@code-yeongyu/senpi-ai 2026.7.26 → 2026.7.28

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

@@ -0,0 +1,294 @@
+import { getProviderEnvValue } from "./provider-env.js";
+export const PROMPT_CACHE_TTL_SHORT_SECONDS = 300;
+export const PROMPT_CACHE_TTL_LONG_SECONDS = 3600;
+export function isAnthropicApiBaseUrl(baseUrl) {
+    try {
+        return new URL(baseUrl).hostname === "api.anthropic.com";
+    }
+    catch {
+        return false;
+    }
+}
+const CLAUDE_FABLE_OR_MYTHOS_MODEL_ID = /^claude-(?:fable|mythos)(?:-|$)/i;
+/**
+ * Default for `supportsToolReferences`: first-party Anthropic models except
+ * Haiku (rejects client-side tool_reference blocks) and models that predate
+ * tool search (Claude 3.x, Opus/Sonnet 4.0, Opus 4.1).
+ */
+function defaultSupportsToolReferences(model) {
+    if (model.provider !== "anthropic" || model.id.includes("haiku"))
+        return false;
+    const version = model.id.match(/^claude-(?:opus|sonnet|fable)-(\d+)(?:-(\d+))?(?:-|$)/);
+    if (!version)
+        return false;
+    const major = Number(version[1]);
+    const minor = version[2] && version[2].length < 8 ? Number(version[2]) : 0;
+    return major > 4 || (major === 4 && minor >= 5);
+}
+export function getAnthropicCompat(model) {
+    // Auto-detect session affinity and cache control support from provider
+    const isFireworks = model.provider === "fireworks";
+    const isCloudflareAiGatewayAnthropic = model.provider === "cloudflare-ai-gateway" && model.baseUrl.includes("anthropic");
+    const isXiaomi = model.provider === "xiaomi" || model.provider.startsWith("xiaomi-token-plan-");
+    return {
+        supportsEagerToolInputStreaming: model.compat?.supportsEagerToolInputStreaming ?? !isFireworks,
+        supportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? !isFireworks,
+        sendSessionAffinityHeaders: model.compat?.sendSessionAffinityHeaders ?? !!(isFireworks || isCloudflareAiGatewayAnthropic),
+        supportsCacheControlOnTools: model.compat?.supportsCacheControlOnTools ?? !isFireworks,
+        supportsDisabledThinking: model.compat?.supportsDisabledThinking ?? !isXiaomi,
+        supportsTemperature: model.compat?.supportsTemperature ?? true,
+        supportsToolChoice: model.compat?.supportsToolChoice ?? true,
+        supportsForcedToolChoice: model.compat?.supportsForcedToolChoice ?? !CLAUDE_FABLE_OR_MYTHOS_MODEL_ID.test(model.id),
+        allowEmptySignature: model.compat?.allowEmptySignature ?? false,
+        unsignedThinkingReplay: model.compat?.unsignedThinkingReplay ?? (model.compat?.allowEmptySignature ? "empty-signature" : "text"),
+        supportsStrictTools: model.compat?.supportsStrictTools ?? false,
+        supportsToolReferences: model.compat?.supportsToolReferences ?? defaultSupportsToolReferences(model),
+        // Default: first-party Anthropic only. Anthropic-compatible providers
+        // (kimi-coding, fireworks, copilot, gateways) may execute the server-side
+        // search but reject the replayed server_tool_use / web_search_tool_result
+        // blocks on the next request (kimi-coding 400s with `tool_call_id is not
+        // found`).
+        supportsWebSearch: model.compat?.supportsWebSearch ?? isAnthropicApiBaseUrl(model.baseUrl),
+    };
+}
+/**
+ * Detect compatibility settings from provider and baseUrl for known providers.
+ * Provider takes precedence over URL-based detection since it's explicitly configured.
+ * Returns a fully resolved OpenAICompletionsCompat object with all fields set.
+ */
+function detectOpenAICompletionsCompat(model) {
+    const provider = model.provider;
+    const baseUrl = model.baseUrl;
+    const isZai = provider === "zai" ||
+        provider === "zai-coding-cn" ||
+        baseUrl.includes("api.z.ai") ||
+        baseUrl.includes("open.bigmodel.cn");
+    const isTogether = provider === "together" || baseUrl.includes("api.together.ai") || baseUrl.includes("api.together.xyz");
+    const isMoonshot = provider === "moonshotai" || provider === "moonshotai-cn" || baseUrl.includes("api.moonshot.");
+    const isOpenRouter = provider === "openrouter" || baseUrl.includes("openrouter.ai");
+    const isCloudflareWorkersAI = provider === "cloudflare-workers-ai" || baseUrl.includes("api.cloudflare.com");
+    const isCloudflareAiGateway = provider === "cloudflare-ai-gateway" || baseUrl.includes("gateway.ai.cloudflare.com");
+    const isNvidia = provider === "nvidia" || baseUrl.includes("integrate.api.nvidia.com");
+    const isAntLing = provider === "ant-ling" || baseUrl.includes("api.ant-ling.com");
+    const isNonStandard = isNvidia ||
+        provider === "cerebras" ||
+        baseUrl.includes("cerebras.ai") ||
+        provider === "xai" ||
+        baseUrl.includes("api.x.ai") ||
+        isTogether ||
+        baseUrl.includes("chutes.ai") ||
+        baseUrl.includes("deepseek.com") ||
+        isZai ||
+        isMoonshot ||
+        provider === "opencode" ||
+        baseUrl.includes("opencode.ai") ||
+        isCloudflareWorkersAI ||
+        isCloudflareAiGateway ||
+        isAntLing;
+    const useMaxTokens = baseUrl.includes("chutes.ai") || isMoonshot || isCloudflareAiGateway || isTogether || isNvidia || isAntLing;
+    const isGrok = provider === "xai" || baseUrl.includes("api.x.ai");
+    const isDeepSeek = provider === "deepseek" || baseUrl.includes("deepseek.com");
+    const isOpenRouterDeveloperRoleModel = isOpenRouter && (model.id.startsWith("anthropic/") || model.id.startsWith("openai/"));
+    const cacheControlFormat = provider === "openrouter" && model.id.startsWith("anthropic/") ? "anthropic" : undefined;
+    return {
+        supportsStore: !isNonStandard,
+        supportsDeveloperRole: isOpenRouterDeveloperRoleModel || (!isNonStandard && !isOpenRouter),
+        supportsReasoningEffort: !isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia && !isAntLing,
+        supportsUsageInStreaming: true,
+        maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
+        requiresToolResultName: false,
+        requiresAssistantAfterToolResult: false,
+        requiresThinkingAsText: false,
+        requiresReasoningContentOnAssistantMessages: isDeepSeek,
+        thinkingFormat: isDeepSeek
+            ? "deepseek"
+            : isZai
+                ? "zai"
+                : isTogether
+                    ? "together"
+                    : isAntLing
+                        ? "ant-ling"
+                        : isOpenRouter
+                            ? "openrouter"
+                            : "openai",
+        openRouterRouting: {},
+        vercelGatewayRouting: {},
+        chatTemplateKwargs: {},
+        zaiToolStream: false,
+        supportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,
+        toolSchemaFlavor: isMoonshot ? "moonshot-mfjs" : undefined,
+        supportsDisabledThinking: true,
+        toolCallFormat: undefined,
+        supportsOpenAIGrammarTools: false,
+        cacheControlFormat,
+        sendSessionAffinityHeaders: false,
+        deferredToolsMode: undefined,
+        sessionAffinityFormat: isOpenRouter ? "openrouter" : "openai",
+        supportsLongCacheRetention: !(isTogether ||
+            isCloudflareWorkersAI ||
+            isCloudflareAiGateway ||
+            isNvidia ||
+            isAntLing),
+    };
+}
+/**
+ * Get resolved compatibility settings for a model.
+ * Auto-detects from provider/URL then overrides with explicit model.compat.
+ */
+export function getOpenAICompletionsCompat(model) {
+    const detected = detectOpenAICompletionsCompat(model);
+    if (!model.compat)
+        return detected;
+    return {
+        supportsStore: model.compat.supportsStore ?? detected.supportsStore,
+        supportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,
+        supportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,
+        supportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? detected.supportsUsageInStreaming,
+        maxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,
+        requiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,
+        requiresAssistantAfterToolResult: model.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,
+        requiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,
+        requiresReasoningContentOnAssistantMessages: model.compat.requiresReasoningContentOnAssistantMessages ??
+            detected.requiresReasoningContentOnAssistantMessages,
+        thinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,
+        supportsDisabledThinking: model.compat.supportsDisabledThinking ?? detected.supportsDisabledThinking,
+        openRouterRouting: model.compat.openRouterRouting ?? detected.openRouterRouting,
+        vercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,
+        chatTemplateKwargs: model.compat.chatTemplateKwargs ?? detected.chatTemplateKwargs,
+        zaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,
+        supportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,
+        toolCallFormat: model.compat.toolCallFormat ?? detected.toolCallFormat,
+        supportsOpenAIGrammarTools: model.compat.supportsOpenAIGrammarTools ?? detected.supportsOpenAIGrammarTools,
+        cacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,
+        sendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,
+        deferredToolsMode: model.compat.deferredToolsMode ?? detected.deferredToolsMode,
+        sessionAffinityFormat: model.compat.sessionAffinityFormat ?? detected.sessionAffinityFormat,
+        supportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,
+    };
+}
+export function getBedrockModelMatchCandidates(modelId, modelName) {
+    const values = modelName ? [modelId, modelName] : [modelId];
+    return values.flatMap((value) => {
+        const lower = value.toLowerCase();
+        return [lower, lower.replace(/[\s_.:]+/g, "-")];
+    });
+}
+/**
+ * Check if the model supports prompt caching.
+ * Supported: Claude 3.5 Haiku, Claude 3.7 Sonnet, Claude 4.x models, Claude 5 models
+ *
+ * For base models and system-defined inference profiles the model ID / ARN
+ * contains the model name, so we can decide locally.
+ *
+ * For application inference profiles (whose ARNs don't contain the model name),
+ * also checks model.name which is user-controlled via models.json or registerProvider.
+ * As a last resort, set AWS_BEDROCK_FORCE_CACHE=1 to enable cache points.
+ * Amazon Nova models have automatic caching and don't need explicit cache points.
+ */
+export function supportsPromptCaching(model, env) {
+    const candidates = getBedrockModelMatchCandidates(model.id, model.name);
+    const hasClaudeRef = candidates.some((s) => s.includes("claude"));
+    if (!hasClaudeRef) {
+        // Application inference profiles don't contain the model name in the ARN.
+        // Allow users to force cache points via environment variable.
+        if (getProviderEnvValue("AWS_BEDROCK_FORCE_CACHE", env) === "1")
+            return true;
+        return false;
+    }
+    // Claude 5 models (fable-5, opus-5, sonnet-5)
+    if (candidates.some((s) => s.includes("fable-5") || s.includes("opus-5") || s.includes("sonnet-5")))
+        return true;
+    // Claude 4.x models (opus-4, sonnet-4, haiku-4)
+    if (candidates.some((s) => s.includes("-4-")))
+        return true;
+    // Claude 3.7 Sonnet
+    if (candidates.some((s) => s.includes("claude-3-7-sonnet")))
+        return true;
+    // Claude 3.5 Haiku
+    if (candidates.some((s) => s.includes("claude-3-5-haiku")))
+        return true;
+    return false;
+}
+function resolveAnthropicCacheRetention(cacheRetention, env, fallback = "short") {
+    if (cacheRetention) {
+        return cacheRetention;
+    }
+    if (getProviderEnvValue("PI_CACHE_RETENTION", env) === "long") {
+        return "long";
+    }
+    if (typeof process !== "undefined" && process.env.PI_CACHE_RETENTION !== undefined) {
+        return "short";
+    }
+    return fallback;
+}
+function resolveOpenAICompletionsCacheRetention(cacheRetention, env) {
+    if (cacheRetention) {
+        return cacheRetention;
+    }
+    if (getProviderEnvValue("PI_CACHE_RETENTION", env) === "long") {
+        return "long";
+    }
+    return "short";
+}
+function resolveBedrockCacheRetention(cacheRetention, env) {
+    if (cacheRetention) {
+        return cacheRetention;
+    }
+    if (getProviderEnvValue("PI_CACHE_RETENTION", env) === "long") {
+        return "long";
+    }
+    return "short";
+}
+function resolveOpenAIResponsesCacheRetention(cacheRetention, env) {
+    if (cacheRetention) {
+        return cacheRetention;
+    }
+    if (getProviderEnvValue("PI_CACHE_RETENTION", env) === "long") {
+        return "long";
+    }
+    return "short";
+}
+export function resolvePromptCacheTtlSeconds(model, env) {
+    switch (model.api) {
+        case "anthropic-messages": {
+            const anthropicModel = model;
+            const retention = resolveAnthropicCacheRetention(anthropicModel.cacheRetention, env, "long");
+            if (retention === "none")
+                return undefined;
+            return retention === "long" &&
+                isAnthropicApiBaseUrl(anthropicModel.baseUrl) &&
+                getAnthropicCompat(anthropicModel).supportsLongCacheRetention
+                ? PROMPT_CACHE_TTL_LONG_SECONDS
+                : PROMPT_CACHE_TTL_SHORT_SECONDS;
+        }
+        case "bedrock-converse-stream": {
+            const bedrockModel = model;
+            const retention = resolveBedrockCacheRetention(bedrockModel.cacheRetention, env);
+            if (retention === "none" || !supportsPromptCaching(bedrockModel, env))
+                return undefined;
+            return retention === "long" ? PROMPT_CACHE_TTL_LONG_SECONDS : PROMPT_CACHE_TTL_SHORT_SECONDS;
+        }
+        case "openai-completions": {
+            const completionsModel = model;
+            const retention = resolveOpenAICompletionsCacheRetention(completionsModel.cacheRetention, env);
+            if (retention === "none")
+                return undefined;
+            const compat = getOpenAICompletionsCompat(completionsModel);
+            if (compat.cacheControlFormat === "anthropic") {
+                return retention === "long" && compat.supportsLongCacheRetention
+                    ? PROMPT_CACHE_TTL_LONG_SECONDS
+                    : PROMPT_CACHE_TTL_SHORT_SECONDS;
+            }
+            return PROMPT_CACHE_TTL_SHORT_SECONDS;
+        }
+        case "openai-responses":
+        case "openai-codex-responses":
+        case "azure-openai-responses": {
+            const retention = resolveOpenAIResponsesCacheRetention(model.cacheRetention, env);
+            return retention === "none" ? undefined : PROMPT_CACHE_TTL_SHORT_SECONDS;
+        }
+        default:
+            return undefined;
+    }
+}
+//# sourceMappingURL=prompt-cache-ttl.js.map

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

@@ -0,0 +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;AAElD,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,MAAM,+BAA+B,GAAG,kCAAkC,CAAC;AAE3E;;;;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;IAElC,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,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,CAAC;QAC9F,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,+BAA+B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAC1F,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,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;AAYD;;;;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;IAElF,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,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC;QAChC,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,IAAI,UAAU,IAAI,qBAAqB,IAAI,UAAU,IAAI,QAAQ,IAAI,SAAS,CAAC;IAE7G,MAAM,MAAM,GAAG,QAAQ,KAAK,KAAK,IAAI,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IAClE,MAAM,UAAU,GAAG,QAAQ,KAAK,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;IAC/E,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,kBAAkB,GAAG,QAAQ,KAAK,YAAY,IAAI,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IAEpH,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,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,KAAK;QACjC,iBAAiB,EAAE,SAAS;QAC5B,qBAAqB,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ;QAC7D,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,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,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,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,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;;;;;;;;;;;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,MAAM,UAAU,4BAA4B,CAAC,KAAiB,EAAE,GAAiB;IAChF,QAAQ,KAAK,CAAC,GAAG,EAAE,CAAC;QACnB,KAAK,oBAAoB,EAAE,CAAC;YAC3B,MAAM,cAAc,GAAG,KAAoC,CAAC;YAC5D,MAAM,SAAS,GAAG,8BAA8B,CAAC,cAAc,CAAC,cAAc,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;YAC7F,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,SAAS,CAAC;YAC3C,OAAO,SAAS,KAAK,MAAM;gBAC1B,qBAAqB,CAAC,cAAc,CAAC,OAAO,CAAC;gBAC7C,kBAAkB,CAAC,cAAc,CAAC,CAAC,0BAA0B;gBAC7D,CAAC,CAAC,6BAA6B;gBAC/B,CAAC,CAAC,8BAA8B,CAAC;QACnC,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,SAAS,CAAC;YACxF,OAAO,SAAS,KAAK,MAAM,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,8BAA8B,CAAC;QAC9F,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,SAAS,CAAC;YAC3C,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,6BAA6B;oBAC/B,CAAC,CAAC,8BAA8B,CAAC;YACnC,CAAC;YACD,OAAO,8BAA8B,CAAC;QACvC,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,OAAO,SAAS,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,8BAA8B,CAAC;QAC1E,CAAC;QACD;YACC,OAAO,SAAS,CAAC;IACnB,CAAC;AACF,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\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\nconst CLAUDE_FABLE_OR_MYTHOS_MODEL_ID = /^claude-(?:fable|mythos)(?:-|$)/i;\n\n/**\n * Default for `supportsToolReferences`: first-party Anthropic models except\n * Haiku (rejects client-side tool_reference blocks) and models that predate\n * tool search (Claude 3.x, Opus/Sonnet 4.0, Opus 4.1).\n */\nfunction defaultSupportsToolReferences(model: Model<\"anthropic-messages\">): boolean {\n\tif (model.provider !== \"anthropic\" || model.id.includes(\"haiku\")) return false;\n\tconst version = model.id.match(/^claude-(?:opus|sonnet|fable)-(\\d+)(?:-(\\d+))?(?:-|$)/);\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<Omit<AnthropicMessagesCompat, \"forceAdaptiveThinking\">> {\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\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),\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 ?? !CLAUDE_FABLE_OR_MYTHOS_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\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\"cacheControlFormat\" | \"toolCallFormat\" | \"deferredToolsMode\" | \"toolSchemaFlavor\"\n> & {\n\tcacheControlFormat?: OpenAICompletionsCompat[\"cacheControlFormat\"];\n\ttoolCallFormat?: OpenAICompletionsCompat[\"toolCallFormat\"];\n\tdeferredToolsMode?: OpenAICompletionsCompat[\"deferredToolsMode\"];\n\ttoolSchemaFlavor?: OpenAICompletionsCompat[\"toolSchemaFlavor\"];\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\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\tbaseUrl.includes(\"deepseek.com\") ||\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\") || isMoonshot || isCloudflareAiGateway || isTogether || isNvidia || isAntLing;\n\n\tconst isGrok = provider === \"xai\" || baseUrl.includes(\"api.x.ai\");\n\tconst isDeepSeek = provider === \"deepseek\" || baseUrl.includes(\"deepseek.com\");\n\tconst isOpenRouterDeveloperRoleModel =\n\t\tisOpenRouter && (model.id.startsWith(\"anthropic/\") || model.id.startsWith(\"openai/\"));\n\tconst cacheControlFormat = provider === \"openrouter\" && model.id.startsWith(\"anthropic/\") ? \"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\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: false,\n\t\tdeferredToolsMode: undefined,\n\t\tsessionAffinityFormat: isOpenRouter ? \"openrouter\" : \"openai\",\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\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\tzaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,\n\t\tsupportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,\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\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\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\nexport function resolvePromptCacheTtlSeconds(model: Model<Api>, env?: ProviderEnv): number | undefined {\n\tswitch (model.api) {\n\t\tcase \"anthropic-messages\": {\n\t\t\tconst anthropicModel = model as Model<\"anthropic-messages\">;\n\t\t\tconst retention = resolveAnthropicCacheRetention(anthropicModel.cacheRetention, env, \"long\");\n\t\t\tif (retention === \"none\") return undefined;\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? PROMPT_CACHE_TTL_LONG_SECONDS\n\t\t\t\t: 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 undefined;\n\t\t\treturn retention === \"long\" ? PROMPT_CACHE_TTL_LONG_SECONDS : 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 undefined;\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? PROMPT_CACHE_TTL_LONG_SECONDS\n\t\t\t\t\t: PROMPT_CACHE_TTL_SHORT_SECONDS;\n\t\t\t}\n\t\t\treturn 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\treturn retention === \"none\" ? undefined : PROMPT_CACHE_TTL_SHORT_SECONDS;\n\t\t}\n\t\tdefault:\n\t\t\treturn undefined;\n\t}\n}\n"]}

package/dist/utils/provider-retry.d.ts

@@ -12,5 +12,10 @@ interface ProviderRetryOptions {
  * disable the limit.
  */
 export declare function retryProviderRequest<T>(request: () => Promise<T>, options?: ProviderRetryOptions): Promise<T>;
+interface ProviderStreamAttempt<TChunk, TMetadata> {
+    stream: AsyncIterable<TChunk>;
+    metadata: TMetadata;
+}
+export declare function retryProviderStreamRequest<TChunk, TMetadata>(request: () => Promise<ProviderStreamAttempt<TChunk, TMetadata>>, options?: ProviderRetryOptions): Promise<ProviderStreamAttempt<TChunk, TMetadata>>;
 export {};
 //# sourceMappingURL=provider-retry.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"provider-retry.d.ts","sourceRoot":"","sources":["../../src/utils/provider-retry.ts"],"names":[],"mappings":"AAEA,UAAU,oBAAoB;IAC7B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB;AA0FD;;;;;;;GAOG;AACH,wBAAsB,oBAAoB,CAAC,CAAC,EAC3C,OAAO,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACzB,OAAO,GAAE,oBAAyB,GAChC,OAAO,CAAC,CAAC,CAAC,CAiBZ"}
+{"version":3,"file":"provider-retry.d.ts","sourceRoot":"","sources":["../../src/utils/provider-retry.ts"],"names":[],"mappings":"AAGA,UAAU,oBAAoB;IAC7B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB;AA4FD;;;;;;;GAOG;AACH,wBAAsB,oBAAoB,CAAC,CAAC,EAC3C,OAAO,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACzB,OAAO,GAAE,oBAAyB,GAChC,OAAO,CAAC,CAAC,CAAC,CAiBZ;AAED,UAAU,qBAAqB,CAAC,MAAM,EAAE,SAAS;IAChD,MAAM,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;IAC9B,QAAQ,EAAE,SAAS,CAAC;CACpB;AAgCD,wBAAsB,0BAA0B,CAAC,MAAM,EAAE,SAAS,EACjE,OAAO,EAAE,MAAM,OAAO,CAAC,qBAAqB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC,EAChE,OAAO,GAAE,oBAAyB,GAChC,OAAO,CAAC,qBAAqB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC,CAUnD"}

package/dist/utils/provider-retry.js

@@ -1,6 +1,11 @@
 const DEFAULT_MAX_RETRY_DELAY_MS = 60_000;
+const DIGITALOCEAN_STREAM_FAILURE_MESSAGE = "Upstream error from DigitalOcean: stream failed";
 function isProviderError(error) {
-    if (!(error instanceof Error) || !("status" in error) || !("headers" in error))
+    if (!(error instanceof Error))
+        return false;
+    if (error.message === DIGITALOCEAN_STREAM_FAILURE_MESSAGE)
+        return true;
+    if (!("status" in error) || !("headers" in error))
         return false;
     return ((error.status === undefined || typeof error.status === "number") &&
         (error.headers === undefined || error.headers instanceof Headers));
@@ -91,4 +96,44 @@ export async function retryProviderRequest(request, options = {}) {
         }
     }
 }
+async function* replayPrefetchedStream(first, iterator) {
+    if (first.done)
+        return;
+    let completed = false;
+    let providerFailed = false;
+    try {
+        yield first.value;
+        for (;;) {
+            let next;
+            try {
+                next = await iterator.next();
+            }
+            catch (error) {
+                providerFailed = true;
+                throw error;
+            }
+            if (next.done) {
+                completed = true;
+                return;
+            }
+            yield next.value;
+        }
+    }
+    finally {
+        if (!completed && !providerFailed) {
+            await iterator.return?.();
+        }
+    }
+}
+export async function retryProviderStreamRequest(request, options = {}) {
+    return retryProviderRequest(async () => {
+        const attempt = await request();
+        const iterator = attempt.stream[Symbol.asyncIterator]();
+        const first = await iterator.next();
+        return {
+            stream: replayPrefetchedStream(first, iterator),
+            metadata: attempt.metadata,
+        };
+    }, options);
+}
 //# sourceMappingURL=provider-retry.js.map

package/dist/utils/provider-retry.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider-retry.js","sourceRoot":"","sources":["../../src/utils/provider-retry.ts"],"names":[],"mappings":"AAAA,MAAM,0BAA0B,GAAG,MAAM,CAAC;AAa1C,SAAS,eAAe,CAAC,KAAc;IACtC,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC,IAAI,CAAC,CAAC,QAAQ,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,SAAS,IAAI,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC7F,OAAO,CACN,CAAC,KAAK,CAAC,MAAM,KAAK,SAAS,IAAI,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;QAChE,CAAC,KAAK,CAAC,OAAO,KAAK,SAAS,IAAI,KAAK,CAAC,OAAO,YAAY,OAAO,CAAC,CACjE,CAAC;AACH,CAAC;AAED,gGAAgG;AAChG,SAAS,wBAAwB,CAAC,KAAoB;IACrD,MAAM,WAAW,GAAG,KAAK,CAAC,OAAO,EAAE,GAAG,CAAC,gBAAgB,CAAC,CAAC;IACzD,IAAI,WAAW,KAAK,MAAM;QAAE,OAAO,IAAI,CAAC;IACxC,IAAI,WAAW,KAAK,OAAO;QAAE,OAAO,KAAK,CAAC;IAE1C,IAAI,KAAK,CAAC,MAAM,KAAK,SAAS;QAAE,OAAO,IAAI,CAAC;IAC5C,OAAO,CACN,KAAK,CAAC,MAAM,KAAK,GAAG;QACpB,KAAK,CAAC,MAAM,KAAK,GAAG;QACpB,KAAK,CAAC,MAAM,KAAK,GAAG;QACpB,CAAC,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,IAAI,GAAG,CAAC,CACzD,CAAC;AACH,CAAC;AAED,SAAS,0BAA0B,CAClC,OAAe,EACf,eAAmC,EACnC,oBAA4B;IAE5B,MAAM,UAAU,GAAG,eAAe,IAAI,0BAA0B,CAAC;IACjE,IAAI,UAAU,GAAG,CAAC,IAAI,OAAO,GAAG,UAAU,EAAE,CAAC;QAC5C,MAAM,IAAI,KAAK,CACd,oBAAoB,IAAI,CAAC,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,uBAAuB,IAAI,CAAC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,OAAO,oBAAoB,EAAE,CAC7H,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,eAAe,CAAC,KAAoB,EAAE,UAAkB,EAAE,eAAmC;IACrG,MAAM,YAAY,GAAG,KAAK,CAAC,OAAO,EAAE,GAAG,CAAC,gBAAgB,CAAC,CAAC;IAC1D,IAAI,YAAY,EAAE,CAAC;QAClB,MAAM,KAAK,GAAG,MAAM,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC;QAC9C,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC;YAAE,OAAO,0BAA0B,CAAC,KAAK,EAAE,eAAe,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IACpG,CAAC;IAED,MAAM,UAAU,GAAG,KAAK,CAAC,OAAO,EAAE,GAAG,CAAC,aAAa,CAAC,CAAC;IACrD,IAAI,UAAU,EAAE,CAAC;QAChB,MAAM,OAAO,GAAG,MAAM,CAAC,UAAU,CAAC,UAAU,CAAC,CAAC;QAC9C,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,OAAO,GAAG,IAAI,CAAC;QAC7F,OAAO,0BAA0B,CAAC,OAAO,EAAE,eAAe,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IAC5E,CAAC;IAED,MAAM,gBAAgB,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC;IACnE,OAAO,gBAAgB,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC;AACtD,CAAC;AAED,SAAS,gBAAgB;IACxB,MAAM,KAAK,GAAG,IAAI,KAAK,CAAC,iBAAiB,CAAC,CAAC;IAC3C,KAAK,CAAC,IAAI,GAAG,YAAY,CAAC;IAC1B,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,cAAc,CAAC,EAAU,EAAE,MAAoB;IACvD,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACtC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,MAAM,CAAC,gBAAgB,EAAE,CAAC,CAAC;YAC3B,OAAO;QACR,CAAC;QAED,MAAM,OAAO,GAAG,GAAG,EAAE;YACpB,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,gBAAgB,EAAE,CAAC,CAAC;QAC5B,CAAC,CAAC;QACF,MAAM,OAAO,GAAG,UAAU,CACzB,GAAG,EAAE;YACJ,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,OAAO,EAAE,CAAC;QACX,CAAC,EACD,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CACf,CAAC;QACF,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IAC5D,CAAC,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,CAAC,KAAK,UAAU,oBAAoB,CACzC,OAAyB,EACzB,OAAO,GAAyB,EAAE;IAElC,MAAM,UAAU,GAAG,OAAO,CAAC,UAAU,IAAI,CAAC,CAAC;IAC3C,IAAI,gBAAgB,GAAG,UAAU,CAAC;IAElC,SAAS,CAAC;QACT,IAAI,CAAC;YACJ,8EAA8E;YAC9E,OAAO,MAAM,OAAO,EAAE,CAAC;QACxB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO;gBAAE,MAAM,gBAAgB,EAAE,CAAC;YACtD,IAAI,gBAAgB,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,KAAK,CAAC,IAAI,CAAC,wBAAwB,CAAC,KAAK,CAAC;gBAAE,MAAM,KAAK,CAAC;YAEtG,MAAM,UAAU,GAAG,UAAU,GAAG,gBAAgB,CAAC;YACjD,gBAAgB,EAAE,CAAC;YACnB,MAAM,cAAc,CAAC,eAAe,CAAC,KAAK,EAAE,UAAU,EAAE,OAAO,CAAC,eAAe,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;QACnG,CAAC;IACF,CAAC;AACF,CAAC","sourcesContent":["const DEFAULT_MAX_RETRY_DELAY_MS = 60_000;\n\ninterface ProviderRetryOptions {\n\tmaxRetries?: number;\n\tmaxRetryDelayMs?: number;\n\tsignal?: AbortSignal;\n}\n\ninterface ProviderError extends Error {\n\tstatus: number | undefined;\n\theaders: Headers | undefined;\n}\n\nfunction isProviderError(error: unknown): error is ProviderError {\n\tif (!(error instanceof Error) || !(\"status\" in error) || !(\"headers\" in error)) return false;\n\treturn (\n\t\t(error.status === undefined || typeof error.status === \"number\") &&\n\t\t(error.headers === undefined || error.headers instanceof Headers)\n\t);\n}\n\n/** Mirrors the pinned OpenAI/Anthropic SDK retry policy; review when either SDK is upgraded. */\nfunction isRetryableProviderError(error: ProviderError): boolean {\n\tconst shouldRetry = error.headers?.get(\"x-should-retry\");\n\tif (shouldRetry === \"true\") return true;\n\tif (shouldRetry === \"false\") return false;\n\n\tif (error.status === undefined) return true;\n\treturn (\n\t\terror.status === 408 ||\n\t\terror.status === 409 ||\n\t\terror.status === 429 ||\n\t\t(typeof error.status === \"number\" && error.status >= 500)\n\t);\n}\n\nfunction validateServerRetryDelayMs(\n\tdelayMs: number,\n\tmaxRetryDelayMs: number | undefined,\n\tproviderErrorMessage: string,\n): number {\n\tconst maxDelayMs = maxRetryDelayMs ?? DEFAULT_MAX_RETRY_DELAY_MS;\n\tif (maxDelayMs > 0 && delayMs > maxDelayMs) {\n\t\tthrow new Error(\n\t\t\t`Server requested ${Math.ceil(delayMs / 1000)}s retry delay (max: ${Math.ceil(maxDelayMs / 1000)}s). ${providerErrorMessage}`,\n\t\t);\n\t}\n\treturn delayMs;\n}\n\nfunction getRetryDelayMs(error: ProviderError, retryIndex: number, maxRetryDelayMs: number | undefined): number {\n\tconst retryAfterMs = error.headers?.get(\"retry-after-ms\");\n\tif (retryAfterMs) {\n\t\tconst value = Number.parseFloat(retryAfterMs);\n\t\tif (!Number.isNaN(value)) return validateServerRetryDelayMs(value, maxRetryDelayMs, error.message);\n\t}\n\n\tconst retryAfter = error.headers?.get(\"retry-after\");\n\tif (retryAfter) {\n\t\tconst seconds = Number.parseFloat(retryAfter);\n\t\tconst delayMs = Number.isNaN(seconds) ? Date.parse(retryAfter) - Date.now() : seconds * 1000;\n\t\treturn validateServerRetryDelayMs(delayMs, maxRetryDelayMs, error.message);\n\t}\n\n\tconst exponentialDelay = Math.min(0.5 * 2 ** retryIndex, 8) * 1000;\n\treturn exponentialDelay * (1 - Math.random() * 0.25);\n}\n\nfunction createAbortError(): Error {\n\tconst error = new Error(\"Request aborted\");\n\terror.name = \"AbortError\";\n\treturn error;\n}\n\nfunction abortableSleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(createAbortError());\n\t\t\treturn;\n\t\t}\n\n\t\tconst onAbort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(createAbortError());\n\t\t};\n\t\tconst timeout = setTimeout(\n\t\t\t() => {\n\t\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\t\tresolve();\n\t\t\t},\n\t\t\tMath.max(0, ms),\n\t\t);\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t});\n}\n\n/**\n * Reproduce the retry behavior used by the OpenAI and Anthropic SDKs while making\n * their backoff sleep interruptible. Their built-in retry timers ignore the\n * request AbortSignal, so callers must invoke the SDK with `maxRetries: 0` and\n * wrap the request with this helper. Provider-requested delays above\n * `maxRetryDelayMs` fail immediately (60 seconds by default); set it to zero to\n * disable the limit.\n */\nexport async function retryProviderRequest<T>(\n\trequest: () => Promise<T>,\n\toptions: ProviderRetryOptions = {},\n): Promise<T> {\n\tconst maxRetries = options.maxRetries ?? 0;\n\tlet retriesRemaining = maxRetries;\n\n\tfor (;;) {\n\t\ttry {\n\t\t\t// Each retry is a fresh SDK request, so X-Stainless-Retry-Count remains zero.\n\t\t\treturn await request();\n\t\t} catch (error) {\n\t\t\tif (options.signal?.aborted) throw createAbortError();\n\t\t\tif (retriesRemaining <= 0 || !isProviderError(error) || !isRetryableProviderError(error)) throw error;\n\n\t\t\tconst retryIndex = maxRetries - retriesRemaining;\n\t\t\tretriesRemaining--;\n\t\t\tawait abortableSleep(getRetryDelayMs(error, retryIndex, options.maxRetryDelayMs), options.signal);\n\t\t}\n\t}\n}\n"]}
+{"version":3,"file":"provider-retry.js","sourceRoot":"","sources":["../../src/utils/provider-retry.ts"],"names":[],"mappings":"AAAA,MAAM,0BAA0B,GAAG,MAAM,CAAC;AAC1C,MAAM,mCAAmC,GAAG,iDAAiD,CAAC;AAa9F,SAAS,eAAe,CAAC,KAAc;IACtC,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5C,IAAI,KAAK,CAAC,OAAO,KAAK,mCAAmC;QAAE,OAAO,IAAI,CAAC;IACvE,IAAI,CAAC,CAAC,QAAQ,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,SAAS,IAAI,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAChE,OAAO,CACN,CAAC,KAAK,CAAC,MAAM,KAAK,SAAS,IAAI,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;QAChE,CAAC,KAAK,CAAC,OAAO,KAAK,SAAS,IAAI,KAAK,CAAC,OAAO,YAAY,OAAO,CAAC,CACjE,CAAC;AACH,CAAC;AAED,gGAAgG;AAChG,SAAS,wBAAwB,CAAC,KAAoB;IACrD,MAAM,WAAW,GAAG,KAAK,CAAC,OAAO,EAAE,GAAG,CAAC,gBAAgB,CAAC,CAAC;IACzD,IAAI,WAAW,KAAK,MAAM;QAAE,OAAO,IAAI,CAAC;IACxC,IAAI,WAAW,KAAK,OAAO;QAAE,OAAO,KAAK,CAAC;IAE1C,IAAI,KAAK,CAAC,MAAM,KAAK,SAAS;QAAE,OAAO,IAAI,CAAC;IAC5C,OAAO,CACN,KAAK,CAAC,MAAM,KAAK,GAAG;QACpB,KAAK,CAAC,MAAM,KAAK,GAAG;QACpB,KAAK,CAAC,MAAM,KAAK,GAAG;QACpB,CAAC,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,IAAI,GAAG,CAAC,CACzD,CAAC;AACH,CAAC;AAED,SAAS,0BAA0B,CAClC,OAAe,EACf,eAAmC,EACnC,oBAA4B;IAE5B,MAAM,UAAU,GAAG,eAAe,IAAI,0BAA0B,CAAC;IACjE,IAAI,UAAU,GAAG,CAAC,IAAI,OAAO,GAAG,UAAU,EAAE,CAAC;QAC5C,MAAM,IAAI,KAAK,CACd,oBAAoB,IAAI,CAAC,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,uBAAuB,IAAI,CAAC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,OAAO,oBAAoB,EAAE,CAC7H,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,eAAe,CAAC,KAAoB,EAAE,UAAkB,EAAE,eAAmC;IACrG,MAAM,YAAY,GAAG,KAAK,CAAC,OAAO,EAAE,GAAG,CAAC,gBAAgB,CAAC,CAAC;IAC1D,IAAI,YAAY,EAAE,CAAC;QAClB,MAAM,KAAK,GAAG,MAAM,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC;QAC9C,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC;YAAE,OAAO,0BAA0B,CAAC,KAAK,EAAE,eAAe,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IACpG,CAAC;IAED,MAAM,UAAU,GAAG,KAAK,CAAC,OAAO,EAAE,GAAG,CAAC,aAAa,CAAC,CAAC;IACrD,IAAI,UAAU,EAAE,CAAC;QAChB,MAAM,OAAO,GAAG,MAAM,CAAC,UAAU,CAAC,UAAU,CAAC,CAAC;QAC9C,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,OAAO,GAAG,IAAI,CAAC;QAC7F,OAAO,0BAA0B,CAAC,OAAO,EAAE,eAAe,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IAC5E,CAAC;IAED,MAAM,gBAAgB,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC;IACnE,OAAO,gBAAgB,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC;AACtD,CAAC;AAED,SAAS,gBAAgB;IACxB,MAAM,KAAK,GAAG,IAAI,KAAK,CAAC,iBAAiB,CAAC,CAAC;IAC3C,KAAK,CAAC,IAAI,GAAG,YAAY,CAAC;IAC1B,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,cAAc,CAAC,EAAU,EAAE,MAAoB;IACvD,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACtC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,MAAM,CAAC,gBAAgB,EAAE,CAAC,CAAC;YAC3B,OAAO;QACR,CAAC;QAED,MAAM,OAAO,GAAG,GAAG,EAAE;YACpB,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,gBAAgB,EAAE,CAAC,CAAC;QAC5B,CAAC,CAAC;QACF,MAAM,OAAO,GAAG,UAAU,CACzB,GAAG,EAAE;YACJ,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,OAAO,EAAE,CAAC;QACX,CAAC,EACD,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CACf,CAAC;QACF,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IAC5D,CAAC,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,CAAC,KAAK,UAAU,oBAAoB,CACzC,OAAyB,EACzB,OAAO,GAAyB,EAAE;IAElC,MAAM,UAAU,GAAG,OAAO,CAAC,UAAU,IAAI,CAAC,CAAC;IAC3C,IAAI,gBAAgB,GAAG,UAAU,CAAC;IAElC,SAAS,CAAC;QACT,IAAI,CAAC;YACJ,8EAA8E;YAC9E,OAAO,MAAM,OAAO,EAAE,CAAC;QACxB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO;gBAAE,MAAM,gBAAgB,EAAE,CAAC;YACtD,IAAI,gBAAgB,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,KAAK,CAAC,IAAI,CAAC,wBAAwB,CAAC,KAAK,CAAC;gBAAE,MAAM,KAAK,CAAC;YAEtG,MAAM,UAAU,GAAG,UAAU,GAAG,gBAAgB,CAAC;YACjD,gBAAgB,EAAE,CAAC;YACnB,MAAM,cAAc,CAAC,eAAe,CAAC,KAAK,EAAE,UAAU,EAAE,OAAO,CAAC,eAAe,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;QACnG,CAAC;IACF,CAAC;AACF,CAAC;AAOD,KAAK,SAAS,CAAC,CAAC,sBAAsB,CACrC,KAA6B,EAC7B,QAA+B;IAE/B,IAAI,KAAK,CAAC,IAAI;QAAE,OAAO;IACvB,IAAI,SAAS,GAAG,KAAK,CAAC;IACtB,IAAI,cAAc,GAAG,KAAK,CAAC;IAC3B,IAAI,CAAC;QACJ,MAAM,KAAK,CAAC,KAAK,CAAC;QAClB,SAAS,CAAC;YACT,IAAI,IAA4B,CAAC;YACjC,IAAI,CAAC;gBACJ,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;YAC9B,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBAChB,cAAc,GAAG,IAAI,CAAC;gBACtB,MAAM,KAAK,CAAC;YACb,CAAC;YACD,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;gBACf,SAAS,GAAG,IAAI,CAAC;gBACjB,OAAO;YACR,CAAC;YACD,MAAM,IAAI,CAAC,KAAK,CAAC;QAClB,CAAC;IACF,CAAC;YAAS,CAAC;QACV,IAAI,CAAC,SAAS,IAAI,CAAC,cAAc,EAAE,CAAC;YACnC,MAAM,QAAQ,CAAC,MAAM,EAAE,EAAE,CAAC;QAC3B,CAAC;IACF,CAAC;AACF,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAC/C,OAAgE,EAChE,OAAO,GAAyB,EAAE;IAElC,OAAO,oBAAoB,CAAC,KAAK,IAAI,EAAE;QACtC,MAAM,OAAO,GAAG,MAAM,OAAO,EAAE,CAAC;QAChC,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,EAAE,CAAC;QACxD,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACpC,OAAO;YACN,MAAM,EAAE,sBAAsB,CAAC,KAAK,EAAE,QAAQ,CAAC;YAC/C,QAAQ,EAAE,OAAO,CAAC,QAAQ;SAC1B,CAAC;IACH,CAAC,EAAE,OAAO,CAAC,CAAC;AACb,CAAC","sourcesContent":["const DEFAULT_MAX_RETRY_DELAY_MS = 60_000;\nconst DIGITALOCEAN_STREAM_FAILURE_MESSAGE = \"Upstream error from DigitalOcean: stream failed\";\n\ninterface ProviderRetryOptions {\n\tmaxRetries?: number;\n\tmaxRetryDelayMs?: number;\n\tsignal?: AbortSignal;\n}\n\ninterface ProviderError extends Error {\n\tstatus: number | undefined;\n\theaders: Headers | undefined;\n}\n\nfunction isProviderError(error: unknown): error is ProviderError {\n\tif (!(error instanceof Error)) return false;\n\tif (error.message === DIGITALOCEAN_STREAM_FAILURE_MESSAGE) return true;\n\tif (!(\"status\" in error) || !(\"headers\" in error)) return false;\n\treturn (\n\t\t(error.status === undefined || typeof error.status === \"number\") &&\n\t\t(error.headers === undefined || error.headers instanceof Headers)\n\t);\n}\n\n/** Mirrors the pinned OpenAI/Anthropic SDK retry policy; review when either SDK is upgraded. */\nfunction isRetryableProviderError(error: ProviderError): boolean {\n\tconst shouldRetry = error.headers?.get(\"x-should-retry\");\n\tif (shouldRetry === \"true\") return true;\n\tif (shouldRetry === \"false\") return false;\n\n\tif (error.status === undefined) return true;\n\treturn (\n\t\terror.status === 408 ||\n\t\terror.status === 409 ||\n\t\terror.status === 429 ||\n\t\t(typeof error.status === \"number\" && error.status >= 500)\n\t);\n}\n\nfunction validateServerRetryDelayMs(\n\tdelayMs: number,\n\tmaxRetryDelayMs: number | undefined,\n\tproviderErrorMessage: string,\n): number {\n\tconst maxDelayMs = maxRetryDelayMs ?? DEFAULT_MAX_RETRY_DELAY_MS;\n\tif (maxDelayMs > 0 && delayMs > maxDelayMs) {\n\t\tthrow new Error(\n\t\t\t`Server requested ${Math.ceil(delayMs / 1000)}s retry delay (max: ${Math.ceil(maxDelayMs / 1000)}s). ${providerErrorMessage}`,\n\t\t);\n\t}\n\treturn delayMs;\n}\n\nfunction getRetryDelayMs(error: ProviderError, retryIndex: number, maxRetryDelayMs: number | undefined): number {\n\tconst retryAfterMs = error.headers?.get(\"retry-after-ms\");\n\tif (retryAfterMs) {\n\t\tconst value = Number.parseFloat(retryAfterMs);\n\t\tif (!Number.isNaN(value)) return validateServerRetryDelayMs(value, maxRetryDelayMs, error.message);\n\t}\n\n\tconst retryAfter = error.headers?.get(\"retry-after\");\n\tif (retryAfter) {\n\t\tconst seconds = Number.parseFloat(retryAfter);\n\t\tconst delayMs = Number.isNaN(seconds) ? Date.parse(retryAfter) - Date.now() : seconds * 1000;\n\t\treturn validateServerRetryDelayMs(delayMs, maxRetryDelayMs, error.message);\n\t}\n\n\tconst exponentialDelay = Math.min(0.5 * 2 ** retryIndex, 8) * 1000;\n\treturn exponentialDelay * (1 - Math.random() * 0.25);\n}\n\nfunction createAbortError(): Error {\n\tconst error = new Error(\"Request aborted\");\n\terror.name = \"AbortError\";\n\treturn error;\n}\n\nfunction abortableSleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(createAbortError());\n\t\t\treturn;\n\t\t}\n\n\t\tconst onAbort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(createAbortError());\n\t\t};\n\t\tconst timeout = setTimeout(\n\t\t\t() => {\n\t\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\t\tresolve();\n\t\t\t},\n\t\t\tMath.max(0, ms),\n\t\t);\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t});\n}\n\n/**\n * Reproduce the retry behavior used by the OpenAI and Anthropic SDKs while making\n * their backoff sleep interruptible. Their built-in retry timers ignore the\n * request AbortSignal, so callers must invoke the SDK with `maxRetries: 0` and\n * wrap the request with this helper. Provider-requested delays above\n * `maxRetryDelayMs` fail immediately (60 seconds by default); set it to zero to\n * disable the limit.\n */\nexport async function retryProviderRequest<T>(\n\trequest: () => Promise<T>,\n\toptions: ProviderRetryOptions = {},\n): Promise<T> {\n\tconst maxRetries = options.maxRetries ?? 0;\n\tlet retriesRemaining = maxRetries;\n\n\tfor (;;) {\n\t\ttry {\n\t\t\t// Each retry is a fresh SDK request, so X-Stainless-Retry-Count remains zero.\n\t\t\treturn await request();\n\t\t} catch (error) {\n\t\t\tif (options.signal?.aborted) throw createAbortError();\n\t\t\tif (retriesRemaining <= 0 || !isProviderError(error) || !isRetryableProviderError(error)) throw error;\n\n\t\t\tconst retryIndex = maxRetries - retriesRemaining;\n\t\t\tretriesRemaining--;\n\t\t\tawait abortableSleep(getRetryDelayMs(error, retryIndex, options.maxRetryDelayMs), options.signal);\n\t\t}\n\t}\n}\n\ninterface ProviderStreamAttempt<TChunk, TMetadata> {\n\tstream: AsyncIterable<TChunk>;\n\tmetadata: TMetadata;\n}\n\nasync function* replayPrefetchedStream<TChunk>(\n\tfirst: IteratorResult<TChunk>,\n\titerator: AsyncIterator<TChunk>,\n): AsyncGenerator<TChunk> {\n\tif (first.done) return;\n\tlet completed = false;\n\tlet providerFailed = false;\n\ttry {\n\t\tyield first.value;\n\t\tfor (;;) {\n\t\t\tlet next: IteratorResult<TChunk>;\n\t\t\ttry {\n\t\t\t\tnext = await iterator.next();\n\t\t\t} catch (error) {\n\t\t\t\tproviderFailed = true;\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tif (next.done) {\n\t\t\t\tcompleted = true;\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tyield next.value;\n\t\t}\n\t} finally {\n\t\tif (!completed && !providerFailed) {\n\t\t\tawait iterator.return?.();\n\t\t}\n\t}\n}\n\nexport async function retryProviderStreamRequest<TChunk, TMetadata>(\n\trequest: () => Promise<ProviderStreamAttempt<TChunk, TMetadata>>,\n\toptions: ProviderRetryOptions = {},\n): Promise<ProviderStreamAttempt<TChunk, TMetadata>> {\n\treturn retryProviderRequest(async () => {\n\t\tconst attempt = await request();\n\t\tconst iterator = attempt.stream[Symbol.asyncIterator]();\n\t\tconst first = await iterator.next();\n\t\treturn {\n\t\t\tstream: replayPrefetchedStream(first, iterator),\n\t\t\tmetadata: attempt.metadata,\n\t\t};\n\t}, options);\n}\n"]}

package/dist/utils/retry.d.ts

@@ -50,4 +50,11 @@ export declare function retryAssistantCall(produce: () => Promise<AssistantMessa
  * before restarting the assistant turn.
  */
 export declare function isRetryableAssistantError(message: AssistantMessage): boolean;
+/**
+ * Classifies a raw error-message string with the same transient-vs-terminal
+ * rules as {@link isRetryableAssistantError}, for callers that hold a thrown
+ * `Error` instead of an `AssistantMessage` (e.g. compaction summarization
+ * failures that must decide between degrading and surfacing loudly).
+ */
+export declare function isRetryableErrorMessage(errorMessage: string): boolean;
 //# sourceMappingURL=retry.d.ts.map

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;AAqGpD;;;;;GAKG;AACH,MAAM,WAAW,WAAW;IAC3B,OAAO,EAAE,OAAO,CAAC;IACjB,qFAAqF;IACrF,UAAU,EAAE,MAAM,CAAC;IACnB,0FAA0F;IAC1F,WAAW,EAAE,MAAM,CAAC;CACpB;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;;;;;;;;;;;;;;;;;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,CA4C3B;AAED;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAY5E"}
+{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAwGpD;;;;;GAKG;AACH,MAAM,WAAW,WAAW;IAC3B,OAAO,EAAE,OAAO,CAAC;IACjB,qFAAqF;IACrF,UAAU,EAAE,MAAM,CAAC;IACnB,0FAA0F;IAC1F,WAAW,EAAE,MAAM,CAAC;CACpB;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;;;;;;;;;;;;;;;;;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,CA4C3B;AAED;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAU5E;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,YAAY,EAAE,MAAM,GAAG,OAAO,CAIrE"}

package/dist/utils/retry.js

@@ -27,6 +27,9 @@ const RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([
     "502",
     "503",
     "504",
+    // Cloudflare 522 (Connection timed out): origin stopped responding; transient
+    // like the other 5xx gateway statuses, surfaced as "Error: error code: 522".
+    "522",
     "524",
     "service.?unavailable",
     "server.?error",
@@ -179,7 +182,17 @@ export function isRetryableAssistantError(message) {
         !message.errorMessage) {
         return false;
     }
-    const errorMessage = message.errorMessage;
+    return isRetryableErrorMessage(message.errorMessage);
+}
+/**
+ * Classifies a raw error-message string with the same transient-vs-terminal
+ * rules as {@link isRetryableAssistantError}, for callers that hold a thrown
+ * `Error` instead of an `AssistantMessage` (e.g. compaction summarization
+ * failures that must decide between degrading and surfacing loudly).
+ */
+export function isRetryableErrorMessage(errorMessage) {
+    if (!errorMessage)
+        return false;
     if (NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN.test(errorMessage))
         return false;
     return RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage);

package/dist/utils/retry.js.map

@@ -1 +1 @@
-{"version":3,"file":"retry.js","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAEA,SAAS,yBAAyB,CAAC,QAA2B;IAC7D,OAAO,IAAI,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,CAAC;AAC5C,CAAC;AAED,MAAM,0CAA0C,GAAG,yBAAyB,CAAC;IAC5E,8EAA8E;IAC9E,2EAA2E;IAC3E,mBAAmB;IACnB,qBAAqB;IAErB,6EAA6E;IAC7E,yDAAyD;IACzD,6BAA6B;IAC7B,mBAAmB;IAEnB,4EAA4E;IAC5E,4EAA4E;IAC5E,oBAAoB;IACpB,eAAe;IACf,gBAAgB;IAChB,SAAS;CACT,CAAC,CAAC;AAEH,MAAM,gCAAgC,GAAG,yBAAyB,CAAC;IAClE,0EAA0E;IAC1E,YAAY;IACZ,aAAa;IACb,mBAAmB;IACnB,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,sBAAsB;IACtB,eAAe;IACf,iBAAiB;IAEjB,8EAA8E;IAC9E,+CAA+C;IAC/C,2BAA2B;IAE3B,2EAA2E;IAC3E,2EAA2E;IAC3E,2EAA2E;IAC3E,gBAAgB;IAChB,mBAAmB;IACnB,qBAAqB;IACrB,kBAAkB;IAClB,mBAAmB;IACnB,cAAc;IACd,aAAa;IACb,WAAW;IACX,WAAW;IACX,mBAAmB;IACnB,uBAAuB;IACvB,sBAAsB;IACtB,gBAAgB;IAChB,8BAA8B;IAC9B,YAAY;IACZ,SAAS;IACT,YAAY;IAEZ,+EAA+E;IAC/E,mBAAmB;IACnB,kBAAkB;IAElB,yEAAyE;IACzE,8EAA8E;IAC9E,yEAAyE;IACzE,eAAe;IACf,kCAAkC;IAClC,+CAA+C;IAC/C,sCAAsC;IAEtC,4EAA4E;IAC5E,iEAAiE;IACjE,aAAa;IAEb,6EAA6E;IAC7E,6BAA6B;IAC7B,4BAA4B;IAC5B,wBAAwB;IACxB,2BAA2B;IAE3B,gFAAgF;IAChF,4EAA4E;IAC5E,4EAA4E;IAC5E,6EAA6E;IAC7E,+EAA+E;IAC/E,gFAAgF;IAChF,gFAAgF;IAChF,oEAAoE;IACpE,qCAAqC;IAErC,yCAAyC;IACzC,mBAAmB;CACnB,CAAC,CAAC;AA+BH,MAAM,oBAAqB,SAAQ,KAAK;IACvC;QACC,KAAK,CAAC,SAAS,CAAC,CAAC;IAClB,CAAC;CACD;AAED,SAAS,KAAK,CAAC,EAAU,EAAE,MAAoB;IAC9C,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACtC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,MAAM,CAAC,IAAI,oBAAoB,EAAE,CAAC,CAAC;YACnC,OAAO;QACR,CAAC;QACD,MAAM,OAAO,GAAG,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACxC,MAAM,EAAE,gBAAgB,CACvB,OAAO,EACP,GAAG,EAAE;YACJ,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,IAAI,oBAAoB,EAAE,CAAC,CAAC;QACpC,CAAC,EACD,EAAE,IAAI,EAAE,IAAI,EAAE,CACd,CAAC;IACH,CAAC,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACvC,OAAwC,EACxC,MAA+B,EAC/B,MAA+B,EAC/B,SAA0B;IAE1B,MAAM,WAAW,GAAG,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAE5D,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,SAAgE,CAAC;IACrE,SAAS,CAAC;QACT,MAAM,QAAQ,GAAG,MAAM,OAAO,EAAE,CAAC;QAEjC,sEAAsE;QACtE,IAAI,QAAQ,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;YACvC,IAAI,SAAS;gBAAE,MAAM,SAAS,EAAE,eAAe,EAAE,CAAC,KAAK,EAAE,SAAS,CAAC,OAAO,CAAC,CAAC;YAC5E,OAAO,QAAQ,CAAC;QACjB,CAAC;QAED,wDAAwD;QACxD,IAAI,QAAQ,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;YACrC,IAAI,SAAS;gBAAE,MAAM,SAAS,EAAE,eAAe,EAAE,CAAC,IAAI,EAAE,SAAS,CAAC,OAAO,CAAC,CAAC;YAC3E,OAAO,QAAQ,CAAC;QACjB,CAAC;QAED,sEAAsE;QACtE,IAAI,OAAO,IAAI,WAAW,IAAI,CAAC,yBAAyB,CAAC,QAAQ,CAAC,EAAE,CAAC;YACpE,IAAI,SAAS;gBAAE,MAAM,SAAS,EAAE,eAAe,EAAE,CAAC,KAAK,EAAE,SAAS,CAAC,OAAO,EAAE,QAAQ,CAAC,YAAY,CAAC,CAAC;YACnG,OAAO,QAAQ,CAAC;QACjB,CAAC;QAED,OAAO,EAAE,CAAC;QACV,SAAS,GAAG,EAAE,OAAO,EAAE,YAAY,EAAE,QAAQ,CAAC,YAAY,IAAI,eAAe,EAAE,CAAC;QAChF,MAAM,OAAO,GAAG,MAAO,CAAC,WAAW,GAAG,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC;QACzD,MAAM,SAAS,EAAE,gBAAgB,EAAE,CAAC,OAAO,EAAE,WAAW,EAAE,OAAO,EAAE,SAAS,CAAC,YAAY,CAAC,CAAC;QAE3F,8EAA8E;QAC9E,qFAAqF;QACrF,IAAI,CAAC;YACJ,MAAM,KAAK,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,MAAM,SAAS,EAAE,eAAe,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,SAAS,CAAC,YAAY,CAAC,CAAC;YAC3E,IAAI,KAAK,YAAY,oBAAoB,EAAE,CAAC;gBAC3C,OAAO,EAAE,GAAG,QAAQ,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,SAAS,EAAE,CAAC;YACxE,CAAC;YACD,MAAM,KAAK,CAAC;QACb,CAAC;QACD,MAAM,SAAS,EAAE,mBAAmB,EAAE,EAAE,CAAC;IAC1C,CAAC;AACF,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,yBAAyB,CAAC,OAAyB;IAClE,IACC,OAAO,CAAC,UAAU,KAAK,OAAO;QAC9B,OAAO,CAAC,WAAW,EAAE,IAAI,KAAK,SAAS;QACvC,OAAO,CAAC,WAAW,EAAE,IAAI,KAAK,WAAW;QACzC,CAAC,OAAO,CAAC,YAAY,EACpB,CAAC;QACF,OAAO,KAAK,CAAC;IACd,CAAC;IACD,MAAM,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;IAC1C,IAAI,0CAA0C,CAAC,IAAI,CAAC,YAAY,CAAC;QAAE,OAAO,KAAK,CAAC;IAChF,OAAO,gCAAgC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;AAC5D,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\nfunction buildProviderErrorPattern(patterns: readonly string[]): RegExp {\n\treturn new RegExp(patterns.join(\"|\"), \"i\");\n}\n\nconst NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN = buildProviderErrorPattern([\n\t// OpenCode Go/free-tier limits returned as 429 JSON error types by OpenCode's\n\t// Zen API. These are subscription/account limits, not transient throttles.\n\t\"GoUsageLimitError\",\n\t\"FreeUsageLimitError\",\n\n\t// OpenCode Go subscription-limit text asks users to enable available-balance\n\t// usage after rolling/weekly/monthly limits are reached.\n\t\"Monthly usage limit reached\",\n\t\"available balance\",\n\n\t// Generic quota/budget/billing exhaustion. `insufficient_quota` is OpenAI's\n\t// quota/billing error code; the other strings cover common gateway wording.\n\t\"insufficient_quota\",\n\t\"out of budget\",\n\t\"quota exceeded\",\n\t\"billing\",\n]);\n\nconst RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([\n\t// Generic provider load, HTTP status, and server-side transient failures.\n\t\"overloaded\",\n\t\"rate.?limit\",\n\t\"too many requests\",\n\t\"429\",\n\t\"500\",\n\t\"502\",\n\t\"503\",\n\t\"504\",\n\t\"524\",\n\t\"service.?unavailable\",\n\t\"server.?error\",\n\t\"internal.?error\",\n\n\t// Wrapper/provider text for transient upstream failures, including OpenRouter\n\t// \"Provider returned error\" responses (#2264).\n\t\"provider.?returned.?error\",\n\n\t// Network, proxy, and fetch transport failures. This includes OpenAI Codex\n\t// raw-fetch failures such as \"upstream connect\", \"connection refused\", and\n\t// \"reset before headers\" (#733), plus OpenRouter connection drops (#3317).\n\t\"network.?error\",\n\t\"connection.?error\",\n\t\"connection.?refused\",\n\t\"connection.?lost\",\n\t\"other side closed\",\n\t\"fetch failed\",\n\t\"getaddrinfo\",\n\t\"ENOTFOUND\",\n\t\"EAI_AGAIN\",\n\t\"upstream.?connect\",\n\t\"upstream.?unavailable\",\n\t\"reset before headers\",\n\t\"socket hang up\",\n\t\"socket connection was closed\",\n\t\"timed? out\",\n\t\"timeout\",\n\t\"terminated\",\n\n\t// WebSocket transports can report close/error text instead of HTTP/fetch text.\n\t\"websocket.?closed\",\n\t\"websocket.?error\",\n\n\t// Premature stream endings from SDKs and transports. Anthropic can throw\n\t// \"stream ended without ...\" and \"Anthropic stream ended before message_stop\"\n\t// (#4433); Bedrock/Smithy can throw an HTTP/2 no-response error (#3594).\n\t\"ended without\",\n\t\"stream ended before message_stop\",\n\t\"stream ended before a terminal response event\",\n\t\"http2 request did not get a response\",\n\n\t// Provider-requested retry delay cap failures should flow through the outer\n\t// retry policy so callers can surface/abort the backoff (#1123).\n\t\"retry delay\",\n\n\t// Explicit retry guidance emitted mid-stream by OpenAI Responses and Bedrock\n\t// stream exceptions (#6019).\n\t\"you can retry your request\",\n\t\"try your request again\",\n\t\"please retry your request\",\n\n\t// Anthropic server-tool pairing rejections, e.g. \"`web_search` tool use with id\n\t// `srvtoolu_...` was found without a corresponding `web_search_tool_result`\n\t// block\". A turn closed before its deferred server tool could answer leaves\n\t// the unpairable half in history, and replaying it 400s every later request.\n\t// The Anthropic request builder repairs the replayed history, so a retry sends\n\t// a valid payload; keeping the class retryable is what lets retry and the model\n\t// fallback chain unwedge such a session instead of dead-ending it. The trailing\n\t// backtick keeps the pattern on Anthropic's pairing-error template.\n\t\"was found without a corresponding `\",\n\n\t// gRPC based providers (e.g. NVIDIA NIM)\n\t\"ResourceExhausted\",\n]);\n\n/**\n * Retry policy: bounded attempts with exponential backoff (`baseDelayMs * 2^(attempt-1)`).\n * Matches `settings.retry` (`enabled`, `maxRetries`, `baseDelayMs`) in coding-agent; kept\n * here so the classifier and the policy-driven retry loop live together and stay reusable\n * by the SDK and other callers.\n */\nexport interface RetryPolicy {\n\tenabled: boolean;\n\t/** Max retry attempts (0 = no retries). The initial call never counts as a retry. */\n\tmaxRetries: number;\n\t/** Base delay in ms. Per-attempt delay is `baseDelayMs * 2^(attempt-1)` before jitter. */\n\tbaseDelayMs: number;\n}\n\n/** Optional callbacks emitted by {@link retryAssistantCall} around each retry. */\nexport interface RetryCallbacks {\n\t/** Emitted before the backoff sleep of each retry attempt (1-indexed). */\n\tonRetryScheduled?: (\n\t\tattempt: number,\n\t\tmaxAttempts: number,\n\t\tdelayMs: number,\n\t\terrorMessage: string,\n\t) => void | Promise<void>;\n\t/** Emitted after the backoff sleep, immediately before the retried call starts. */\n\tonRetryAttemptStart?: () => void | Promise<void>;\n\t/** Emitted once when the loop ends: success if a later call completed normally. */\n\tonRetryFinished?: (success: boolean, attempt: number, finalError?: string) => void | Promise<void>;\n}\n\nclass RetrySleepAbortError extends Error {\n\tconstructor() {\n\t\tsuper(\"Aborted\");\n\t}\n}\n\nfunction sleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new RetrySleepAbortError());\n\t\t\treturn;\n\t\t}\n\t\tconst timeout = setTimeout(resolve, ms);\n\t\tsignal?.addEventListener(\n\t\t\t\"abort\",\n\t\t\t() => {\n\t\t\t\tclearTimeout(timeout);\n\t\t\t\treject(new RetrySleepAbortError());\n\t\t\t},\n\t\t\t{ once: true },\n\t\t);\n\t});\n}\n\n/**\n * Run a single assistant-producing call with bounded retry on transient errors.\n *\n * Behavior:\n * - A successful response is returned immediately. Aborts are terminal and never\n *   retried, but reported as unsuccessful if they happen after a retry was scheduled.\n *   Aborts during the backoff sleep are normalized to an aborted `AssistantMessage`\n *   too, so callers do not need to care when cancellation happened.\n * - A non-retryable error (per {@link isRetryableAssistantError}, including quota/\n *   billing exhaustion) is returned immediately so deterministic errors fail fast.\n * - Otherwise retries up to `maxRetries` times with exponential backoff, emitting\n *   `onRetryScheduled` before each sleep, `onRetryAttemptStart` after each sleep before\n *   the retried call starts, and `onRetryFinished` once at the end (whether the loop\n *   ends in success, exhausted retries, or an aborted backoff).\n *\n * When `policy` is undefined or disabled, the first response is returned unchanged\n * (equivalent to calling `produce()` directly).\n */\nexport async function retryAssistantCall(\n\tproduce: () => Promise<AssistantMessage>,\n\tpolicy: RetryPolicy | undefined,\n\tsignal: AbortSignal | undefined,\n\tcallbacks?: RetryCallbacks,\n): Promise<AssistantMessage> {\n\tconst maxAttempts = policy?.enabled ? policy.maxRetries : 0;\n\n\tlet attempt = 0;\n\tlet lastRetry: { attempt: number; errorMessage: string } | undefined;\n\tfor (;;) {\n\t\tconst response = await produce();\n\n\t\t// Abort: terminal but not successful. Never retry an aborted message.\n\t\tif (response.stopReason === \"aborted\") {\n\t\t\tif (lastRetry) await callbacks?.onRetryFinished?.(false, lastRetry.attempt);\n\t\t\treturn response;\n\t\t}\n\n\t\t// Success: non-error, non-abort responses return as-is.\n\t\tif (response.stopReason !== \"error\") {\n\t\t\tif (lastRetry) await callbacks?.onRetryFinished?.(true, lastRetry.attempt);\n\t\t\treturn response;\n\t\t}\n\n\t\t// Non-retryable, or budget exhausted: return the final error message.\n\t\tif (attempt >= maxAttempts || !isRetryableAssistantError(response)) {\n\t\t\tif (lastRetry) await callbacks?.onRetryFinished?.(false, lastRetry.attempt, response.errorMessage);\n\t\t\treturn response;\n\t\t}\n\n\t\tattempt++;\n\t\tlastRetry = { attempt, errorMessage: response.errorMessage || \"Unknown error\" };\n\t\tconst delayMs = policy!.baseDelayMs * 2 ** (attempt - 1);\n\t\tawait callbacks?.onRetryScheduled?.(attempt, maxAttempts, delayMs, lastRetry.errorMessage);\n\n\t\t// Normalize aborts during retry backoff to the same AssistantMessage shape as\n\t\t// provider stream aborts, so callers do not need to care when cancellation happened.\n\t\ttry {\n\t\t\tawait sleep(delayMs, signal);\n\t\t} catch (error) {\n\t\t\tawait callbacks?.onRetryFinished?.(false, attempt, lastRetry.errorMessage);\n\t\t\tif (error instanceof RetrySleepAbortError) {\n\t\t\t\treturn { ...response, stopReason: \"aborted\", errorMessage: undefined };\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t\tawait callbacks?.onRetryAttemptStart?.();\n\t}\n}\n\n/**\n * Classifies whether a failed assistant message looks like a transient provider\n * or transport error, so callers can decide if the last assistant turn should be\n * restarted.\n *\n * This does not implement retry policy. Callers should first handle context\n * overflow separately, then apply their own retry budget, backoff, and reporting\n * before restarting the assistant turn.\n */\nexport function isRetryableAssistantError(message: AssistantMessage): boolean {\n\tif (\n\t\tmessage.stopReason !== \"error\" ||\n\t\tmessage.stopDetails?.type === \"refusal\" ||\n\t\tmessage.stopDetails?.type === \"sensitive\" ||\n\t\t!message.errorMessage\n\t) {\n\t\treturn false;\n\t}\n\tconst errorMessage = message.errorMessage;\n\tif (NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN.test(errorMessage)) return false;\n\treturn RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage);\n}\n"]}
+{"version":3,"file":"retry.js","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAEA,SAAS,yBAAyB,CAAC,QAA2B;IAC7D,OAAO,IAAI,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,CAAC;AAC5C,CAAC;AAED,MAAM,0CAA0C,GAAG,yBAAyB,CAAC;IAC5E,8EAA8E;IAC9E,2EAA2E;IAC3E,mBAAmB;IACnB,qBAAqB;IAErB,6EAA6E;IAC7E,yDAAyD;IACzD,6BAA6B;IAC7B,mBAAmB;IAEnB,4EAA4E;IAC5E,4EAA4E;IAC5E,oBAAoB;IACpB,eAAe;IACf,gBAAgB;IAChB,SAAS;CACT,CAAC,CAAC;AAEH,MAAM,gCAAgC,GAAG,yBAAyB,CAAC;IAClE,0EAA0E;IAC1E,YAAY;IACZ,aAAa;IACb,mBAAmB;IACnB,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,8EAA8E;IAC9E,6EAA6E;IAC7E,KAAK;IACL,KAAK;IACL,sBAAsB;IACtB,eAAe;IACf,iBAAiB;IAEjB,8EAA8E;IAC9E,+CAA+C;IAC/C,2BAA2B;IAE3B,2EAA2E;IAC3E,2EAA2E;IAC3E,2EAA2E;IAC3E,gBAAgB;IAChB,mBAAmB;IACnB,qBAAqB;IACrB,kBAAkB;IAClB,mBAAmB;IACnB,cAAc;IACd,aAAa;IACb,WAAW;IACX,WAAW;IACX,mBAAmB;IACnB,uBAAuB;IACvB,sBAAsB;IACtB,gBAAgB;IAChB,8BAA8B;IAC9B,YAAY;IACZ,SAAS;IACT,YAAY;IAEZ,+EAA+E;IAC/E,mBAAmB;IACnB,kBAAkB;IAElB,yEAAyE;IACzE,8EAA8E;IAC9E,yEAAyE;IACzE,eAAe;IACf,kCAAkC;IAClC,+CAA+C;IAC/C,sCAAsC;IAEtC,4EAA4E;IAC5E,iEAAiE;IACjE,aAAa;IAEb,6EAA6E;IAC7E,6BAA6B;IAC7B,4BAA4B;IAC5B,wBAAwB;IACxB,2BAA2B;IAE3B,gFAAgF;IAChF,4EAA4E;IAC5E,4EAA4E;IAC5E,6EAA6E;IAC7E,+EAA+E;IAC/E,gFAAgF;IAChF,gFAAgF;IAChF,oEAAoE;IACpE,qCAAqC;IAErC,yCAAyC;IACzC,mBAAmB;CACnB,CAAC,CAAC;AA+BH,MAAM,oBAAqB,SAAQ,KAAK;IACvC;QACC,KAAK,CAAC,SAAS,CAAC,CAAC;IAClB,CAAC;CACD;AAED,SAAS,KAAK,CAAC,EAAU,EAAE,MAAoB;IAC9C,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACtC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,MAAM,CAAC,IAAI,oBAAoB,EAAE,CAAC,CAAC;YACnC,OAAO;QACR,CAAC;QACD,MAAM,OAAO,GAAG,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACxC,MAAM,EAAE,gBAAgB,CACvB,OAAO,EACP,GAAG,EAAE;YACJ,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,IAAI,oBAAoB,EAAE,CAAC,CAAC;QACpC,CAAC,EACD,EAAE,IAAI,EAAE,IAAI,EAAE,CACd,CAAC;IACH,CAAC,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACvC,OAAwC,EACxC,MAA+B,EAC/B,MAA+B,EAC/B,SAA0B;IAE1B,MAAM,WAAW,GAAG,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAE5D,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,SAAgE,CAAC;IACrE,SAAS,CAAC;QACT,MAAM,QAAQ,GAAG,MAAM,OAAO,EAAE,CAAC;QAEjC,sEAAsE;QACtE,IAAI,QAAQ,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;YACvC,IAAI,SAAS;gBAAE,MAAM,SAAS,EAAE,eAAe,EAAE,CAAC,KAAK,EAAE,SAAS,CAAC,OAAO,CAAC,CAAC;YAC5E,OAAO,QAAQ,CAAC;QACjB,CAAC;QAED,wDAAwD;QACxD,IAAI,QAAQ,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;YACrC,IAAI,SAAS;gBAAE,MAAM,SAAS,EAAE,eAAe,EAAE,CAAC,IAAI,EAAE,SAAS,CAAC,OAAO,CAAC,CAAC;YAC3E,OAAO,QAAQ,CAAC;QACjB,CAAC;QAED,sEAAsE;QACtE,IAAI,OAAO,IAAI,WAAW,IAAI,CAAC,yBAAyB,CAAC,QAAQ,CAAC,EAAE,CAAC;YACpE,IAAI,SAAS;gBAAE,MAAM,SAAS,EAAE,eAAe,EAAE,CAAC,KAAK,EAAE,SAAS,CAAC,OAAO,EAAE,QAAQ,CAAC,YAAY,CAAC,CAAC;YACnG,OAAO,QAAQ,CAAC;QACjB,CAAC;QAED,OAAO,EAAE,CAAC;QACV,SAAS,GAAG,EAAE,OAAO,EAAE,YAAY,EAAE,QAAQ,CAAC,YAAY,IAAI,eAAe,EAAE,CAAC;QAChF,MAAM,OAAO,GAAG,MAAO,CAAC,WAAW,GAAG,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC;QACzD,MAAM,SAAS,EAAE,gBAAgB,EAAE,CAAC,OAAO,EAAE,WAAW,EAAE,OAAO,EAAE,SAAS,CAAC,YAAY,CAAC,CAAC;QAE3F,8EAA8E;QAC9E,qFAAqF;QACrF,IAAI,CAAC;YACJ,MAAM,KAAK,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,MAAM,SAAS,EAAE,eAAe,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,SAAS,CAAC,YAAY,CAAC,CAAC;YAC3E,IAAI,KAAK,YAAY,oBAAoB,EAAE,CAAC;gBAC3C,OAAO,EAAE,GAAG,QAAQ,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,SAAS,EAAE,CAAC;YACxE,CAAC;YACD,MAAM,KAAK,CAAC;QACb,CAAC;QACD,MAAM,SAAS,EAAE,mBAAmB,EAAE,EAAE,CAAC;IAC1C,CAAC;AACF,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,yBAAyB,CAAC,OAAyB;IAClE,IACC,OAAO,CAAC,UAAU,KAAK,OAAO;QAC9B,OAAO,CAAC,WAAW,EAAE,IAAI,KAAK,SAAS;QACvC,OAAO,CAAC,WAAW,EAAE,IAAI,KAAK,WAAW;QACzC,CAAC,OAAO,CAAC,YAAY,EACpB,CAAC;QACF,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,uBAAuB,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC;AACtD,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,uBAAuB,CAAC,YAAoB;IAC3D,IAAI,CAAC,YAAY;QAAE,OAAO,KAAK,CAAC;IAChC,IAAI,0CAA0C,CAAC,IAAI,CAAC,YAAY,CAAC;QAAE,OAAO,KAAK,CAAC;IAChF,OAAO,gCAAgC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;AAC5D,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\nfunction buildProviderErrorPattern(patterns: readonly string[]): RegExp {\n\treturn new RegExp(patterns.join(\"|\"), \"i\");\n}\n\nconst NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN = buildProviderErrorPattern([\n\t// OpenCode Go/free-tier limits returned as 429 JSON error types by OpenCode's\n\t// Zen API. These are subscription/account limits, not transient throttles.\n\t\"GoUsageLimitError\",\n\t\"FreeUsageLimitError\",\n\n\t// OpenCode Go subscription-limit text asks users to enable available-balance\n\t// usage after rolling/weekly/monthly limits are reached.\n\t\"Monthly usage limit reached\",\n\t\"available balance\",\n\n\t// Generic quota/budget/billing exhaustion. `insufficient_quota` is OpenAI's\n\t// quota/billing error code; the other strings cover common gateway wording.\n\t\"insufficient_quota\",\n\t\"out of budget\",\n\t\"quota exceeded\",\n\t\"billing\",\n]);\n\nconst RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([\n\t// Generic provider load, HTTP status, and server-side transient failures.\n\t\"overloaded\",\n\t\"rate.?limit\",\n\t\"too many requests\",\n\t\"429\",\n\t\"500\",\n\t\"502\",\n\t\"503\",\n\t\"504\",\n\t// Cloudflare 522 (Connection timed out): origin stopped responding; transient\n\t// like the other 5xx gateway statuses, surfaced as \"Error: error code: 522\".\n\t\"522\",\n\t\"524\",\n\t\"service.?unavailable\",\n\t\"server.?error\",\n\t\"internal.?error\",\n\n\t// Wrapper/provider text for transient upstream failures, including OpenRouter\n\t// \"Provider returned error\" responses (#2264).\n\t\"provider.?returned.?error\",\n\n\t// Network, proxy, and fetch transport failures. This includes OpenAI Codex\n\t// raw-fetch failures such as \"upstream connect\", \"connection refused\", and\n\t// \"reset before headers\" (#733), plus OpenRouter connection drops (#3317).\n\t\"network.?error\",\n\t\"connection.?error\",\n\t\"connection.?refused\",\n\t\"connection.?lost\",\n\t\"other side closed\",\n\t\"fetch failed\",\n\t\"getaddrinfo\",\n\t\"ENOTFOUND\",\n\t\"EAI_AGAIN\",\n\t\"upstream.?connect\",\n\t\"upstream.?unavailable\",\n\t\"reset before headers\",\n\t\"socket hang up\",\n\t\"socket connection was closed\",\n\t\"timed? out\",\n\t\"timeout\",\n\t\"terminated\",\n\n\t// WebSocket transports can report close/error text instead of HTTP/fetch text.\n\t\"websocket.?closed\",\n\t\"websocket.?error\",\n\n\t// Premature stream endings from SDKs and transports. Anthropic can throw\n\t// \"stream ended without ...\" and \"Anthropic stream ended before message_stop\"\n\t// (#4433); Bedrock/Smithy can throw an HTTP/2 no-response error (#3594).\n\t\"ended without\",\n\t\"stream ended before message_stop\",\n\t\"stream ended before a terminal response event\",\n\t\"http2 request did not get a response\",\n\n\t// Provider-requested retry delay cap failures should flow through the outer\n\t// retry policy so callers can surface/abort the backoff (#1123).\n\t\"retry delay\",\n\n\t// Explicit retry guidance emitted mid-stream by OpenAI Responses and Bedrock\n\t// stream exceptions (#6019).\n\t\"you can retry your request\",\n\t\"try your request again\",\n\t\"please retry your request\",\n\n\t// Anthropic server-tool pairing rejections, e.g. \"`web_search` tool use with id\n\t// `srvtoolu_...` was found without a corresponding `web_search_tool_result`\n\t// block\". A turn closed before its deferred server tool could answer leaves\n\t// the unpairable half in history, and replaying it 400s every later request.\n\t// The Anthropic request builder repairs the replayed history, so a retry sends\n\t// a valid payload; keeping the class retryable is what lets retry and the model\n\t// fallback chain unwedge such a session instead of dead-ending it. The trailing\n\t// backtick keeps the pattern on Anthropic's pairing-error template.\n\t\"was found without a corresponding `\",\n\n\t// gRPC based providers (e.g. NVIDIA NIM)\n\t\"ResourceExhausted\",\n]);\n\n/**\n * Retry policy: bounded attempts with exponential backoff (`baseDelayMs * 2^(attempt-1)`).\n * Matches `settings.retry` (`enabled`, `maxRetries`, `baseDelayMs`) in coding-agent; kept\n * here so the classifier and the policy-driven retry loop live together and stay reusable\n * by the SDK and other callers.\n */\nexport interface RetryPolicy {\n\tenabled: boolean;\n\t/** Max retry attempts (0 = no retries). The initial call never counts as a retry. */\n\tmaxRetries: number;\n\t/** Base delay in ms. Per-attempt delay is `baseDelayMs * 2^(attempt-1)` before jitter. */\n\tbaseDelayMs: number;\n}\n\n/** Optional callbacks emitted by {@link retryAssistantCall} around each retry. */\nexport interface RetryCallbacks {\n\t/** Emitted before the backoff sleep of each retry attempt (1-indexed). */\n\tonRetryScheduled?: (\n\t\tattempt: number,\n\t\tmaxAttempts: number,\n\t\tdelayMs: number,\n\t\terrorMessage: string,\n\t) => void | Promise<void>;\n\t/** Emitted after the backoff sleep, immediately before the retried call starts. */\n\tonRetryAttemptStart?: () => void | Promise<void>;\n\t/** Emitted once when the loop ends: success if a later call completed normally. */\n\tonRetryFinished?: (success: boolean, attempt: number, finalError?: string) => void | Promise<void>;\n}\n\nclass RetrySleepAbortError extends Error {\n\tconstructor() {\n\t\tsuper(\"Aborted\");\n\t}\n}\n\nfunction sleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(new RetrySleepAbortError());\n\t\t\treturn;\n\t\t}\n\t\tconst timeout = setTimeout(resolve, ms);\n\t\tsignal?.addEventListener(\n\t\t\t\"abort\",\n\t\t\t() => {\n\t\t\t\tclearTimeout(timeout);\n\t\t\t\treject(new RetrySleepAbortError());\n\t\t\t},\n\t\t\t{ once: true },\n\t\t);\n\t});\n}\n\n/**\n * Run a single assistant-producing call with bounded retry on transient errors.\n *\n * Behavior:\n * - A successful response is returned immediately. Aborts are terminal and never\n *   retried, but reported as unsuccessful if they happen after a retry was scheduled.\n *   Aborts during the backoff sleep are normalized to an aborted `AssistantMessage`\n *   too, so callers do not need to care when cancellation happened.\n * - A non-retryable error (per {@link isRetryableAssistantError}, including quota/\n *   billing exhaustion) is returned immediately so deterministic errors fail fast.\n * - Otherwise retries up to `maxRetries` times with exponential backoff, emitting\n *   `onRetryScheduled` before each sleep, `onRetryAttemptStart` after each sleep before\n *   the retried call starts, and `onRetryFinished` once at the end (whether the loop\n *   ends in success, exhausted retries, or an aborted backoff).\n *\n * When `policy` is undefined or disabled, the first response is returned unchanged\n * (equivalent to calling `produce()` directly).\n */\nexport async function retryAssistantCall(\n\tproduce: () => Promise<AssistantMessage>,\n\tpolicy: RetryPolicy | undefined,\n\tsignal: AbortSignal | undefined,\n\tcallbacks?: RetryCallbacks,\n): Promise<AssistantMessage> {\n\tconst maxAttempts = policy?.enabled ? policy.maxRetries : 0;\n\n\tlet attempt = 0;\n\tlet lastRetry: { attempt: number; errorMessage: string } | undefined;\n\tfor (;;) {\n\t\tconst response = await produce();\n\n\t\t// Abort: terminal but not successful. Never retry an aborted message.\n\t\tif (response.stopReason === \"aborted\") {\n\t\t\tif (lastRetry) await callbacks?.onRetryFinished?.(false, lastRetry.attempt);\n\t\t\treturn response;\n\t\t}\n\n\t\t// Success: non-error, non-abort responses return as-is.\n\t\tif (response.stopReason !== \"error\") {\n\t\t\tif (lastRetry) await callbacks?.onRetryFinished?.(true, lastRetry.attempt);\n\t\t\treturn response;\n\t\t}\n\n\t\t// Non-retryable, or budget exhausted: return the final error message.\n\t\tif (attempt >= maxAttempts || !isRetryableAssistantError(response)) {\n\t\t\tif (lastRetry) await callbacks?.onRetryFinished?.(false, lastRetry.attempt, response.errorMessage);\n\t\t\treturn response;\n\t\t}\n\n\t\tattempt++;\n\t\tlastRetry = { attempt, errorMessage: response.errorMessage || \"Unknown error\" };\n\t\tconst delayMs = policy!.baseDelayMs * 2 ** (attempt - 1);\n\t\tawait callbacks?.onRetryScheduled?.(attempt, maxAttempts, delayMs, lastRetry.errorMessage);\n\n\t\t// Normalize aborts during retry backoff to the same AssistantMessage shape as\n\t\t// provider stream aborts, so callers do not need to care when cancellation happened.\n\t\ttry {\n\t\t\tawait sleep(delayMs, signal);\n\t\t} catch (error) {\n\t\t\tawait callbacks?.onRetryFinished?.(false, attempt, lastRetry.errorMessage);\n\t\t\tif (error instanceof RetrySleepAbortError) {\n\t\t\t\treturn { ...response, stopReason: \"aborted\", errorMessage: undefined };\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t\tawait callbacks?.onRetryAttemptStart?.();\n\t}\n}\n\n/**\n * Classifies whether a failed assistant message looks like a transient provider\n * or transport error, so callers can decide if the last assistant turn should be\n * restarted.\n *\n * This does not implement retry policy. Callers should first handle context\n * overflow separately, then apply their own retry budget, backoff, and reporting\n * before restarting the assistant turn.\n */\nexport function isRetryableAssistantError(message: AssistantMessage): boolean {\n\tif (\n\t\tmessage.stopReason !== \"error\" ||\n\t\tmessage.stopDetails?.type === \"refusal\" ||\n\t\tmessage.stopDetails?.type === \"sensitive\" ||\n\t\t!message.errorMessage\n\t) {\n\t\treturn false;\n\t}\n\treturn isRetryableErrorMessage(message.errorMessage);\n}\n\n/**\n * Classifies a raw error-message string with the same transient-vs-terminal\n * rules as {@link isRetryableAssistantError}, for callers that hold a thrown\n * `Error` instead of an `AssistantMessage` (e.g. compaction summarization\n * failures that must decide between degrading and surfacing loudly).\n */\nexport function isRetryableErrorMessage(errorMessage: string): boolean {\n\tif (!errorMessage) return false;\n\tif (NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN.test(errorMessage)) return false;\n\treturn RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage);\n}\n"]}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@code-yeongyu/senpi-ai",
-	"version": "2026.7.26",
+	"version": "2026.7.28",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",