxF,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,gCAAgC,EAC/B,KAAK,CAAC,MAAM,CAAC,gCAAgC,IAAI,QAAQ,CAAC,gCAAgC;QAC3F,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,2CAA2C,EAC1C,KAAK,CAAC,MAAM,CAAC,2CAA2C;YACxD,QAAQ,CAAC,2CAA2C;QACrD,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,iBAAiB,IAAI,QAAQ,CAAC,iBAAiB;QAC/E,oBAAoB,EAAE,KAAK,CAAC,MAAM,CAAC,oBAAoB,IAAI,QAAQ,CAAC,oBAAoB;QACxF,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAC5E,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,2BAA2B,EAAE,KAAK,CAAC,MAAM,CAAC,2BAA2B,IAAI,QAAQ,CAAC,2BAA2B;QAC7G,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAC5E,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;QAC1G,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;QAC1G,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,iBAAiB,IAAI,QAAQ,CAAC,iBAAiB;QAC/E,qBAAqB,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,qBAAqB;QAC3F,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,uBAAuB,EAAE,KAAK,CAAC,MAAM,CAAC,uBAAuB,IAAI,QAAQ,CAAC,uBAAuB;QACjG,YAAY,EAAE,KAAK,CAAC,MAAM,CAAC,YAAY,IAAI,QAAQ,CAAC,YAAY;QAChE,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;KAC1G,CAAC;AACH,CAAC;AAED,MAAM,UAAU,8BAA8B,CAAC,OAAe,EAAE,SAAkB;IACjF,MAAM,MAAM,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;IAC5D,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE;QAC/B,MAAM,KAAK,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;QAClC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,CAAC;IACjD,CAAC,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAuC;IAC9E,MAAM,UAAU,GAAG,8BAA8B,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IACxE,OAAO,UAAU,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CACpC,CAAC,UAAU,EAAE,YAAY,EAAE,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,YAAY,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,YAAY,CAAC,CAAC,CAChG,CAAC;AACH,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,qBAAqB,CAAC,KAAuC,EAAE,GAAiB;IAC/F,MAAM,UAAU,GAAG,8BAA8B,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IAExE,MAAM,YAAY,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClE,IAAI,CAAC,YAAY,EAAE,CAAC;QACnB,0EAA0E;QAC1E,8DAA8D;QAC9D,IAAI,mBAAmB,CAAC,yBAAyB,EAAE,GAAG,CAAC,KAAK,GAAG;YAAE,OAAO,IAAI,CAAC;QAC7E,OAAO,KAAK,CAAC;IACd,CAAC;IACD,8CAA8C;IAC9C,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACjH,gDAAgD;IAChD,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IAC3D,oBAAoB;IACpB,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,mBAAmB,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACzE,mBAAmB;IACnB,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACxE,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,8BAA8B,CACtC,cAA+B,EAC/B,GAAiB,EACjB,QAAQ,GAAmB,OAAO;IAElC,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,OAAO,CAAC,GAAG,CAAC,kBAAkB,KAAK,SAAS,EAAE,CAAC;QACpF,OAAO,OAAO,CAAC;IAChB,CAAC;IACD,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED,SAAS,sCAAsC,CAAC,cAA+B,EAAE,GAAiB;IACjG,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,4BAA4B,CAAC,cAA+B,EAAE,GAAiB;IACvF,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,oCAAoC,CAAC,cAA+B,EAAE,GAAiB;IAC/F,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,mGAAmG;AACnG,SAAS,qBAAqB,CAAC,KAAkC;IAChE,OAAO,KAAK,CAAC,QAAQ,KAAK,UAAU,IAAI,UAAU,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,kBAAkB,CAAC;AAC1F,CAAC;AAED,4EAA4E;AAC5E,MAAM,8BAA8B,GAAG,iEAAiE,CAAC;AAEzG;;;;GAIG;AACH,SAAS,4BAA4B,CAAC,KAAiB;IACtD,IAAI,KAAK,CAAC,GAAG,KAAK,kBAAkB,EAAE,CAAC;QACtC,MAAM,cAAc,GAAG,KAAkC,CAAC;QAC1D,IAAI,cAAc,CAAC,MAAM,EAAE,+BAA+B,KAAK,IAAI;YAAE,OAAO,IAAI,CAAC;QACjF,IAAI,cAAc,CAAC,QAAQ,KAAK,QAAQ,IAAI,UAAU,CAAC,cAAc,CAAC,OAAO,CAAC,KAAK,gBAAgB;YAAE,OAAO,KAAK,CAAC;IACnH,CAAC;IACD,OAAO,8BAA8B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;AACtD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,0BAA0B,CAAC,KAAiB,EAAE,GAAiB;IAC9E,QAAQ,KAAK,CAAC,GAAG,EAAE,CAAC;QACnB,KAAK,kBAAkB;YACtB,wFAAwF;YACxF,OAAO,GAAG,CAAC,8BAA8B,CAAC,CAAC;QAC5C,KAAK,oBAAoB,EAAE,CAAC;YAC3B,MAAM,cAAc,GAAG,KAAoC,CAAC;YAC5D,MAAM,SAAS,GAAG,8BAA8B,CAAC,cAAc,CAAC,cAAc,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;YAC9F,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,eAAe,CAAC;YACjD,OAAO,SAAS,KAAK,MAAM;gBAC1B,qBAAqB,CAAC,cAAc,CAAC,OAAO,CAAC;gBAC7C,kBAAkB,CAAC,cAAc,CAAC,CAAC,0BAA0B;gBAC7D,CAAC,CAAC,GAAG,CAAC,6BAA6B,CAAC;gBACpC,CAAC,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;QACxC,CAAC;QACD,KAAK,yBAAyB,EAAE,CAAC;YAChC,MAAM,YAAY,GAAG,KAAyC,CAAC;YAC/D,MAAM,SAAS,GAAG,4BAA4B,CAAC,YAAY,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;YACjF,IAAI,SAAS,KAAK,MAAM,IAAI,CAAC,qBAAqB,CAAC,YAAY,EAAE,GAAG,CAAC;gBAAE,OAAO,eAAe,CAAC;YAC9F,OAAO,SAAS,KAAK,MAAM,IAAI,uBAAuB,CAAC,YAAY,CAAC;gBACnE,CAAC,CAAC,GAAG,CAAC,6BAA6B,CAAC;gBACpC,CAAC,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;QACxC,CAAC;QACD,KAAK,oBAAoB,EAAE,CAAC;YAC3B,MAAM,gBAAgB,GAAG,KAAoC,CAAC;YAC9D,MAAM,SAAS,GAAG,sCAAsC,CAAC,gBAAgB,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;YAC/F,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,eAAe,CAAC;YACjD,IAAI,qBAAqB,CAAC,gBAAgB,CAAC;gBAAE,OAAO,wBAAwB,CAAC;YAC7E,MAAM,MAAM,GAAG,0BAA0B,CAAC,gBAAgB,CAAC,CAAC;YAC5D,IAAI,MAAM,CAAC,kBAAkB,KAAK,WAAW,EAAE,CAAC;gBAC/C,OAAO,SAAS,KAAK,MAAM,IAAI,MAAM,CAAC,0BAA0B;oBAC/D,CAAC,CAAC,GAAG,CAAC,6BAA6B,CAAC;oBACpC,CAAC,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;YACxC,CAAC;YACD,OAAO,GAAG,CAAC,8BAA8B,CAAC,CAAC;QAC5C,CAAC;QACD,KAAK,kBAAkB,CAAC;QACxB,KAAK,wBAAwB,CAAC;QAC9B,KAAK,wBAAwB,EAAE,CAAC;YAC/B,MAAM,SAAS,GAAG,oCAAoC,CAAC,KAAK,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;YAClF,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,eAAe,CAAC;YACjD,OAAO,4BAA4B,CAAC,KAAK,CAAC;gBACzC,CAAC,CAAC,GAAG,CAAC,wCAAwC,CAAC;gBAC/C,CAAC,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;QACxC,CAAC;QACD;YACC,OAAO,eAAe,CAAC;IACzB,CAAC;AACF,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,4BAA4B,CAAC,KAAiB,EAAE,GAAiB;IAChF,MAAM,QAAQ,GAAG,0BAA0B,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IACxD,OAAO,QAAQ,CAAC,IAAI,KAAK,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;AAClE,CAAC","sourcesContent":["import type {\n\tAnthropicMessagesCompat,\n\tApi,\n\tCacheRetention,\n\tModel,\n\tOpenAICompletionsCompat,\n\tProviderEnv,\n} from \"../types.ts\";\nimport { getProviderEnvValue } from \"./provider-env.ts\";\n\nexport const PROMPT_CACHE_TTL_SHORT_SECONDS = 300;\nexport const PROMPT_CACHE_TTL_LONG_SECONDS = 3600;\n/**\n * OpenAI GPT-5.6 and later: `prompt_cache_options.ttl` accepts only `\"30m\"`, and the cache stays\n * eligible at least 30 minutes after the latest write or reuse.\n */\nexport const PROMPT_CACHE_TTL_OPENAI_EXTENDED_SECONDS = 1800;\n\n/**\n * How long a provider promises to keep a prompt prefix cached.\n *\n * - `ttl`: an explicit expiry contract; `ttlSeconds` is the conservative lifetime after the last write or reuse.\n * - `best-effort`: the provider caches automatically but promises no expiry (direct DeepSeek clears unused entries\n *   after hours to days), so nothing needs to wake or ping solely to preserve the cache.\n * - `none`: caching is disabled or the lane has no known cache contract.\n */\nexport type PromptCacheLifetime =\n\t| { readonly kind: \"ttl\"; readonly ttlSeconds: number }\n\t| { readonly kind: \"best-effort\" }\n\t| { readonly kind: \"none\" };\n\nconst NO_PROMPT_CACHE: PromptCacheLifetime = { kind: \"none\" };\nconst BEST_EFFORT_PROMPT_CACHE: PromptCacheLifetime = { kind: \"best-effort\" };\n\nfunction ttl(ttlSeconds: number): PromptCacheLifetime {\n\treturn { kind: \"ttl\", ttlSeconds };\n}\n\nfunction hostnameOf(baseUrl: string): string | undefined {\n\ttry {\n\t\treturn new URL(baseUrl).hostname.toLowerCase();\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport function isAnthropicApiBaseUrl(baseUrl: string): boolean {\n\ttry {\n\t\treturn new URL(baseUrl).hostname === \"api.anthropic.com\";\n\t} catch {\n\t\treturn false;\n\t}\n}\n\n/**\n * Families that reject forced tool use (`tool_choice` `any` / `tool` return 400): every Fable and\n * Mythos release, and Claude Opus 5.5 (`claude-opus-5-5`; gateways may spell it `claude-opus-5.5`).\n */\nconst FORCED_TOOL_CHOICE_REJECTING_MODEL_ID = /^claude-(?:(?:fable|mythos)(?:-|$)|opus-5[.-]5(?:[.-]|$))/i;\n\n/**\n * Default for `supportsToolReferences`: first-party Anthropic models except\n * Haiku (rejects client-side tool_reference blocks) and models that predate\n * tool search (Claude 3.x, Opus/Sonnet 4.0, Opus 4.1).\n */\nfunction defaultSupportsToolReferences(model: Model<\"anthropic-messages\">): boolean {\n\tif (model.provider !== \"anthropic\" || model.id.includes(\"haiku\")) return false;\n\tconst version = model.id.match(/^claude-(?:opus|sonnet|fable)-(\\d+)(?:-(\\d+))?(?:-|$)/);\n\tif (!version) return false;\n\tconst major = Number(version[1]);\n\tconst minor = version[2] && version[2].length < 8 ? Number(version[2]) : 0;\n\treturn major > 4 || (major === 4 && minor >= 5);\n}\n\nexport function getAnthropicCompat(\n\tmodel: Model<\"anthropic-messages\">,\n): Required<\n\tOmit<AnthropicMessagesCompat, \"forceAdaptiveThinking\" | \"supportsMidConvoEffort\" | \"sessionAffinityFormat\">\n> &\n\tPick<AnthropicMessagesCompat, \"sessionAffinityFormat\"> {\n\t// Auto-detect session affinity and cache control support from provider\n\tconst isFireworks = model.provider === \"fireworks\";\n\tconst isCloudflareAiGatewayAnthropic =\n\t\tmodel.provider === \"cloudflare-ai-gateway\" && model.baseUrl.includes(\"anthropic\");\n\tconst isXiaomi = model.provider === \"xiaomi\" || model.provider.startsWith(\"xiaomi-token-plan-\");\n\t// OpenRouter carries prompt-cache affinity on its own x-session-id header and\n\t// rejects x-session-affinity (earendil-works/pi#9102).\n\tconst isOpenRouter = model.provider === \"openrouter\" || model.baseUrl.includes(\"openrouter.ai\");\n\treturn {\n\t\tsupportsEagerToolInputStreaming: model.compat?.supportsEagerToolInputStreaming ?? !isFireworks,\n\t\tsupportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? !isFireworks,\n\t\tsendSessionAffinityHeaders:\n\t\t\tmodel.compat?.sendSessionAffinityHeaders ?? !!(isFireworks || isCloudflareAiGatewayAnthropic || isOpenRouter),\n\t\tsessionAffinityFormat: model.compat?.sessionAffinityFormat ?? (isOpenRouter ? \"openrouter\" : undefined),\n\t\tsupportsCacheControlOnTools: model.compat?.supportsCacheControlOnTools ?? !isFireworks,\n\t\tsupportsDisabledThinking: model.compat?.supportsDisabledThinking ?? !isXiaomi,\n\t\tsupportsTemperature: model.compat?.supportsTemperature ?? true,\n\t\tsupportsToolChoice: model.compat?.supportsToolChoice ?? true,\n\t\tsupportsForcedToolChoice:\n\t\t\tmodel.compat?.supportsForcedToolChoice ?? !FORCED_TOOL_CHOICE_REJECTING_MODEL_ID.test(model.id),\n\t\tallowEmptySignature: model.compat?.allowEmptySignature ?? false,\n\t\tunsignedThinkingReplay:\n\t\t\tmodel.compat?.unsignedThinkingReplay ?? (model.compat?.allowEmptySignature ? \"empty-signature\" : \"text\"),\n\t\tallowedFallbackModels: model.compat?.allowedFallbackModels ?? [],\n\t\tsupportsStrictTools: model.compat?.supportsStrictTools ?? false,\n\t\tsupportsToolReferences: model.compat?.supportsToolReferences ?? defaultSupportsToolReferences(model),\n\t\t// Default: first-party Anthropic only. Anthropic-compatible providers\n\t\t// (kimi-coding, fireworks, copilot, gateways) may execute the server-side\n\t\t// search but reject the replayed server_tool_use / web_search_tool_result\n\t\t// blocks on the next request (kimi-coding 400s with `tool_call_id is not\n\t\t// found`).\n\t\tsupportsWebSearch: model.compat?.supportsWebSearch ?? isAnthropicApiBaseUrl(model.baseUrl),\n\t};\n}\n\nexport type ResolvedOpenAICompletionsCompat = Omit<\n\tRequired<OpenAICompletionsCompat>,\n\t| \"vllmPriority\"\n\t| \"cacheControlFormat\"\n\t| \"toolCallFormat\"\n\t| \"deferredToolsMode\"\n\t| \"toolSchemaFlavor\"\n\t| \"supportsPromptCacheKey\"\n\t| \"chatTemplateArgs\"\n\t| \"supportsThinkingTokenBudget\"\n\t| \"thinkingTokenBudgetField\"\n\t| \"veniceParameters\"\n> & {\n\tcacheControlFormat?: OpenAICompletionsCompat[\"cacheControlFormat\"];\n\tsupportsPromptCacheKey?: OpenAICompletionsCompat[\"supportsPromptCacheKey\"];\n\ttoolCallFormat?: OpenAICompletionsCompat[\"toolCallFormat\"];\n\tdeferredToolsMode?: OpenAICompletionsCompat[\"deferredToolsMode\"];\n\ttoolSchemaFlavor?: OpenAICompletionsCompat[\"toolSchemaFlavor\"];\n\tchatTemplateArgs?: OpenAICompletionsCompat[\"chatTemplateArgs\"];\n\tsupportsThinkingTokenBudget?: OpenAICompletionsCompat[\"supportsThinkingTokenBudget\"];\n\tthinkingTokenBudgetField?: OpenAICompletionsCompat[\"thinkingTokenBudgetField\"];\n\t/** vLLM `priority`; off by default and never set on the generated catalog. */\n\tvllmPriority?: OpenAICompletionsCompat[\"vllmPriority\"];\n\t/** Venice `venice_parameters`; only set on the generated Venice catalog. */\n\tveniceParameters?: OpenAICompletionsCompat[\"veniceParameters\"];\n};\n\n/**\n * Detect compatibility settings from provider and baseUrl for known providers.\n * Provider takes precedence over URL-based detection since it's explicitly configured.\n * Returns a fully resolved OpenAICompletionsCompat object with all fields set.\n */\nfunction detectOpenAICompletionsCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst provider = model.provider;\n\tconst baseUrl = model.baseUrl;\n\n\tconst isZai =\n\t\tprovider === \"zai\" ||\n\t\tprovider === \"zai-coding-cn\" ||\n\t\tbaseUrl.includes(\"api.z.ai\") ||\n\t\tbaseUrl.includes(\"open.bigmodel.cn\");\n\tconst isTogether =\n\t\tprovider === \"together\" || baseUrl.includes(\"api.together.ai\") || baseUrl.includes(\"api.together.xyz\");\n\tconst isMoonshot = provider === \"moonshotai\" || provider === \"moonshotai-cn\" || baseUrl.includes(\"api.moonshot.\");\n\tconst isOpenRouter = provider === \"openrouter\" || baseUrl.includes(\"openrouter.ai\");\n\tconst isCloudflareWorkersAI = provider === \"cloudflare-workers-ai\" || baseUrl.includes(\"api.cloudflare.com\");\n\tconst isCloudflareAiGateway = provider === \"cloudflare-ai-gateway\" || baseUrl.includes(\"gateway.ai.cloudflare.com\");\n\tconst isNvidia = provider === \"nvidia\" || baseUrl.includes(\"integrate.api.nvidia.com\");\n\tconst isAntLing = provider === \"ant-ling\" || baseUrl.includes(\"api.ant-ling.com\");\n\tconst isDeepSeek = provider === \"deepseek\" || baseUrl.toLowerCase().includes(\"deepseek.com\");\n\n\tconst isNonStandard =\n\t\tisNvidia ||\n\t\tprovider === \"cerebras\" ||\n\t\tbaseUrl.includes(\"cerebras.ai\") ||\n\t\tprovider === \"xai\" ||\n\t\tbaseUrl.includes(\"api.x.ai\") ||\n\t\tisTogether ||\n\t\tbaseUrl.includes(\"chutes.ai\") ||\n\t\tisDeepSeek ||\n\t\tisZai ||\n\t\tisMoonshot ||\n\t\tprovider === \"opencode\" ||\n\t\tbaseUrl.includes(\"opencode.ai\") ||\n\t\tisCloudflareWorkersAI ||\n\t\tisCloudflareAiGateway ||\n\t\tisAntLing;\n\n\tconst useMaxTokens =\n\t\tbaseUrl.includes(\"chutes.ai\") ||\n\t\tisDeepSeek ||\n\t\tisMoonshot ||\n\t\tisCloudflareAiGateway ||\n\t\tisTogether ||\n\t\tisNvidia ||\n\t\tisAntLing ||\n\t\tisZai;\n\n\tconst isGrok = provider === \"xai\" || baseUrl.includes(\"api.x.ai\");\n\tconst isOpenRouterDeveloperRoleModel =\n\t\tisOpenRouter && (model.id.startsWith(\"anthropic/\") || model.id.startsWith(\"openai/\"));\n\tconst openRouterCacheControlPrefixes = [\"anthropic/\", \"qwen/\", \"google/\"];\n\tconst cacheControlModelId = model.id.startsWith(\"~\") ? model.id.slice(1) : model.id;\n\tconst supportsOpenRouterCacheControl = openRouterCacheControlPrefixes.some((prefix) =>\n\t\tcacheControlModelId.startsWith(prefix),\n\t);\n\tconst cacheControlFormat = provider === \"openrouter\" && supportsOpenRouterCacheControl ? \"anthropic\" : undefined;\n\n\treturn {\n\t\tsupportsStore: !isNonStandard,\n\t\tsupportsDeveloperRole: isOpenRouterDeveloperRoleModel || (!isNonStandard && !isOpenRouter),\n\t\tsupportsReasoningEffort:\n\t\t\t!isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia && !isAntLing,\n\t\tsupportsUsageInStreaming: true,\n\t\tsupportsFinishReason: true,\n\t\tmaxTokensField: useMaxTokens ? \"max_tokens\" : \"max_completion_tokens\",\n\t\trequiresToolResultName: false,\n\t\trequiresAssistantAfterToolResult: false,\n\t\trequiresThinkingAsText: false,\n\t\trequiresReasoningContentOnAssistantMessages: isDeepSeek,\n\t\tthinkingFormat: isDeepSeek\n\t\t\t? \"deepseek\"\n\t\t\t: isZai\n\t\t\t\t? \"zai\"\n\t\t\t\t: isTogether\n\t\t\t\t\t? \"together\"\n\t\t\t\t\t: isAntLing\n\t\t\t\t\t\t? \"ant-ling\"\n\t\t\t\t\t\t: isOpenRouter\n\t\t\t\t\t\t\t? \"openrouter\"\n\t\t\t\t\t\t\t: \"openai\",\n\t\topenRouterRouting: {},\n\t\tvercelGatewayRouting: {},\n\t\tchatTemplateKwargs: {},\n\t\tzaiToolStream: false,\n\t\tsupportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,\n\t\ttoolSchemaFlavor: isMoonshot ? \"moonshot-mfjs\" : undefined,\n\t\tsupportsDisabledThinking: true,\n\t\ttoolCallFormat: undefined,\n\t\tsupportsOpenAIGrammarTools: false,\n\t\tcacheControlFormat,\n\t\tsendSessionAffinityHeaders: isOpenRouter,\n\t\tdeferredToolsMode: undefined,\n\t\tsessionAffinityFormat: isOpenRouter ? \"openrouter\" : \"openai\",\n\t\tsupportsPromptCacheKey: isMoonshot || baseUrl.includes(\"api.openai.com\"),\n\t\tsupportsMaxOutputTokens: true,\n\t\tsupportsLongCacheRetention: !(\n\t\t\tisTogether ||\n\t\t\tisCloudflareWorkersAI ||\n\t\t\tisCloudflareAiGateway ||\n\t\t\tisNvidia ||\n\t\t\tisAntLing\n\t\t),\n\t};\n}\n\n/**\n * Get resolved compatibility settings for a model.\n * Auto-detects from provider/URL then overrides with explicit model.compat.\n */\nexport function getOpenAICompletionsCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst detected = detectOpenAICompletionsCompat(model);\n\tif (!model.compat) return detected;\n\n\treturn {\n\t\tsupportsStore: model.compat.supportsStore ?? detected.supportsStore,\n\t\tsupportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,\n\t\tsupportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,\n\t\tsupportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? detected.supportsUsageInStreaming,\n\t\tsupportsFinishReason: model.compat.supportsFinishReason ?? detected.supportsFinishReason,\n\t\tmaxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,\n\t\trequiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,\n\t\trequiresAssistantAfterToolResult:\n\t\t\tmodel.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,\n\t\trequiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,\n\t\trequiresReasoningContentOnAssistantMessages:\n\t\t\tmodel.compat.requiresReasoningContentOnAssistantMessages ??\n\t\t\tdetected.requiresReasoningContentOnAssistantMessages,\n\t\tthinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,\n\t\tsupportsDisabledThinking: model.compat.supportsDisabledThinking ?? detected.supportsDisabledThinking,\n\t\topenRouterRouting: model.compat.openRouterRouting ?? detected.openRouterRouting,\n\t\tvercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,\n\t\tchatTemplateKwargs: model.compat.chatTemplateKwargs ?? detected.chatTemplateKwargs,\n\t\tchatTemplateArgs: model.compat.chatTemplateArgs ?? detected.chatTemplateArgs,\n\t\tzaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,\n\t\tsupportsThinkingTokenBudget: model.compat.supportsThinkingTokenBudget ?? detected.supportsThinkingTokenBudget,\n\t\tthinkingTokenBudgetField: model.compat.thinkingTokenBudgetField ?? detected.thinkingTokenBudgetField,\n\t\tsupportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,\n\t\ttoolSchemaFlavor: model.compat.toolSchemaFlavor ?? detected.toolSchemaFlavor,\n\t\ttoolCallFormat: model.compat.toolCallFormat ?? detected.toolCallFormat,\n\t\tsupportsOpenAIGrammarTools: model.compat.supportsOpenAIGrammarTools ?? detected.supportsOpenAIGrammarTools,\n\t\tcacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,\n\t\tsendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,\n\t\tdeferredToolsMode: model.compat.deferredToolsMode ?? detected.deferredToolsMode,\n\t\tsessionAffinityFormat: model.compat.sessionAffinityFormat ?? detected.sessionAffinityFormat,\n\t\tsupportsPromptCacheKey: model.compat.supportsPromptCacheKey ?? detected.supportsPromptCacheKey,\n\t\tsupportsMaxOutputTokens: model.compat.supportsMaxOutputTokens ?? detected.supportsMaxOutputTokens,\n\t\tvllmPriority: model.compat.vllmPriority ?? detected.vllmPriority,\n\t\tsupportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,\n\t};\n}\n\nexport function getBedrockModelMatchCandidates(modelId: string, modelName?: string): string[] {\n\tconst values = modelName ? [modelId, modelName] : [modelId];\n\treturn values.flatMap((value) => {\n\t\tconst lower = value.toLowerCase();\n\t\treturn [lower, lower.replace(/[\\s_.:]+/g, \"-\")];\n\t});\n}\n\nexport function supportsOneHourCacheTtl(model: Model<\"bedrock-converse-stream\">): boolean {\n\tconst candidates = getBedrockModelMatchCandidates(model.id, model.name);\n\treturn candidates.some((candidate) =>\n\t\t[\"opus-4-5\", \"sonnet-4-5\", \"haiku-4-5\"].some((modelVersion) => candidate.includes(modelVersion)),\n\t);\n}\n\n/**\n * Check if the model supports prompt caching.\n * Supported: Claude 3.5 Haiku, Claude 3.7 Sonnet, Claude 4.x models, Claude 5 models\n *\n * For base models and system-defined inference profiles the model ID / ARN\n * contains the model name, so we can decide locally.\n *\n * For application inference profiles (whose ARNs don't contain the model name),\n * also checks model.name which is user-controlled via models.json or registerProvider.\n * As a last resort, set AWS_BEDROCK_FORCE_CACHE=1 to enable cache points.\n * Amazon Nova models have automatic caching and don't need explicit cache points.\n */\nexport function supportsPromptCaching(model: Model<\"bedrock-converse-stream\">, env?: ProviderEnv): boolean {\n\tconst candidates = getBedrockModelMatchCandidates(model.id, model.name);\n\n\tconst hasClaudeRef = candidates.some((s) => s.includes(\"claude\"));\n\tif (!hasClaudeRef) {\n\t\t// Application inference profiles don't contain the model name in the ARN.\n\t\t// Allow users to force cache points via environment variable.\n\t\tif (getProviderEnvValue(\"AWS_BEDROCK_FORCE_CACHE\", env) === \"1\") return true;\n\t\treturn false;\n\t}\n\t// Claude 5 models (fable-5, opus-5, sonnet-5)\n\tif (candidates.some((s) => s.includes(\"fable-5\") || s.includes(\"opus-5\") || s.includes(\"sonnet-5\"))) return true;\n\t// Claude 4.x models (opus-4, sonnet-4, haiku-4)\n\tif (candidates.some((s) => s.includes(\"-4-\"))) return true;\n\t// Claude 3.7 Sonnet\n\tif (candidates.some((s) => s.includes(\"claude-3-7-sonnet\"))) return true;\n\t// Claude 3.5 Haiku\n\tif (candidates.some((s) => s.includes(\"claude-3-5-haiku\"))) return true;\n\treturn false;\n}\n\nfunction resolveAnthropicCacheRetention(\n\tcacheRetention?: CacheRetention,\n\tenv?: ProviderEnv,\n\tfallback: CacheRetention = \"short\",\n): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\tif (typeof process !== \"undefined\" && process.env.PI_CACHE_RETENTION !== undefined) {\n\t\treturn \"short\";\n\t}\n\treturn fallback;\n}\n\nfunction resolveOpenAICompletionsCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction resolveBedrockCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction resolveOpenAIResponsesCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\n/** Direct DeepSeek API: its automatic disk cache is best-effort and carries no expiry contract. */\nfunction isDirectDeepSeekModel(model: Model<\"openai-completions\">): boolean {\n\treturn model.provider === \"deepseek\" || hostnameOf(model.baseUrl) === \"api.deepseek.com\";\n}\n\n/** GPT-5.6 and later (`gpt-5.6-*`, `gpt-5.10`, `gpt-6-*`, `gpt-7`, ...). */\nconst OPENAI_EXTENDED_CACHE_MODEL_ID = /^gpt-(?:5\\.(?:[6-9]|\\d{2,})|(?:[6-9]|\\d{2,})(?:\\.\\d+)?)(?:-|$)/i;\n\n/**\n * The OpenAI-operated Responses lanes (OpenAI API, Azure OpenAI, ChatGPT subscription) serving a GPT-5.6+\n * model, where OpenAI documents a cache lifetime of at least 30 minutes. Gateways that merely proxy the same\n * model ids keep the conservative short lifetime because their routing does not carry that contract.\n */\nfunction hasOpenAIExtendedPromptCache(model: Model<Api>): boolean {\n\tif (model.api === \"openai-responses\") {\n\t\tconst responsesModel = model as Model<\"openai-responses\">;\n\t\tif (responsesModel.compat?.supportsExplicitPromptCacheMode === true) return true;\n\t\tif (responsesModel.provider !== \"openai\" && hostnameOf(responsesModel.baseUrl) !== \"api.openai.com\") return false;\n\t}\n\treturn OPENAI_EXTENDED_CACHE_MODEL_ID.test(model.id);\n}\n\n/**\n * Classify the active model's prompt-cache lifetime from the provider's documented cache contract.\n * `cacheRetention: \"none\"` (or `PI_CACHE_RETENTION` resolving to it) always wins.\n */\nexport function resolvePromptCacheLifetime(model: Model<Api>, env?: ProviderEnv): PromptCacheLifetime {\n\tswitch (model.api) {\n\t\tcase \"claude-sdk-oauth\":\n\t\t\t// The Claude SDK owns prompt caching for this lane and uses Anthropic's default 5m TTL.\n\t\t\treturn ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\tcase \"anthropic-messages\": {\n\t\t\tconst anthropicModel = model as Model<\"anthropic-messages\">;\n\t\t\tconst retention = resolveAnthropicCacheRetention(anthropicModel.cacheRetention, env, \"short\");\n\t\t\tif (retention === \"none\") return NO_PROMPT_CACHE;\n\t\t\treturn retention === \"long\" &&\n\t\t\t\tisAnthropicApiBaseUrl(anthropicModel.baseUrl) &&\n\t\t\t\tgetAnthropicCompat(anthropicModel).supportsLongCacheRetention\n\t\t\t\t? ttl(PROMPT_CACHE_TTL_LONG_SECONDS)\n\t\t\t\t: ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t}\n\t\tcase \"bedrock-converse-stream\": {\n\t\t\tconst bedrockModel = model as Model<\"bedrock-converse-stream\">;\n\t\t\tconst retention = resolveBedrockCacheRetention(bedrockModel.cacheRetention, env);\n\t\t\tif (retention === \"none\" || !supportsPromptCaching(bedrockModel, env)) return NO_PROMPT_CACHE;\n\t\t\treturn retention === \"long\" && supportsOneHourCacheTtl(bedrockModel)\n\t\t\t\t? ttl(PROMPT_CACHE_TTL_LONG_SECONDS)\n\t\t\t\t: ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t}\n\t\tcase \"openai-completions\": {\n\t\t\tconst completionsModel = model as Model<\"openai-completions\">;\n\t\t\tconst retention = resolveOpenAICompletionsCacheRetention(completionsModel.cacheRetention, env);\n\t\t\tif (retention === \"none\") return NO_PROMPT_CACHE;\n\t\t\tif (isDirectDeepSeekModel(completionsModel)) return BEST_EFFORT_PROMPT_CACHE;\n\t\t\tconst compat = getOpenAICompletionsCompat(completionsModel);\n\t\t\tif (compat.cacheControlFormat === \"anthropic\") {\n\t\t\t\treturn retention === \"long\" && compat.supportsLongCacheRetention\n\t\t\t\t\t? ttl(PROMPT_CACHE_TTL_LONG_SECONDS)\n\t\t\t\t\t: ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t\t}\n\t\t\treturn ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t}\n\t\tcase \"openai-responses\":\n\t\tcase \"openai-codex-responses\":\n\t\tcase \"azure-openai-responses\": {\n\t\t\tconst retention = resolveOpenAIResponsesCacheRetention(model.cacheRetention, env);\n\t\t\tif (retention === \"none\") return NO_PROMPT_CACHE;\n\t\t\treturn hasOpenAIExtendedPromptCache(model)\n\t\t\t\t? ttl(PROMPT_CACHE_TTL_OPENAI_EXTENDED_SECONDS)\n\t\t\t\t: ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t}\n\t\tdefault:\n\t\t\treturn NO_PROMPT_CACHE;\n\t}\n}\n\n/**\n * Explicit prompt-cache TTL in seconds, or `undefined` when the lane has no expiry contract: caching disabled,\n * unknown, or automatic best-effort (see {@link resolvePromptCacheLifetime}).\n */\nexport function resolvePromptCacheTtlSeconds(model: Model<Api>, env?: ProviderEnv): number | undefined {\n\tconst lifetime = resolvePromptCacheLifetime(model, env);\n\treturn lifetime.kind === \"ttl\" ? lifetime.ttlSeconds : undefined;\n}\n"]}
+{"version":3,"file":"prompt-cache-ttl.js","sourceRoot":"","sources":["../../src/utils/prompt-cache-ttl.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AAExD,MAAM,CAAC,MAAM,8BAA8B,GAAG,GAAG,CAAC;AAClD,MAAM,CAAC,MAAM,6BAA6B,GAAG,IAAI,CAAC;AAClD;;;GAGG;AACH,MAAM,CAAC,MAAM,wCAAwC,GAAG,IAAI,CAAC;AAe7D,MAAM,eAAe,GAAwB,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC;AAC9D,MAAM,wBAAwB,GAAwB,EAAE,IAAI,EAAE,aAAa,EAAE,CAAC;AAE9E,SAAS,GAAG,CAAC,UAAkB;IAC9B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC;AACpC,CAAC;AAED,SAAS,UAAU,CAAC,OAAe;IAClC,IAAI,CAAC;QACJ,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;IAChD,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AACF,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,OAAe;IACpD,IAAI,CAAC;QACJ,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,QAAQ,KAAK,mBAAmB,CAAC;IAC1D,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AACF,CAAC;AAED;;;;GAIG;AACH,MAAM,qCAAqC,GAAG,uEAAuE,CAAC;AAEtH;;;;GAIG;AACH,SAAS,6BAA6B,CAAC,KAAkC;IACxE,IAAI,KAAK,CAAC,QAAQ,KAAK,WAAW,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC;QAAE,OAAO,KAAK,CAAC;IAC/E,MAAM,OAAO,GAAG,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,uDAAuD,CAAC,CAAC;IACxF,IAAI,CAAC,OAAO;QAAE,OAAO,KAAK,CAAC;IAC3B,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;IACjC,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3E,OAAO,KAAK,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC;AACjD,CAAC;AAED,MAAM,UAAU,kBAAkB,CACjC,KAAkC;IAKlC,uEAAuE;IACvE,MAAM,WAAW,GAAG,KAAK,CAAC,QAAQ,KAAK,WAAW,CAAC;IACnD,MAAM,8BAA8B,GACnC,KAAK,CAAC,QAAQ,KAAK,uBAAuB,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC;IACnF,MAAM,QAAQ,GAAG,KAAK,CAAC,QAAQ,KAAK,QAAQ,IAAI,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,oBAAoB,CAAC,CAAC;IAChG,8EAA8E;IAC9E,uDAAuD;IACvD,MAAM,YAAY,GAAG,KAAK,CAAC,QAAQ,KAAK,YAAY,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;IAChG,OAAO;QACN,+BAA+B,EAAE,KAAK,CAAC,MAAM,EAAE,+BAA+B,IAAI,CAAC,WAAW;QAC9F,0BAA0B,EAAE,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,CAAC,WAAW;QACpF,0BAA0B,EACzB,KAAK,CAAC,MAAM,EAAE,0BAA0B,IAAI,CAAC,CAAC,CAAC,WAAW,IAAI,8BAA8B,IAAI,YAAY,CAAC;QAC9G,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC;QACvG,2BAA2B,EAAE,KAAK,CAAC,MAAM,EAAE,2BAA2B,IAAI,CAAC,WAAW;QACtF,wBAAwB,EAAE,KAAK,CAAC,MAAM,EAAE,wBAAwB,IAAI,CAAC,QAAQ;QAC7E,mBAAmB,EAAE,KAAK,CAAC,MAAM,EAAE,mBAAmB,IAAI,IAAI;QAC9D,kBAAkB,EAAE,KAAK,CAAC,MAAM,EAAE,kBAAkB,IAAI,IAAI;QAC5D,wBAAwB,EACvB,KAAK,CAAC,MAAM,EAAE,wBAAwB,IAAI,CAAC,qCAAqC,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAChG,mBAAmB,EAAE,KAAK,CAAC,MAAM,EAAE,mBAAmB,IAAI,KAAK;QAC/D,sBAAsB,EACrB,KAAK,CAAC,MAAM,EAAE,sBAAsB,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,mBAAmB,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,MAAM,CAAC;QACzG,qBAAqB,EAAE,KAAK,CAAC,MAAM,EAAE,qBAAqB,IAAI,EAAE;QAChE,mBAAmB,EAAE,KAAK,CAAC,MAAM,EAAE,mBAAmB,IAAI,KAAK;QAC/D,sBAAsB,EAAE,KAAK,CAAC,MAAM,EAAE,sBAAsB,IAAI,6BAA6B,CAAC,KAAK,CAAC;QACpG,sEAAsE;QACtE,0EAA0E;QAC1E,0EAA0E;QAC1E,yEAAyE;QACzE,WAAW;QACX,iBAAiB,EAAE,KAAK,CAAC,MAAM,EAAE,iBAAiB,IAAI,qBAAqB,CAAC,KAAK,CAAC,OAAO,CAAC;KAC1F,CAAC;AACH,CAAC;AAgCD;;;;GAIG;AACH,SAAS,6BAA6B,CAAC,KAAkC;IACxE,MAAM,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAChC,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;IAE9B,MAAM,KAAK,GACV,QAAQ,KAAK,KAAK;QAClB,QAAQ,KAAK,eAAe;QAC5B,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC;QAC5B,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;IACtC,MAAM,UAAU,GACf,QAAQ,KAAK,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;IACxG,MAAM,UAAU,GAAG,QAAQ,KAAK,YAAY,IAAI,QAAQ,KAAK,eAAe,IAAI,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;IAClH,MAAM,YAAY,GAAG,QAAQ,KAAK,YAAY,IAAI,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;IACpF,MAAM,qBAAqB,GAAG,QAAQ,KAAK,uBAAuB,IAAI,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IAC7G,MAAM,qBAAqB,GAAG,QAAQ,KAAK,uBAAuB,IAAI,OAAO,CAAC,QAAQ,CAAC,2BAA2B,CAAC,CAAC;IACpH,MAAM,QAAQ,GAAG,QAAQ,KAAK,QAAQ,IAAI,OAAO,CAAC,QAAQ,CAAC,0BAA0B,CAAC,CAAC;IACvF,MAAM,SAAS,GAAG,QAAQ,KAAK,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;IAClF,MAAM,UAAU,GAAG,QAAQ,KAAK,UAAU,IAAI,OAAO,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;IAE7F,MAAM,aAAa,GAClB,QAAQ;QACR,QAAQ,KAAK,UAAU;QACvB,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC;QAC/B,QAAQ,KAAK,KAAK;QAClB,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC;QAC5B,UAAU;QACV,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC;QAC7B,UAAU;QACV,KAAK;QACL,UAAU;QACV,QAAQ,KAAK,UAAU;QACvB,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAC;QAC/B,qBAAqB;QACrB,qBAAqB;QACrB,SAAS,CAAC;IAEX,MAAM,YAAY,GACjB,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC;QAC7B,UAAU;QACV,UAAU;QACV,qBAAqB;QACrB,UAAU;QACV,QAAQ;QACR,SAAS;QACT,KAAK,CAAC;IAEP,MAAM,MAAM,GAAG,QAAQ,KAAK,KAAK,IAAI,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IAClE,MAAM,8BAA8B,GACnC,YAAY,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,YAAY,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC;IACvF,MAAM,8BAA8B,GAAG,CAAC,YAAY,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IAC1E,MAAM,mBAAmB,GAAG,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC;IACpF,MAAM,8BAA8B,GAAG,8BAA8B,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CACrF,mBAAmB,CAAC,UAAU,CAAC,MAAM,CAAC,CACtC,CAAC;IACF,MAAM,kBAAkB,GAAG,QAAQ,KAAK,YAAY,IAAI,8BAA8B,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IAEjH,OAAO;QACN,aAAa,EAAE,CAAC,aAAa;QAC7B,qBAAqB,EAAE,8BAA8B,IAAI,CAAC,CAAC,aAAa,IAAI,CAAC,YAAY,CAAC;QAC1F,uBAAuB,EACtB,CAAC,MAAM,IAAI,CAAC,KAAK,IAAI,CAAC,UAAU,IAAI,CAAC,UAAU,IAAI,CAAC,qBAAqB,IAAI,CAAC,QAAQ,IAAI,CAAC,SAAS;QACrG,wBAAwB,EAAE,IAAI;QAC9B,oBAAoB,EAAE,IAAI;QAC1B,cAAc,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,uBAAuB;QACrE,sBAAsB,EAAE,KAAK;QAC7B,gCAAgC,EAAE,KAAK;QACvC,sBAAsB,EAAE,KAAK;QAC7B,2CAA2C,EAAE,UAAU;QACvD,cAAc,EAAE,UAAU;YACzB,CAAC,CAAC,UAAU;YACZ,CAAC,CAAC,KAAK;gBACN,CAAC,CAAC,KAAK;gBACP,CAAC,CAAC,UAAU;oBACX,CAAC,CAAC,UAAU;oBACZ,CAAC,CAAC,SAAS;wBACV,CAAC,CAAC,UAAU;wBACZ,CAAC,CAAC,YAAY;4BACb,CAAC,CAAC,YAAY;4BACd,CAAC,CAAC,QAAQ;QACf,iBAAiB,EAAE,EAAE;QACrB,oBAAoB,EAAE,EAAE;QACxB,kBAAkB,EAAE,EAAE;QACtB,aAAa,EAAE,KAAK;QACpB,kBAAkB,EAAE,CAAC,UAAU,IAAI,CAAC,UAAU,IAAI,CAAC,qBAAqB,IAAI,CAAC,QAAQ;QACrF,gBAAgB,EAAE,UAAU,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,SAAS;QAC1D,wBAAwB,EAAE,IAAI;QAC9B,cAAc,EAAE,SAAS;QACzB,0BAA0B,EAAE,KAAK;QACjC,kBAAkB;QAClB,0BAA0B,EAAE,YAAY;QACxC,iBAAiB,EAAE,SAAS;QAC5B,qBAAqB,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ;QAC7D,sBAAsB,EAAE,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC;QACxE,uBAAuB,EAAE,IAAI;QAC7B,0BAA0B,EAAE,CAAC,CAC5B,UAAU;YACV,qBAAqB;YACrB,qBAAqB;YACrB,QAAQ;YACR,SAAS,CACT;KACD,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,0BAA0B,CAAC,KAAkC;IAC5E,MAAM,QAAQ,GAAG,6BAA6B,CAAC,KAAK,CAAC,CAAC;IACtD,IAAI,CAAC,KAAK,CAAC,MAAM;QAAE,OAAO,QAAQ,CAAC;IAEnC,OAAO;QACN,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,qBAAqB,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,qBAAqB;QAC3F,uBAAuB,EAAE,KAAK,CAAC,MAAM,CAAC,uBAAuB,IAAI,QAAQ,CAAC,uBAAuB;QACjG,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,oBAAoB,EAAE,KAAK,CAAC,MAAM,CAAC,oBAAoB,IAAI,QAAQ,CAAC,oBAAoB;QACxF,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,gCAAgC,EAC/B,KAAK,CAAC,MAAM,CAAC,gCAAgC,IAAI,QAAQ,CAAC,gCAAgC;QAC3F,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,2CAA2C,EAC1C,KAAK,CAAC,MAAM,CAAC,2CAA2C;YACxD,QAAQ,CAAC,2CAA2C;QACrD,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,iBAAiB,IAAI,QAAQ,CAAC,iBAAiB;QAC/E,oBAAoB,EAAE,KAAK,CAAC,MAAM,CAAC,oBAAoB,IAAI,QAAQ,CAAC,oBAAoB;QACxF,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAC5E,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,2BAA2B,EAAE,KAAK,CAAC,MAAM,CAAC,2BAA2B,IAAI,QAAQ,CAAC,2BAA2B;QAC7G,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAC5E,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;QAC1G,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;QAC1G,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,iBAAiB,IAAI,QAAQ,CAAC,iBAAiB;QAC/E,qBAAqB,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,qBAAqB;QAC3F,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,uBAAuB,EAAE,KAAK,CAAC,MAAM,CAAC,uBAAuB,IAAI,QAAQ,CAAC,uBAAuB;QACjG,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB;QAC/D,YAAY,EAAE,KAAK,CAAC,MAAM,CAAC,YAAY,IAAI,QAAQ,CAAC,YAAY;QAChE,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;KAC1G,CAAC;AACH,CAAC;AAED,MAAM,UAAU,8BAA8B,CAAC,OAAe,EAAE,SAAkB;IACjF,MAAM,MAAM,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;IAC5D,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE;QAC/B,MAAM,KAAK,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;QAClC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,CAAC;IACjD,CAAC,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAuC;IAC9E,MAAM,UAAU,GAAG,8BAA8B,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IACxE,OAAO,UAAU,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CACpC,CAAC,UAAU,EAAE,YAAY,EAAE,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,YAAY,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,YAAY,CAAC,CAAC,CAChG,CAAC;AACH,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,qBAAqB,CAAC,KAAuC,EAAE,GAAiB;IAC/F,MAAM,UAAU,GAAG,8BAA8B,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IAExE,MAAM,YAAY,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClE,IAAI,CAAC,YAAY,EAAE,CAAC;QACnB,0EAA0E;QAC1E,8DAA8D;QAC9D,IAAI,mBAAmB,CAAC,yBAAyB,EAAE,GAAG,CAAC,KAAK,GAAG;YAAE,OAAO,IAAI,CAAC;QAC7E,OAAO,KAAK,CAAC;IACd,CAAC;IACD,8CAA8C;IAC9C,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACjH,gDAAgD;IAChD,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IAC3D,oBAAoB;IACpB,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,mBAAmB,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACzE,mBAAmB;IACnB,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACxE,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,8BAA8B,CACtC,cAA+B,EAC/B,GAAiB,EACjB,QAAQ,GAAmB,OAAO;IAElC,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,OAAO,CAAC,GAAG,CAAC,kBAAkB,KAAK,SAAS,EAAE,CAAC;QACpF,OAAO,OAAO,CAAC;IAChB,CAAC;IACD,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED,SAAS,sCAAsC,CAAC,cAA+B,EAAE,GAAiB;IACjG,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,4BAA4B,CAAC,cAA+B,EAAE,GAAiB;IACvF,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,oCAAoC,CAAC,cAA+B,EAAE,GAAiB;IAC/F,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,mGAAmG;AACnG,SAAS,qBAAqB,CAAC,KAAkC;IAChE,OAAO,KAAK,CAAC,QAAQ,KAAK,UAAU,IAAI,UAAU,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,kBAAkB,CAAC;AAC1F,CAAC;AAED,4EAA4E;AAC5E,MAAM,8BAA8B,GAAG,iEAAiE,CAAC;AAEzG;;;;GAIG;AACH,SAAS,4BAA4B,CAAC,KAAiB;IACtD,IAAI,KAAK,CAAC,GAAG,KAAK,kBAAkB,EAAE,CAAC;QACtC,MAAM,cAAc,GAAG,KAAkC,CAAC;QAC1D,IAAI,cAAc,CAAC,MAAM,EAAE,+BAA+B,KAAK,IAAI;YAAE,OAAO,IAAI,CAAC;QACjF,IAAI,cAAc,CAAC,QAAQ,KAAK,QAAQ,IAAI,UAAU,CAAC,cAAc,CAAC,OAAO,CAAC,KAAK,gBAAgB;YAAE,OAAO,KAAK,CAAC;IACnH,CAAC;IACD,OAAO,8BAA8B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;AACtD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,0BAA0B,CAAC,KAAiB,EAAE,GAAiB;IAC9E,QAAQ,KAAK,CAAC,GAAG,EAAE,CAAC;QACnB,KAAK,kBAAkB;YACtB,wFAAwF;YACxF,OAAO,GAAG,CAAC,8BAA8B,CAAC,CAAC;QAC5C,KAAK,oBAAoB,EAAE,CAAC;YAC3B,MAAM,cAAc,GAAG,KAAoC,CAAC;YAC5D,MAAM,SAAS,GAAG,8BAA8B,CAAC,cAAc,CAAC,cAAc,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;YAC9F,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,eAAe,CAAC;YACjD,OAAO,SAAS,KAAK,MAAM;gBAC1B,qBAAqB,CAAC,cAAc,CAAC,OAAO,CAAC;gBAC7C,kBAAkB,CAAC,cAAc,CAAC,CAAC,0BAA0B;gBAC7D,CAAC,CAAC,GAAG,CAAC,6BAA6B,CAAC;gBACpC,CAAC,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;QACxC,CAAC;QACD,KAAK,yBAAyB,EAAE,CAAC;YAChC,MAAM,YAAY,GAAG,KAAyC,CAAC;YAC/D,MAAM,SAAS,GAAG,4BAA4B,CAAC,YAAY,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;YACjF,IAAI,SAAS,KAAK,MAAM,IAAI,CAAC,qBAAqB,CAAC,YAAY,EAAE,GAAG,CAAC;gBAAE,OAAO,eAAe,CAAC;YAC9F,OAAO,SAAS,KAAK,MAAM,IAAI,uBAAuB,CAAC,YAAY,CAAC;gBACnE,CAAC,CAAC,GAAG,CAAC,6BAA6B,CAAC;gBACpC,CAAC,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;QACxC,CAAC;QACD,KAAK,oBAAoB,EAAE,CAAC;YAC3B,MAAM,gBAAgB,GAAG,KAAoC,CAAC;YAC9D,MAAM,SAAS,GAAG,sCAAsC,CAAC,gBAAgB,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;YAC/F,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,eAAe,CAAC;YACjD,IAAI,qBAAqB,CAAC,gBAAgB,CAAC;gBAAE,OAAO,wBAAwB,CAAC;YAC7E,MAAM,MAAM,GAAG,0BAA0B,CAAC,gBAAgB,CAAC,CAAC;YAC5D,IAAI,MAAM,CAAC,kBAAkB,KAAK,WAAW,EAAE,CAAC;gBAC/C,OAAO,SAAS,KAAK,MAAM,IAAI,MAAM,CAAC,0BAA0B;oBAC/D,CAAC,CAAC,GAAG,CAAC,6BAA6B,CAAC;oBACpC,CAAC,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;YACxC,CAAC;YACD,OAAO,GAAG,CAAC,8BAA8B,CAAC,CAAC;QAC5C,CAAC;QACD,KAAK,kBAAkB,CAAC;QACxB,KAAK,wBAAwB,CAAC;QAC9B,KAAK,wBAAwB,EAAE,CAAC;YAC/B,MAAM,SAAS,GAAG,oCAAoC,CAAC,KAAK,CAAC,cAAc,EAAE,GAAG,CAAC,CAAC;YAClF,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,eAAe,CAAC;YACjD,OAAO,4BAA4B,CAAC,KAAK,CAAC;gBACzC,CAAC,CAAC,GAAG,CAAC,wCAAwC,CAAC;gBAC/C,CAAC,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;QACxC,CAAC;QACD;YACC,OAAO,eAAe,CAAC;IACzB,CAAC;AACF,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,4BAA4B,CAAC,KAAiB,EAAE,GAAiB;IAChF,MAAM,QAAQ,GAAG,0BAA0B,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IACxD,OAAO,QAAQ,CAAC,IAAI,KAAK,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;AAClE,CAAC","sourcesContent":["import type {\n\tAnthropicMessagesCompat,\n\tApi,\n\tCacheRetention,\n\tModel,\n\tOpenAICompletionsCompat,\n\tProviderEnv,\n} from \"../types.ts\";\nimport { getProviderEnvValue } from \"./provider-env.ts\";\n\nexport const PROMPT_CACHE_TTL_SHORT_SECONDS = 300;\nexport const PROMPT_CACHE_TTL_LONG_SECONDS = 3600;\n/**\n * OpenAI GPT-5.6 and later: `prompt_cache_options.ttl` accepts only `\"30m\"`, and the cache stays\n * eligible at least 30 minutes after the latest write or reuse.\n */\nexport const PROMPT_CACHE_TTL_OPENAI_EXTENDED_SECONDS = 1800;\n\n/**\n * How long a provider promises to keep a prompt prefix cached.\n *\n * - `ttl`: an explicit expiry contract; `ttlSeconds` is the conservative lifetime after the last write or reuse.\n * - `best-effort`: the provider caches automatically but promises no expiry (direct DeepSeek clears unused entries\n *   after hours to days), so nothing needs to wake or ping solely to preserve the cache.\n * - `none`: caching is disabled or the lane has no known cache contract.\n */\nexport type PromptCacheLifetime =\n\t| { readonly kind: \"ttl\"; readonly ttlSeconds: number }\n\t| { readonly kind: \"best-effort\" }\n\t| { readonly kind: \"none\" };\n\nconst NO_PROMPT_CACHE: PromptCacheLifetime = { kind: \"none\" };\nconst BEST_EFFORT_PROMPT_CACHE: PromptCacheLifetime = { kind: \"best-effort\" };\n\nfunction ttl(ttlSeconds: number): PromptCacheLifetime {\n\treturn { kind: \"ttl\", ttlSeconds };\n}\n\nfunction hostnameOf(baseUrl: string): string | undefined {\n\ttry {\n\t\treturn new URL(baseUrl).hostname.toLowerCase();\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport function isAnthropicApiBaseUrl(baseUrl: string): boolean {\n\ttry {\n\t\treturn new URL(baseUrl).hostname === \"api.anthropic.com\";\n\t} catch {\n\t\treturn false;\n\t}\n}\n\n/**\n * Families that reject forced tool use (`tool_choice` `any` / `tool` return 400): every Fable and\n * Mythos release, Claude Opus 5.5 and Claude Sonnet 5.5 (`claude-opus-5-5`, `claude-sonnet-5-5`;\n * gateways may spell them with a dot).\n */\nconst FORCED_TOOL_CHOICE_REJECTING_MODEL_ID = /^claude-(?:(?:fable|mythos)(?:-|$)|(?:opus|sonnet)-5[.-]5(?:[.-]|$))/i;\n\n/**\n * Default for `supportsToolReferences`: first-party Anthropic models except\n * Haiku (rejects client-side tool_reference blocks) and models that predate\n * tool search (Claude 3.x, Opus/Sonnet 4.0, Opus 4.1).\n */\nfunction defaultSupportsToolReferences(model: Model<\"anthropic-messages\">): boolean {\n\tif (model.provider !== \"anthropic\" || model.id.includes(\"haiku\")) return false;\n\tconst version = model.id.match(/^claude-(?:opus|sonnet|fable)-(\\d+)(?:-(\\d+))?(?:-|$)/);\n\tif (!version) return false;\n\tconst major = Number(version[1]);\n\tconst minor = version[2] && version[2].length < 8 ? Number(version[2]) : 0;\n\treturn major > 4 || (major === 4 && minor >= 5);\n}\n\nexport function getAnthropicCompat(\n\tmodel: Model<\"anthropic-messages\">,\n): Required<\n\tOmit<AnthropicMessagesCompat, \"forceAdaptiveThinking\" | \"supportsMidConvoEffort\" | \"sessionAffinityFormat\">\n> &\n\tPick<AnthropicMessagesCompat, \"sessionAffinityFormat\"> {\n\t// Auto-detect session affinity and cache control support from provider\n\tconst isFireworks = model.provider === \"fireworks\";\n\tconst isCloudflareAiGatewayAnthropic =\n\t\tmodel.provider === \"cloudflare-ai-gateway\" && model.baseUrl.includes(\"anthropic\");\n\tconst isXiaomi = model.provider === \"xiaomi\" || model.provider.startsWith(\"xiaomi-token-plan-\");\n\t// OpenRouter carries prompt-cache affinity on its own x-session-id header and\n\t// rejects x-session-affinity (earendil-works/pi#9102).\n\tconst isOpenRouter = model.provider === \"openrouter\" || model.baseUrl.includes(\"openrouter.ai\");\n\treturn {\n\t\tsupportsEagerToolInputStreaming: model.compat?.supportsEagerToolInputStreaming ?? !isFireworks,\n\t\tsupportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? !isFireworks,\n\t\tsendSessionAffinityHeaders:\n\t\t\tmodel.compat?.sendSessionAffinityHeaders ?? !!(isFireworks || isCloudflareAiGatewayAnthropic || isOpenRouter),\n\t\tsessionAffinityFormat: model.compat?.sessionAffinityFormat ?? (isOpenRouter ? \"openrouter\" : undefined),\n\t\tsupportsCacheControlOnTools: model.compat?.supportsCacheControlOnTools ?? !isFireworks,\n\t\tsupportsDisabledThinking: model.compat?.supportsDisabledThinking ?? !isXiaomi,\n\t\tsupportsTemperature: model.compat?.supportsTemperature ?? true,\n\t\tsupportsToolChoice: model.compat?.supportsToolChoice ?? true,\n\t\tsupportsForcedToolChoice:\n\t\t\tmodel.compat?.supportsForcedToolChoice ?? !FORCED_TOOL_CHOICE_REJECTING_MODEL_ID.test(model.id),\n\t\tallowEmptySignature: model.compat?.allowEmptySignature ?? false,\n\t\tunsignedThinkingReplay:\n\t\t\tmodel.compat?.unsignedThinkingReplay ?? (model.compat?.allowEmptySignature ? \"empty-signature\" : \"text\"),\n\t\tallowedFallbackModels: model.compat?.allowedFallbackModels ?? [],\n\t\tsupportsStrictTools: model.compat?.supportsStrictTools ?? false,\n\t\tsupportsToolReferences: model.compat?.supportsToolReferences ?? defaultSupportsToolReferences(model),\n\t\t// Default: first-party Anthropic only. Anthropic-compatible providers\n\t\t// (kimi-coding, fireworks, copilot, gateways) may execute the server-side\n\t\t// search but reject the replayed server_tool_use / web_search_tool_result\n\t\t// blocks on the next request (kimi-coding 400s with `tool_call_id is not\n\t\t// found`).\n\t\tsupportsWebSearch: model.compat?.supportsWebSearch ?? isAnthropicApiBaseUrl(model.baseUrl),\n\t};\n}\n\nexport type ResolvedOpenAICompletionsCompat = Omit<\n\tRequired<OpenAICompletionsCompat>,\n\t| \"vllmPriority\"\n\t| \"cacheControlFormat\"\n\t| \"toolCallFormat\"\n\t| \"deferredToolsMode\"\n\t| \"toolSchemaFlavor\"\n\t| \"supportsPromptCacheKey\"\n\t| \"chatTemplateArgs\"\n\t| \"supportsThinkingTokenBudget\"\n\t| \"thinkingTokenBudgetField\"\n\t| \"veniceParameters\"\n\t| \"supportsForcedToolChoice\"\n> & {\n\t/** Declared forced tool_choice support; absent means supported. */\n\tsupportsForcedToolChoice?: OpenAICompletionsCompat[\"supportsForcedToolChoice\"];\n\tcacheControlFormat?: OpenAICompletionsCompat[\"cacheControlFormat\"];\n\tsupportsPromptCacheKey?: OpenAICompletionsCompat[\"supportsPromptCacheKey\"];\n\ttoolCallFormat?: OpenAICompletionsCompat[\"toolCallFormat\"];\n\tdeferredToolsMode?: OpenAICompletionsCompat[\"deferredToolsMode\"];\n\ttoolSchemaFlavor?: OpenAICompletionsCompat[\"toolSchemaFlavor\"];\n\tchatTemplateArgs?: OpenAICompletionsCompat[\"chatTemplateArgs\"];\n\tsupportsThinkingTokenBudget?: OpenAICompletionsCompat[\"supportsThinkingTokenBudget\"];\n\tthinkingTokenBudgetField?: OpenAICompletionsCompat[\"thinkingTokenBudgetField\"];\n\t/** vLLM `priority`; off by default and never set on the generated catalog. */\n\tvllmPriority?: OpenAICompletionsCompat[\"vllmPriority\"];\n\t/** Venice `venice_parameters`; only set on the generated Venice catalog. */\n\tveniceParameters?: OpenAICompletionsCompat[\"veniceParameters\"];\n};\n\n/**\n * Detect compatibility settings from provider and baseUrl for known providers.\n * Provider takes precedence over URL-based detection since it's explicitly configured.\n * Returns a fully resolved OpenAICompletionsCompat object with all fields set.\n */\nfunction detectOpenAICompletionsCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst provider = model.provider;\n\tconst baseUrl = model.baseUrl;\n\n\tconst isZai =\n\t\tprovider === \"zai\" ||\n\t\tprovider === \"zai-coding-cn\" ||\n\t\tbaseUrl.includes(\"api.z.ai\") ||\n\t\tbaseUrl.includes(\"open.bigmodel.cn\");\n\tconst isTogether =\n\t\tprovider === \"together\" || baseUrl.includes(\"api.together.ai\") || baseUrl.includes(\"api.together.xyz\");\n\tconst isMoonshot = provider === \"moonshotai\" || provider === \"moonshotai-cn\" || baseUrl.includes(\"api.moonshot.\");\n\tconst isOpenRouter = provider === \"openrouter\" || baseUrl.includes(\"openrouter.ai\");\n\tconst isCloudflareWorkersAI = provider === \"cloudflare-workers-ai\" || baseUrl.includes(\"api.cloudflare.com\");\n\tconst isCloudflareAiGateway = provider === \"cloudflare-ai-gateway\" || baseUrl.includes(\"gateway.ai.cloudflare.com\");\n\tconst isNvidia = provider === \"nvidia\" || baseUrl.includes(\"integrate.api.nvidia.com\");\n\tconst isAntLing = provider === \"ant-ling\" || baseUrl.includes(\"api.ant-ling.com\");\n\tconst isDeepSeek = provider === \"deepseek\" || baseUrl.toLowerCase().includes(\"deepseek.com\");\n\n\tconst isNonStandard =\n\t\tisNvidia ||\n\t\tprovider === \"cerebras\" ||\n\t\tbaseUrl.includes(\"cerebras.ai\") ||\n\t\tprovider === \"xai\" ||\n\t\tbaseUrl.includes(\"api.x.ai\") ||\n\t\tisTogether ||\n\t\tbaseUrl.includes(\"chutes.ai\") ||\n\t\tisDeepSeek ||\n\t\tisZai ||\n\t\tisMoonshot ||\n\t\tprovider === \"opencode\" ||\n\t\tbaseUrl.includes(\"opencode.ai\") ||\n\t\tisCloudflareWorkersAI ||\n\t\tisCloudflareAiGateway ||\n\t\tisAntLing;\n\n\tconst useMaxTokens =\n\t\tbaseUrl.includes(\"chutes.ai\") ||\n\t\tisDeepSeek ||\n\t\tisMoonshot ||\n\t\tisCloudflareAiGateway ||\n\t\tisTogether ||\n\t\tisNvidia ||\n\t\tisAntLing ||\n\t\tisZai;\n\n\tconst isGrok = provider === \"xai\" || baseUrl.includes(\"api.x.ai\");\n\tconst isOpenRouterDeveloperRoleModel =\n\t\tisOpenRouter && (model.id.startsWith(\"anthropic/\") || model.id.startsWith(\"openai/\"));\n\tconst openRouterCacheControlPrefixes = [\"anthropic/\", \"qwen/\", \"google/\"];\n\tconst cacheControlModelId = model.id.startsWith(\"~\") ? model.id.slice(1) : model.id;\n\tconst supportsOpenRouterCacheControl = openRouterCacheControlPrefixes.some((prefix) =>\n\t\tcacheControlModelId.startsWith(prefix),\n\t);\n\tconst cacheControlFormat = provider === \"openrouter\" && supportsOpenRouterCacheControl ? \"anthropic\" : undefined;\n\n\treturn {\n\t\tsupportsStore: !isNonStandard,\n\t\tsupportsDeveloperRole: isOpenRouterDeveloperRoleModel || (!isNonStandard && !isOpenRouter),\n\t\tsupportsReasoningEffort:\n\t\t\t!isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia && !isAntLing,\n\t\tsupportsUsageInStreaming: true,\n\t\tsupportsFinishReason: true,\n\t\tmaxTokensField: useMaxTokens ? \"max_tokens\" : \"max_completion_tokens\",\n\t\trequiresToolResultName: false,\n\t\trequiresAssistantAfterToolResult: false,\n\t\trequiresThinkingAsText: false,\n\t\trequiresReasoningContentOnAssistantMessages: isDeepSeek,\n\t\tthinkingFormat: isDeepSeek\n\t\t\t? \"deepseek\"\n\t\t\t: isZai\n\t\t\t\t? \"zai\"\n\t\t\t\t: isTogether\n\t\t\t\t\t? \"together\"\n\t\t\t\t\t: isAntLing\n\t\t\t\t\t\t? \"ant-ling\"\n\t\t\t\t\t\t: isOpenRouter\n\t\t\t\t\t\t\t? \"openrouter\"\n\t\t\t\t\t\t\t: \"openai\",\n\t\topenRouterRouting: {},\n\t\tvercelGatewayRouting: {},\n\t\tchatTemplateKwargs: {},\n\t\tzaiToolStream: false,\n\t\tsupportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,\n\t\ttoolSchemaFlavor: isMoonshot ? \"moonshot-mfjs\" : undefined,\n\t\tsupportsDisabledThinking: true,\n\t\ttoolCallFormat: undefined,\n\t\tsupportsOpenAIGrammarTools: false,\n\t\tcacheControlFormat,\n\t\tsendSessionAffinityHeaders: isOpenRouter,\n\t\tdeferredToolsMode: undefined,\n\t\tsessionAffinityFormat: isOpenRouter ? \"openrouter\" : \"openai\",\n\t\tsupportsPromptCacheKey: isMoonshot || baseUrl.includes(\"api.openai.com\"),\n\t\tsupportsMaxOutputTokens: true,\n\t\tsupportsLongCacheRetention: !(\n\t\t\tisTogether ||\n\t\t\tisCloudflareWorkersAI ||\n\t\t\tisCloudflareAiGateway ||\n\t\t\tisNvidia ||\n\t\t\tisAntLing\n\t\t),\n\t};\n}\n\n/**\n * Get resolved compatibility settings for a model.\n * Auto-detects from provider/URL then overrides with explicit model.compat.\n */\nexport function getOpenAICompletionsCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst detected = detectOpenAICompletionsCompat(model);\n\tif (!model.compat) return detected;\n\n\treturn {\n\t\tsupportsStore: model.compat.supportsStore ?? detected.supportsStore,\n\t\tsupportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,\n\t\tsupportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,\n\t\tsupportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? detected.supportsUsageInStreaming,\n\t\tsupportsFinishReason: model.compat.supportsFinishReason ?? detected.supportsFinishReason,\n\t\tmaxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,\n\t\trequiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,\n\t\trequiresAssistantAfterToolResult:\n\t\t\tmodel.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,\n\t\trequiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,\n\t\trequiresReasoningContentOnAssistantMessages:\n\t\t\tmodel.compat.requiresReasoningContentOnAssistantMessages ??\n\t\t\tdetected.requiresReasoningContentOnAssistantMessages,\n\t\tthinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,\n\t\tsupportsDisabledThinking: model.compat.supportsDisabledThinking ?? detected.supportsDisabledThinking,\n\t\topenRouterRouting: model.compat.openRouterRouting ?? detected.openRouterRouting,\n\t\tvercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,\n\t\tchatTemplateKwargs: model.compat.chatTemplateKwargs ?? detected.chatTemplateKwargs,\n\t\tchatTemplateArgs: model.compat.chatTemplateArgs ?? detected.chatTemplateArgs,\n\t\tzaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,\n\t\tsupportsThinkingTokenBudget: model.compat.supportsThinkingTokenBudget ?? detected.supportsThinkingTokenBudget,\n\t\tthinkingTokenBudgetField: model.compat.thinkingTokenBudgetField ?? detected.thinkingTokenBudgetField,\n\t\tsupportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,\n\t\ttoolSchemaFlavor: model.compat.toolSchemaFlavor ?? detected.toolSchemaFlavor,\n\t\ttoolCallFormat: model.compat.toolCallFormat ?? detected.toolCallFormat,\n\t\tsupportsOpenAIGrammarTools: model.compat.supportsOpenAIGrammarTools ?? detected.supportsOpenAIGrammarTools,\n\t\tcacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,\n\t\tsendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,\n\t\tdeferredToolsMode: model.compat.deferredToolsMode ?? detected.deferredToolsMode,\n\t\tsessionAffinityFormat: model.compat.sessionAffinityFormat ?? detected.sessionAffinityFormat,\n\t\tsupportsPromptCacheKey: model.compat.supportsPromptCacheKey ?? detected.supportsPromptCacheKey,\n\t\tsupportsMaxOutputTokens: model.compat.supportsMaxOutputTokens ?? detected.supportsMaxOutputTokens,\n\t\tsupportsForcedToolChoice: model.compat.supportsForcedToolChoice,\n\t\tvllmPriority: model.compat.vllmPriority ?? detected.vllmPriority,\n\t\tsupportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,\n\t};\n}\n\nexport function getBedrockModelMatchCandidates(modelId: string, modelName?: string): string[] {\n\tconst values = modelName ? [modelId, modelName] : [modelId];\n\treturn values.flatMap((value) => {\n\t\tconst lower = value.toLowerCase();\n\t\treturn [lower, lower.replace(/[\\s_.:]+/g, \"-\")];\n\t});\n}\n\nexport function supportsOneHourCacheTtl(model: Model<\"bedrock-converse-stream\">): boolean {\n\tconst candidates = getBedrockModelMatchCandidates(model.id, model.name);\n\treturn candidates.some((candidate) =>\n\t\t[\"opus-4-5\", \"sonnet-4-5\", \"haiku-4-5\"].some((modelVersion) => candidate.includes(modelVersion)),\n\t);\n}\n\n/**\n * Check if the model supports prompt caching.\n * Supported: Claude 3.5 Haiku, Claude 3.7 Sonnet, Claude 4.x models, Claude 5 models\n *\n * For base models and system-defined inference profiles the model ID / ARN\n * contains the model name, so we can decide locally.\n *\n * For application inference profiles (whose ARNs don't contain the model name),\n * also checks model.name which is user-controlled via models.json or registerProvider.\n * As a last resort, set AWS_BEDROCK_FORCE_CACHE=1 to enable cache points.\n * Amazon Nova models have automatic caching and don't need explicit cache points.\n */\nexport function supportsPromptCaching(model: Model<\"bedrock-converse-stream\">, env?: ProviderEnv): boolean {\n\tconst candidates = getBedrockModelMatchCandidates(model.id, model.name);\n\n\tconst hasClaudeRef = candidates.some((s) => s.includes(\"claude\"));\n\tif (!hasClaudeRef) {\n\t\t// Application inference profiles don't contain the model name in the ARN.\n\t\t// Allow users to force cache points via environment variable.\n\t\tif (getProviderEnvValue(\"AWS_BEDROCK_FORCE_CACHE\", env) === \"1\") return true;\n\t\treturn false;\n\t}\n\t// Claude 5 models (fable-5, opus-5, sonnet-5)\n\tif (candidates.some((s) => s.includes(\"fable-5\") || s.includes(\"opus-5\") || s.includes(\"sonnet-5\"))) return true;\n\t// Claude 4.x models (opus-4, sonnet-4, haiku-4)\n\tif (candidates.some((s) => s.includes(\"-4-\"))) return true;\n\t// Claude 3.7 Sonnet\n\tif (candidates.some((s) => s.includes(\"claude-3-7-sonnet\"))) return true;\n\t// Claude 3.5 Haiku\n\tif (candidates.some((s) => s.includes(\"claude-3-5-haiku\"))) return true;\n\treturn false;\n}\n\nfunction resolveAnthropicCacheRetention(\n\tcacheRetention?: CacheRetention,\n\tenv?: ProviderEnv,\n\tfallback: CacheRetention = \"short\",\n): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\tif (typeof process !== \"undefined\" && process.env.PI_CACHE_RETENTION !== undefined) {\n\t\treturn \"short\";\n\t}\n\treturn fallback;\n}\n\nfunction resolveOpenAICompletionsCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction resolveBedrockCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction resolveOpenAIResponsesCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\n/** Direct DeepSeek API: its automatic disk cache is best-effort and carries no expiry contract. */\nfunction isDirectDeepSeekModel(model: Model<\"openai-completions\">): boolean {\n\treturn model.provider === \"deepseek\" || hostnameOf(model.baseUrl) === \"api.deepseek.com\";\n}\n\n/** GPT-5.6 and later (`gpt-5.6-*`, `gpt-5.10`, `gpt-6-*`, `gpt-7`, ...). */\nconst OPENAI_EXTENDED_CACHE_MODEL_ID = /^gpt-(?:5\\.(?:[6-9]|\\d{2,})|(?:[6-9]|\\d{2,})(?:\\.\\d+)?)(?:-|$)/i;\n\n/**\n * The OpenAI-operated Responses lanes (OpenAI API, Azure OpenAI, ChatGPT subscription) serving a GPT-5.6+\n * model, where OpenAI documents a cache lifetime of at least 30 minutes. Gateways that merely proxy the same\n * model ids keep the conservative short lifetime because their routing does not carry that contract.\n */\nfunction hasOpenAIExtendedPromptCache(model: Model<Api>): boolean {\n\tif (model.api === \"openai-responses\") {\n\t\tconst responsesModel = model as Model<\"openai-responses\">;\n\t\tif (responsesModel.compat?.supportsExplicitPromptCacheMode === true) return true;\n\t\tif (responsesModel.provider !== \"openai\" && hostnameOf(responsesModel.baseUrl) !== \"api.openai.com\") return false;\n\t}\n\treturn OPENAI_EXTENDED_CACHE_MODEL_ID.test(model.id);\n}\n\n/**\n * Classify the active model's prompt-cache lifetime from the provider's documented cache contract.\n * `cacheRetention: \"none\"` (or `PI_CACHE_RETENTION` resolving to it) always wins.\n */\nexport function resolvePromptCacheLifetime(model: Model<Api>, env?: ProviderEnv): PromptCacheLifetime {\n\tswitch (model.api) {\n\t\tcase \"claude-sdk-oauth\":\n\t\t\t// The Claude SDK owns prompt caching for this lane and uses Anthropic's default 5m TTL.\n\t\t\treturn ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\tcase \"anthropic-messages\": {\n\t\t\tconst anthropicModel = model as Model<\"anthropic-messages\">;\n\t\t\tconst retention = resolveAnthropicCacheRetention(anthropicModel.cacheRetention, env, \"short\");\n\t\t\tif (retention === \"none\") return NO_PROMPT_CACHE;\n\t\t\treturn retention === \"long\" &&\n\t\t\t\tisAnthropicApiBaseUrl(anthropicModel.baseUrl) &&\n\t\t\t\tgetAnthropicCompat(anthropicModel).supportsLongCacheRetention\n\t\t\t\t? ttl(PROMPT_CACHE_TTL_LONG_SECONDS)\n\t\t\t\t: ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t}\n\t\tcase \"bedrock-converse-stream\": {\n\t\t\tconst bedrockModel = model as Model<\"bedrock-converse-stream\">;\n\t\t\tconst retention = resolveBedrockCacheRetention(bedrockModel.cacheRetention, env);\n\t\t\tif (retention === \"none\" || !supportsPromptCaching(bedrockModel, env)) return NO_PROMPT_CACHE;\n\t\t\treturn retention === \"long\" && supportsOneHourCacheTtl(bedrockModel)\n\t\t\t\t? ttl(PROMPT_CACHE_TTL_LONG_SECONDS)\n\t\t\t\t: ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t}\n\t\tcase \"openai-completions\": {\n\t\t\tconst completionsModel = model as Model<\"openai-completions\">;\n\t\t\tconst retention = resolveOpenAICompletionsCacheRetention(completionsModel.cacheRetention, env);\n\t\t\tif (retention === \"none\") return NO_PROMPT_CACHE;\n\t\t\tif (isDirectDeepSeekModel(completionsModel)) return BEST_EFFORT_PROMPT_CACHE;\n\t\t\tconst compat = getOpenAICompletionsCompat(completionsModel);\n\t\t\tif (compat.cacheControlFormat === \"anthropic\") {\n\t\t\t\treturn retention === \"long\" && compat.supportsLongCacheRetention\n\t\t\t\t\t? ttl(PROMPT_CACHE_TTL_LONG_SECONDS)\n\t\t\t\t\t: ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t\t}\n\t\t\treturn ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t}\n\t\tcase \"openai-responses\":\n\t\tcase \"openai-codex-responses\":\n\t\tcase \"azure-openai-responses\": {\n\t\t\tconst retention = resolveOpenAIResponsesCacheRetention(model.cacheRetention, env);\n\t\t\tif (retention === \"none\") return NO_PROMPT_CACHE;\n\t\t\treturn hasOpenAIExtendedPromptCache(model)\n\t\t\t\t? ttl(PROMPT_CACHE_TTL_OPENAI_EXTENDED_SECONDS)\n\t\t\t\t: ttl(PROMPT_CACHE_TTL_SHORT_SECONDS);\n\t\t}\n\t\tdefault:\n\t\t\treturn NO_PROMPT_CACHE;\n\t}\n}\n\n/**\n * Explicit prompt-cache TTL in seconds, or `undefined` when the lane has no expiry contract: caching disabled,\n * unknown, or automatic best-effort (see {@link resolvePromptCacheLifetime}).\n */\nexport function resolvePromptCacheTtlSeconds(model: Model<Api>, env?: ProviderEnv): number | undefined {\n\tconst lifetime = resolvePromptCacheLifetime(model, env);\n\treturn lifetime.kind === \"ttl\" ? lifetime.ttlSeconds : undefined;\n}\n"]}

package/dist/utils/retry.d.ts

@@ -1,4 +1,7 @@
 import type { AssistantMessage } from "../types.ts";
+export declare const PROVIDER_REQUEST_ID_MARKER = "request id:";
+export declare function formatProviderRequestId(label: string, id: string): string;
+export declare function stripProviderRequestIds(text: string): string;
 /**
  * OpenAI hard account-quota exhaustion (senpi#1969). The wire evidence is a 429
  * whose body is `{"type":"usage_limit_reached","message":"The usage limit has

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

@@ -1 +1 @@
-{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAWpD;;;;;;;;;GASG;AACH,eAAO,MAAM,sBAAsB;IAClC,+EAA+E;aAC/E,KAAK,YAAG,qBAAqB,EAAE,oBAAoB;IACnD,+EAA+E;aAC/E,OAAO,YAAG,qBAAqB,EAAE,oBAAoB,EAAE,8BAA8B;CAC5E,CAAC;AA0CX,wBAAgB,wBAAwB,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAElF;AAoID;;;;;;GAMG;AACH,MAAM,WAAW,WAAW;IAC3B,OAAO,EAAE,OAAO,CAAC;IACjB,qFAAqF;IACrF,UAAU,EAAE,MAAM,CAAC;IACnB,0FAA0F;IAC1F,WAAW,EAAE,MAAM,CAAC;IACpB,+EAA+E;IAC/E,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,oEAAoE;IACpE,MAAM,CAAC,EAAE,MAAM,MAAM,CAAC;CACtB;AAED,eAAO,MAAM,gCAAgC,QAAS,CAAC;AAEvD;;;;;;GAMG;AACH,wBAAgB,YAAY,CAC3B,MAAM,EAAE,IAAI,CAAC,WAAW,EAAE,aAAa,GAAG,iBAAiB,GAAG,QAAQ,CAAC,EACvE,OAAO,EAAE,MAAM,GACb,MAAM,CAMR;AAED,kFAAkF;AAClF,MAAM,WAAW,cAAc;IAC9B,0EAA0E;IAC1E,gBAAgB,CAAC,EAAE,CAClB,OAAO,EAAE,MAAM,EACf,WAAW,EAAE,MAAM,EACnB,OAAO,EAAE,MAAM,EACf,YAAY,EAAE,MAAM,KAChB,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,mFAAmF;IACnF,mBAAmB,CAAC,EAAE,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACjD,mFAAmF;IACnF,eAAe,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACnG;AA0BD;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,wBAAsB,kBAAkB,CAAC,CAAC,EACzC,OAAO,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACzB,WAAW,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,CAAC,CAAC,CAiCZ;AAOD;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAsB,kBAAkB,CACvC,OAAO,EAAE,MAAM,OAAO,CAAC,gBAAgB,CAAC,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,gBAAgB,CAAC,CA6C3B;AAED;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAU5E;AAgBD,wBAAgB,0BAA0B,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAE7E;AA8BD,MAAM,WAAW,+BAA+B;IAC/C,sDAAsD;IACtD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iFAAiF;IACjF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,QAAQ,CAAC,EAAE,wBAAwB,GAAG,iBAAiB,CAAC;CACxD;AAED,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAI7D;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,4BAA4B,CAC3C,YAAY,EAAE,MAAM,GAAG,SAAS,EAChC,OAAO,GAAE,+BAAoC,GAC3C,MAAM,GAAG,SAAS,CAuBpB;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAKzE;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,YAAY,EAAE,MAAM,GAAG,OAAO,CAErE;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,YAAY,EAAE,MAAM,GAAG,eAAe,GAAG,WAAW,GAAG,SAAS,CAKpG"}
+{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAMpD,eAAO,MAAM,0BAA0B,gBAAgB,CAAC;AAIxD,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,GAAG,MAAM,CAEzE;AAED,wBAAgB,uBAAuB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAE5D;AAUD;;;;;;;;;GASG;AACH,eAAO,MAAM,sBAAsB;IAClC,+EAA+E;aAC/E,KAAK,YAAG,qBAAqB,EAAE,oBAAoB;IACnD,+EAA+E;aAC/E,OAAO,YAAG,qBAAqB,EAAE,oBAAoB,EAAE,8BAA8B;CAC5E,CAAC;AA0CX,wBAAgB,wBAAwB,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAElF;AAoID;;;;;;GAMG;AACH,MAAM,WAAW,WAAW;IAC3B,OAAO,EAAE,OAAO,CAAC;IACjB,qFAAqF;IACrF,UAAU,EAAE,MAAM,CAAC;IACnB,0FAA0F;IAC1F,WAAW,EAAE,MAAM,CAAC;IACpB,+EAA+E;IAC/E,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,oEAAoE;IACpE,MAAM,CAAC,EAAE,MAAM,MAAM,CAAC;CACtB;AAED,eAAO,MAAM,gCAAgC,QAAS,CAAC;AAEvD;;;;;;GAMG;AACH,wBAAgB,YAAY,CAC3B,MAAM,EAAE,IAAI,CAAC,WAAW,EAAE,aAAa,GAAG,iBAAiB,GAAG,QAAQ,CAAC,EACvE,OAAO,EAAE,MAAM,GACb,MAAM,CAMR;AAED,kFAAkF;AAClF,MAAM,WAAW,cAAc;IAC9B,0EAA0E;IAC1E,gBAAgB,CAAC,EAAE,CAClB,OAAO,EAAE,MAAM,EACf,WAAW,EAAE,MAAM,EACnB,OAAO,EAAE,MAAM,EACf,YAAY,EAAE,MAAM,KAChB,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,mFAAmF;IACnF,mBAAmB,CAAC,EAAE,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACjD,mFAAmF;IACnF,eAAe,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACnG;AA0BD;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,wBAAsB,kBAAkB,CAAC,CAAC,EACzC,OAAO,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACzB,WAAW,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,CAAC,CAAC,CAiCZ;AAOD;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAsB,kBAAkB,CACvC,OAAO,EAAE,MAAM,OAAO,CAAC,gBAAgB,CAAC,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,gBAAgB,CAAC,CA6C3B;AAED;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAU5E;AAgBD,wBAAgB,0BAA0B,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAE7E;AA8BD,MAAM,WAAW,+BAA+B;IAC/C,sDAAsD;IACtD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iFAAiF;IACjF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,QAAQ,CAAC,EAAE,wBAAwB,GAAG,iBAAiB,CAAC;CACxD;AAED,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAI7D;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,4BAA4B,CAC3C,YAAY,EAAE,MAAM,GAAG,SAAS,EAChC,OAAO,GAAE,+BAAoC,GAC3C,MAAM,GAAG,SAAS,CAuBpB;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAKzE;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,YAAY,EAAE,MAAM,GAAG,OAAO,CAErE;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,YAAY,EAAE,MAAM,GAAG,eAAe,GAAG,WAAW,GAAG,SAAS,CAMpG"}

package/dist/utils/retry.js

@@ -1,4 +1,15 @@
 import { FORWARDED_EMPTY_RESPONSE_ERROR, FORWARDED_EMPTY_TOOL_USE_ERROR } from "./empty-response-errors.js";
+// A provider's support request id is opaque hex like `C6FD:AB660:AEB5548:6ABA4D81`.
+// It can contain `429` or `500`, which message classifiers read as HTTP statuses,
+// so every id is rendered behind this marker and removed before classification.
+export const PROVIDER_REQUEST_ID_MARKER = "request id:";
+const REQUEST_ID_SEGMENT = /request id: [^\s,;)]+/gi;
+export function formatProviderRequestId(label, id) {
+    return `${label} ${PROVIDER_REQUEST_ID_MARKER} ${id}`;
+}
+export function stripProviderRequestIds(text) {
+    return text.replace(REQUEST_ID_SEGMENT, PROVIDER_REQUEST_ID_MARKER);
+}
 function buildProviderErrorPattern(patterns) {
     return new RegExp(patterns.join("|"), "i");
 }
@@ -56,7 +67,7 @@ const QUOTA_EXHAUSTION_PATTERNS = [
 ];
 const QUOTA_EXHAUSTION_PATTERN = buildProviderErrorPattern(QUOTA_EXHAUSTION_PATTERNS);
 export function isQuotaExhaustionMessage(errorMessage) {
-    return errorMessage !== undefined && QUOTA_EXHAUSTION_PATTERN.test(errorMessage);
+    return errorMessage !== undefined && QUOTA_EXHAUSTION_PATTERN.test(stripProviderRequestIds(errorMessage));
 }
 const NON_RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([
     ...QUOTA_EXHAUSTION_PATTERNS,
@@ -463,9 +474,10 @@ export function isRetryableErrorMessage(errorMessage) {
 export function classifyErrorMessage(errorMessage) {
     if (!errorMessage)
         return "unknown";
-    if (NON_RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage))
+    const text = stripProviderRequestIds(errorMessage);
+    if (NON_RETRYABLE_PROVIDER_ERROR_PATTERN.test(text))
         return "non-retryable";
-    if (RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage))
+    if (RETRYABLE_PROVIDER_ERROR_PATTERN.test(text))
         return "retryable";
     return "unknown";
 }