@code-yeongyu/senpi-ai 2026.8.12 → 2026.8.13

package/dist/utils/abort.js

@@ -0,0 +1,49 @@
+function abortReason(signal) {
+    if (signal.reason !== undefined)
+        return signal.reason;
+    const error = new Error("The operation was aborted");
+    error.name = "AbortError";
+    return error;
+}
+/** Create an operation-local signal for public APIs whose signal is optional. */
+export function operationSignal(signal) {
+    return signal ?? new AbortController().signal;
+}
+/**
+ * Stop waiting for an operation when its signal aborts while continuing to
+ * observe the abandoned promise so a later rejection is always handled.
+ */
+export function raceWithAbortSignal(operation, signal) {
+    if (signal.aborted) {
+        void operation.catch(() => { });
+        return Promise.reject(abortReason(signal));
+    }
+    return new Promise((resolve, reject) => {
+        let settled = false;
+        const cleanup = () => signal.removeEventListener("abort", onAbort);
+        const onAbort = () => {
+            if (settled)
+                return;
+            settled = true;
+            cleanup();
+            reject(abortReason(signal));
+        };
+        signal.addEventListener("abort", onAbort, { once: true });
+        void operation.then((value) => {
+            if (settled)
+                return;
+            settled = true;
+            cleanup();
+            resolve(value);
+        }, (error) => {
+            if (settled)
+                return;
+            settled = true;
+            cleanup();
+            reject(error);
+        });
+        if (signal.aborted)
+            onAbort();
+    });
+}
+//# sourceMappingURL=abort.js.map

package/dist/utils/abort.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"abort.js","sourceRoot":"","sources":["../../src/utils/abort.ts"],"names":[],"mappings":"AAAA,SAAS,WAAW,CAAC,MAAmB;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS;QAAE,OAAO,MAAM,CAAC,MAAM,CAAC;IACtD,MAAM,KAAK,GAAG,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;IACrD,KAAK,CAAC,IAAI,GAAG,YAAY,CAAC;IAC1B,OAAO,KAAK,CAAC;AACd,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,eAAe,CAAC,MAAoB;IACnD,OAAO,MAAM,IAAI,IAAI,eAAe,EAAE,CAAC,MAAM,CAAC;AAC/C,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAI,SAAqB,EAAE,MAAmB;IAChF,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,KAAK,SAAS,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;QAC/B,OAAO,OAAO,CAAC,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC;IAC5C,CAAC;IAED,OAAO,IAAI,OAAO,CAAI,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACzC,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QACnE,MAAM,OAAO,GAAG,GAAG,EAAE;YACpB,IAAI,OAAO;gBAAE,OAAO;YACpB,OAAO,GAAG,IAAI,CAAC;YACf,OAAO,EAAE,CAAC;YACV,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC;QAC7B,CAAC,CAAC;QAEF,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;QAC1D,KAAK,SAAS,CAAC,IAAI,CAClB,CAAC,KAAK,EAAE,EAAE;YACT,IAAI,OAAO;gBAAE,OAAO;YACpB,OAAO,GAAG,IAAI,CAAC;YACf,OAAO,EAAE,CAAC;YACV,OAAO,CAAC,KAAK,CAAC,CAAC;QAChB,CAAC,EACD,CAAC,KAAc,EAAE,EAAE;YAClB,IAAI,OAAO;gBAAE,OAAO;YACpB,OAAO,GAAG,IAAI,CAAC;YACf,OAAO,EAAE,CAAC;YACV,MAAM,CAAC,KAAK,CAAC,CAAC;QACf,CAAC,CACD,CAAC;QACF,IAAI,MAAM,CAAC,OAAO;YAAE,OAAO,EAAE,CAAC;IAC/B,CAAC,CAAC,CAAC;AACJ,CAAC","sourcesContent":["function abortReason(signal: AbortSignal): unknown {\n\tif (signal.reason !== undefined) return signal.reason;\n\tconst error = new Error(\"The operation was aborted\");\n\terror.name = \"AbortError\";\n\treturn error;\n}\n\n/** Create an operation-local signal for public APIs whose signal is optional. */\nexport function operationSignal(signal?: AbortSignal): AbortSignal {\n\treturn signal ?? new AbortController().signal;\n}\n\n/**\n * Stop waiting for an operation when its signal aborts while continuing to\n * observe the abandoned promise so a later rejection is always handled.\n */\nexport function raceWithAbortSignal<T>(operation: Promise<T>, signal: AbortSignal): Promise<T> {\n\tif (signal.aborted) {\n\t\tvoid operation.catch(() => {});\n\t\treturn Promise.reject(abortReason(signal));\n\t}\n\n\treturn new Promise<T>((resolve, reject) => {\n\t\tlet settled = false;\n\t\tconst cleanup = () => signal.removeEventListener(\"abort\", onAbort);\n\t\tconst onAbort = () => {\n\t\t\tif (settled) return;\n\t\t\tsettled = true;\n\t\t\tcleanup();\n\t\t\treject(abortReason(signal));\n\t\t};\n\n\t\tsignal.addEventListener(\"abort\", onAbort, { once: true });\n\t\tvoid operation.then(\n\t\t\t(value) => {\n\t\t\t\tif (settled) return;\n\t\t\t\tsettled = true;\n\t\t\t\tcleanup();\n\t\t\t\tresolve(value);\n\t\t\t},\n\t\t\t(error: unknown) => {\n\t\t\t\tif (settled) return;\n\t\t\t\tsettled = true;\n\t\t\t\tcleanup();\n\t\t\t\treject(error);\n\t\t\t},\n\t\t);\n\t\tif (signal.aborted) onAbort();\n\t});\n}\n"]}

package/dist/utils/deferred-tools.js.map

@@ -1 +1 @@
-{"version":3,"file":"deferred-tools.js","sourceRoot":"","sources":["../../src/utils/deferred-tools.ts"],"names":[],"mappings":"AAIA,MAAM,gBAAgB,GAAuB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC;AAE5D,yEAAyE;AACzE,MAAM,UAAU,kBAAkB,CACjC,OAAgB,EAChB,OAAgB,EAChB,aAAa,GAAuB,gBAAgB;IAEpD,MAAM,WAAW,GAAG,IAAI,GAAG,EAAgB,CAAC;IAC5C,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,KAAK,IAAI,EAAE;QAAE,WAAW,CAAC,GAAG,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,CAAC,CAAC;IACxF,IAAI,CAAC,OAAO;QAAE,OAAO,EAAE,SAAS,EAAE,CAAC,GAAG,WAAW,CAAC,MAAM,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,GAAG,EAAE,EAAE,CAAC;IAEnF,MAAM,aAAa,GAAG,IAAI,GAAG,EAAU,CAAC;IACxC,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IACpC,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;QACxC,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YAClC,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;gBACrC,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU;oBAAE,SAAS,CAAC,GAAG,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;YACzE,CAAC;QACF,CAAC;aAAM,IAAI,OAAO,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;YAC1C,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,cAAc,IAAI,EAAE,EAAE,CAAC;gBACjD,MAAM,cAAc,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;gBAC3C,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,cAAc,CAAC;oBAAE,aAAa,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;YACvE,CAAC;QACF,CAAC;IACF,CAAC;IAED,MAAM,SAAS,GAAW,EAAE,CAAC;IAC7B,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAgB,CAAC;IACzC,KAAK,MAAM,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,WAAW,EAAE,CAAC;QACxC,IAAI,aAAa,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;;YACjD,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAC;AAChC,CAAC","sourcesContent":["import type { Context, Tool } from \"../types.ts\";\n\ntype ToolNameNormalizer = (name: string) => string;\n\nconst identityToolName: ToolNameNormalizer = (name) => name;\n\n/** Split current tools into prefix and transcript-loaded definitions. */\nexport function splitDeferredTools(\n\tcontext: Context,\n\tenabled: boolean,\n\tnormalizeName: ToolNameNormalizer = identityToolName,\n): { immediate: Tool[]; deferred: Map<string, Tool> } {\n\tconst uniqueTools = new Map<string, Tool>();\n\tfor (const tool of context.tools ?? []) uniqueTools.set(normalizeName(tool.name), tool);\n\tif (!enabled) return { immediate: [...uniqueTools.values()], deferred: new Map() };\n\n\tconst deferredNames = new Set<string>();\n\tconst usedNames = new Set<string>();\n\tfor (const message of context.messages) {\n\t\tif (message.role === \"assistant\") {\n\t\t\tfor (const block of message.content) {\n\t\t\t\tif (block.type === \"toolCall\") usedNames.add(normalizeName(block.name));\n\t\t\t}\n\t\t} else if (message.role === \"toolResult\") {\n\t\t\tfor (const name of message.addedToolNames ?? []) {\n\t\t\t\tconst normalizedName = normalizeName(name);\n\t\t\t\tif (!usedNames.has(normalizedName)) deferredNames.add(normalizedName);\n\t\t\t}\n\t\t}\n\t}\n\n\tconst immediate: Tool[] = [];\n\tconst deferred = new Map<string, Tool>();\n\tfor (const [name, tool] of uniqueTools) {\n\t\tif (deferredNames.has(name)) deferred.set(name, tool);\n\t\telse immediate.push(tool);\n\t}\n\treturn { immediate, deferred };\n}\n"]}
+{"version":3,"file":"deferred-tools.js","sourceRoot":"","sources":["../../src/utils/deferred-tools.ts"],"names":[],"mappings":"AAIA,MAAM,gBAAgB,GAAuB,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC;AAE5D,yEAAyE;AACzE,MAAM,UAAU,kBAAkB,CACjC,OAAgB,EAChB,OAAgB,EAChB,gBAAoC,gBAAgB;IAEpD,MAAM,WAAW,GAAG,IAAI,GAAG,EAAgB,CAAC;IAC5C,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,KAAK,IAAI,EAAE;QAAE,WAAW,CAAC,GAAG,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,CAAC,CAAC;IACxF,IAAI,CAAC,OAAO;QAAE,OAAO,EAAE,SAAS,EAAE,CAAC,GAAG,WAAW,CAAC,MAAM,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,GAAG,EAAE,EAAE,CAAC;IAEnF,MAAM,aAAa,GAAG,IAAI,GAAG,EAAU,CAAC;IACxC,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IACpC,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;QACxC,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YAClC,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;gBACrC,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU;oBAAE,SAAS,CAAC,GAAG,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;YACzE,CAAC;QACF,CAAC;aAAM,IAAI,OAAO,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;YAC1C,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,cAAc,IAAI,EAAE,EAAE,CAAC;gBACjD,MAAM,cAAc,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;gBAC3C,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,cAAc,CAAC;oBAAE,aAAa,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;YACvE,CAAC;QACF,CAAC;IACF,CAAC;IAED,MAAM,SAAS,GAAW,EAAE,CAAC;IAC7B,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAgB,CAAC;IACzC,KAAK,MAAM,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,WAAW,EAAE,CAAC;QACxC,IAAI,aAAa,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;;YACjD,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAC;AAChC,CAAC","sourcesContent":["import type { Context, Tool } from \"../types.ts\";\n\ntype ToolNameNormalizer = (name: string) => string;\n\nconst identityToolName: ToolNameNormalizer = (name) => name;\n\n/** Split current tools into prefix and transcript-loaded definitions. */\nexport function splitDeferredTools(\n\tcontext: Context,\n\tenabled: boolean,\n\tnormalizeName: ToolNameNormalizer = identityToolName,\n): { immediate: Tool[]; deferred: Map<string, Tool> } {\n\tconst uniqueTools = new Map<string, Tool>();\n\tfor (const tool of context.tools ?? []) uniqueTools.set(normalizeName(tool.name), tool);\n\tif (!enabled) return { immediate: [...uniqueTools.values()], deferred: new Map() };\n\n\tconst deferredNames = new Set<string>();\n\tconst usedNames = new Set<string>();\n\tfor (const message of context.messages) {\n\t\tif (message.role === \"assistant\") {\n\t\t\tfor (const block of message.content) {\n\t\t\t\tif (block.type === \"toolCall\") usedNames.add(normalizeName(block.name));\n\t\t\t}\n\t\t} else if (message.role === \"toolResult\") {\n\t\t\tfor (const name of message.addedToolNames ?? []) {\n\t\t\t\tconst normalizedName = normalizeName(name);\n\t\t\t\tif (!usedNames.has(normalizedName)) deferredNames.add(normalizedName);\n\t\t\t}\n\t\t}\n\t}\n\n\tconst immediate: Tool[] = [];\n\tconst deferred = new Map<string, Tool>();\n\tfor (const [name, tool] of uniqueTools) {\n\t\tif (deferredNames.has(name)) deferred.set(name, tool);\n\t\telse immediate.push(tool);\n\t}\n\treturn { immediate, deferred };\n}\n"]}

package/dist/utils/event-stream.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"event-stream.d.ts","sourceRoot":"","sources":["../../src/utils/event-stream.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,MAAM,aAAa,CAAC;AAG3E,qBAAa,WAAW,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAE,YAAW,aAAa,CAAC,CAAC,CAAC;;IAG7D,OAAO,CAAC,OAAO,CAAgG;IAC/G,OAAO,CAAC,IAAI,CAAS;IAGrB,OAAO,CAAC,kBAAkB,CAAa;IACvC,OAAO,CAAC,kBAAkB,CAAuB;IACjD,OAAO,CAAC,iBAAiB,CAA4B;IACrD,OAAO,CAAC,UAAU,CAAwB;IAC1C,OAAO,CAAC,aAAa,CAAkB;IAEvC,YAAY,UAAU,EAAE,CAAC,KAAK,EAAE,CAAC,KAAK,OAAO,EAAE,aAAa,EAAE,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC,EAQ5E;IAED,IAAI,KAAK,IAAI,CAAC,EAAE,CAEf;IAmBD,IAAI,CAAC,KAAK,EAAE,CAAC,GAAG,IAAI,CAenB;IAED,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,GAAG,IAAI,CAUpB;IAED,IAAI,CAAC,KAAK,EAAE,OAAO,GAAG,IAAI,CAUzB;IAED,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,aAAa,CAAC,CAAC,CAAC,CAezC;IAED,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAEnB;CACD;AAED,qBAAa,2BAA4B,SAAQ,WAAW,CAAC,qBAAqB,EAAE,gBAAgB,CAAC;IACpG,cAYC;CACD;AAED,+EAA+E;AAC/E,wBAAgB,iCAAiC,IAAI,2BAA2B,CAE/E"}
+{"version":3,"file":"event-stream.d.ts","sourceRoot":"","sources":["../../src/utils/event-stream.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,qBAAqB,EAAE,MAAM,aAAa,CAAC;AAG3E,qBAAa,WAAW,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAE,YAAW,aAAa,CAAC,CAAC,CAAC;;IAG7D,OAAO,CAAC,OAAO,CAAgG;IAC/G,OAAO,CAAC,IAAI,CAAS;IAGrB,OAAO,CAAC,kBAAkB,CAAa;IACvC,OAAO,CAAC,kBAAkB,CAAuB;IACjD,OAAO,CAAC,iBAAiB,CAA4B;IACrD,OAAO,CAAC,UAAU,CAAwB;IAC1C,OAAO,CAAC,aAAa,CAAkB;gBAE3B,UAAU,EAAE,CAAC,KAAK,EAAE,CAAC,KAAK,OAAO,EAAE,aAAa,EAAE,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC;IAU7E,IAAI,KAAK,IAAI,CAAC,EAAE,CAEf;IAmBD,IAAI,CAAC,KAAK,EAAE,CAAC,GAAG,IAAI;IAiBpB,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,GAAG,IAAI;IAYrB,IAAI,CAAC,KAAK,EAAE,OAAO,GAAG,IAAI;IAY1B,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,aAAa,CAAC,CAAC,CAAC;IAiB1C,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC;CAGpB;AAED,qBAAa,2BAA4B,SAAQ,WAAW,CAAC,qBAAqB,EAAE,gBAAgB,CAAC;;CAcpG;AAED,+EAA+E;AAC/E,wBAAgB,iCAAiC,IAAI,2BAA2B,CAE/E"}

package/dist/utils/overflow.d.ts

@@ -2,11 +2,13 @@ import type { AssistantMessage } from "../types.ts";
 /**
  * Check if an assistant message represents a context overflow error.
  *
- * This handles two cases:
+ * This handles three cases:
  * 1. Error-based overflow: Most providers return stopReason "error" with a
  *    specific error message pattern.
  * 2. Silent overflow: Some providers accept overflow requests and return
  *    successfully. For these, we check if usage.input exceeds the context window.
+ * 3. Length-stop overflow: Xiaomi MiMo can return "length" with zero output when
+ *    the input fills the context window.
  *
  * ## Reliability by Provider
  *
@@ -52,4 +54,15 @@ import type { AssistantMessage } from "../types.ts";
  * @returns true if the message indicates a context overflow
  */
 export declare function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean;
+/**
+ * Check whether a length stop ended below the caller or model's intended output limit.
+ * Such responses may be caused by context pressure or provider-side truncation, so callers
+ * can make one bounded compact-and-retry attempt. `desiredMaxOutput` must be the original
+ * limit before any context-based clamping.
+ */
+export declare function isRecoverableLength(message: AssistantMessage, desiredMaxOutput: number): boolean;
+/**
+ * Get the overflow patterns for testing purposes.
+ */
+export declare function getOverflowPatterns(): RegExp[];
 //# sourceMappingURL=overflow.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"overflow.d.ts","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AA+EpD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmDG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,gBAAgB,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,OAAO,CA6B5F"}
+{"version":3,"file":"overflow.d.ts","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AA+EpD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqDG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,gBAAgB,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,OAAO,CA6B5F;AACD;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,GAAG,OAAO,CAEhG;AAED;;GAEG;AACH,wBAAgB,mBAAmB,IAAI,MAAM,EAAE,CAE9C"}

package/dist/utils/overflow.js

@@ -76,11 +76,13 @@ const NON_OVERFLOW_PATTERNS = [
 /**
  * Check if an assistant message represents a context overflow error.
  *
- * This handles two cases:
+ * This handles three cases:
  * 1. Error-based overflow: Most providers return stopReason "error" with a
  *    specific error message pattern.
  * 2. Silent overflow: Some providers accept overflow requests and return
  *    successfully. For these, we check if usage.input exceeds the context window.
+ * 3. Length-stop overflow: Xiaomi MiMo can return "length" with zero output when
+ *    the input fills the context window.
  *
  * ## Reliability by Provider
  *
@@ -152,4 +154,19 @@ export function isContextOverflow(message, contextWindow) {
     }
     return false;
 }
+/**
+ * Check whether a length stop ended below the caller or model's intended output limit.
+ * Such responses may be caused by context pressure or provider-side truncation, so callers
+ * can make one bounded compact-and-retry attempt. `desiredMaxOutput` must be the original
+ * limit before any context-based clamping.
+ */
+export function isRecoverableLength(message, desiredMaxOutput) {
+    return message.stopReason === "length" && desiredMaxOutput > 0 && message.usage.output < desiredMaxOutput;
+}
+/**
+ * Get the overflow patterns for testing purposes.
+ */
+export function getOverflowPatterns() {
+    return [...OVERFLOW_PATTERNS];
+}
 //# sourceMappingURL=overflow.js.map

package/dist/utils/overflow.js.map

@@ -1 +1 @@
-{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AACH,MAAM,iBAAiB,GAAG;IACzB,qBAAqB,EAAE,2BAA2B;IAClD,oBAAoB,EAAE,kDAAkD;IACxE,wCAAwC,EAAE,iBAAiB;IAC3D,6BAA6B,EAAE,uCAAuC;IACtE,4FAA4F,EAAE,sCAAsC;IACpI,yCAAyC,EAAE,kBAAkB;IAC7D,+BAA+B,EAAE,aAAa;IAC9C,oCAAoC,EAAE,OAAO;IAC7C,uCAAuC,EAAE,6BAA6B;IACtE,kEAAkE,EAAE,sBAAsB;IAC1F,iFAAiF,EAAE,cAAc;IACjG,2BAA2B,EAAE,iBAAiB;IAC9C,qCAAqC,EAAE,mBAAmB;IAC1D,kCAAkC,EAAE,YAAY;IAChD,+BAA+B,EAAE,UAAU;IAC3C,6BAA6B,EAAE,kBAAkB;IACjD,sDAAsD,EAAE,UAAU;IAClE,+EAA+E,EAAE,aAAa;IAC9F,gCAAgC,EAAE,yDAAyD;IAC3F,oDAAoD,EAAE,iCAAiC;IACvF,kCAAkC,EAAE,8BAA8B;IAClE,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;IAC5C,+CAA+C,EAAE,iCAAiC;CAClF,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,qBAAqB,GAAG;IAC7B,2CAA2C,EAAE,oFAAoF;IACjI,aAAa,EAAE,wBAAwB;IACvC,oBAAoB,EAAE,yBAAyB;CAC/C,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmDG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB;IAClF,uCAAuC;IACvC,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QAC5D,oFAAoF;QACpF,MAAM,aAAa,GAAG,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,CAAC;QACvF,IAAI,CAAC,aAAa,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,EAAE,CAAC;YACpF,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,8EAA8E;IAC9E,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,MAAM,EAAE,CAAC;QACpD,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,GAAG,aAAa,EAAE,CAAC;YACjC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,sFAAsF;IACtF,qFAAqF;IACrF,gEAAgE;IAChE,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpF,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,IAAI,aAAa,GAAG,IAAI,EAAE,CAAC;YACzC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AACd,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\n/**\n * Regex patterns to detect context overflow errors from different providers.\n *\n * These patterns match error messages returned when the input exceeds\n * the model's context window.\n *\n * Provider-specific patterns (with example error messages):\n *\n * - Anthropic: \"prompt is too long: 213462 tokens > 200000 maximum\"\n * - Anthropic: \"413 {\\\"error\\\":{\\\"type\\\":\\\"request_too_large\\\",\\\"message\\\":\\\"Request exceeds the maximum size\\\"}}\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - OpenAI/LiteLLM: \"Requested token count exceeds the model's maximum context length of 131072 tokens\"\n * - OpenAI-compatible: \"Input length (265330) exceeds model's maximum context length (262144).\"\n * - Google: \"The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)\"\n * - xAI: \"This model's maximum prompt length is 131072 but the request contains 537812 tokens\"\n * - Groq: \"Please reduce the length of the messages or completion\"\n * - OpenRouter: \"This endpoint's maximum context length is X tokens. However, you requested about Y tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"the request exceeds the available context size, try increasing it\"\n * - LM Studio: \"tokens to keep from the initial prompt is greater than the context length\"\n * - GitHub Copilot: \"prompt token count of X exceeds the limit of Y\"\n * - MiniMax: \"invalid params, context window exceeds limit\"\n * - Kimi For Coding: \"Your request exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - Cerebras: \"400/413 status code (no body)\"\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Xiaomi MiMo: Truncates input to fill contextWindow exactly, then returns finish_reason \"length\"\n *   with output=0 (no room left to generate). Detected via stopReason \"length\" + zero output +\n *   input filling the context window.\n * - DashScope/Qwen: \"Range of input length should be [1, X]\" (HTTP 400 invalid_parameter_error)\n * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n */\nconst OVERFLOW_PATTERNS = [\n\t/prompt is too long/i, // Anthropic token overflow\n\t/request_too_large/i, // Anthropic request byte-size overflow (HTTP 413)\n\t/input is too long for requested model/i, // Amazon Bedrock\n\t/exceeds the context window/i, // OpenAI (Completions & Responses API)\n\t/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\\d,]+ tokens?|\\s*\\([\\d,]+\\))/i, // OpenAI-compatible proxies (LiteLLM)\n\t/input token count.*exceeds the maximum/i, // Google (Gemini)\n\t/maximum prompt length is \\d+/i, // xAI (Grok)\n\t/reduce the length of the messages/i, // Groq\n\t/maximum context length is \\d+ tokens/i, // OpenRouter (most backends)\n\t/exceeds (?:the )?maximum allowed input length of [\\d,]+ tokens?/i, // OpenRouter/Poolside\n\t/input \\(\\d+ tokens\\) is longer than the model'?s context length \\(\\d+ tokens\\)/i, // Together AI\n\t/exceeds the limit of \\d+/i, // GitHub Copilot\n\t/exceeds the available context size/i, // llama.cpp server\n\t/greater than the context length/i, // LM Studio\n\t/context window exceeds limit/i, // MiniMax\n\t/exceeded model token limit/i, // Kimi For Coding\n\t/too large for model with \\d+ maximum context length/i, // Mistral\n\t/prompt has [\\d,]+ tokens?, but the configured context size is [\\d,]+ tokens?/i, // DS4 server\n\t/model_context_window_exceeded/i, // z.ai non-standard finish_reason surfaced as error text\n\t/prompt too long; exceeded (?:max )?context length/i, // Ollama explicit overflow error\n\t/range of input length should be/i, // DashScope / Qwen Token Plan\n\t/context[_ ]length[_ ]exceeded/i, // Generic fallback\n\t/too many tokens/i, // Generic fallback\n\t/token limit exceeded/i, // Generic fallback\n\t/^4(?:00|13)\\s*(?:status code)?\\s*\\(no body\\)/i, // Cerebras: 400/413 with no body\n];\n\n/**\n * Patterns that indicate non-overflow errors (e.g. rate limiting, server errors).\n * Error messages matching any of these are excluded from overflow detection\n * even if they also match an OVERFLOW_PATTERN.\n *\n * Example: Bedrock formats throttling errors as \"ThrottlingException: Too many tokens,\n * please wait before trying again.\" which would match the /too many tokens/i overflow\n * pattern without this exclusion.\n */\nconst NON_OVERFLOW_PATTERNS = [\n\t/^(Throttling error|Service unavailable):/i, // AWS Bedrock non-overflow errors (human-readable prefixes from formatBedrockError)\n\t/rate limit/i, // Generic rate limiting\n\t/too many requests/i, // Generic HTTP 429 style\n];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles two cases:\n * 1. Error-based overflow: Most providers return stopReason \"error\" with a\n *    specific error message pattern.\n * 2. Silent overflow: Some providers accept overflow requests and return\n *    successfully. For these, we check if usage.input exceeds the context window.\n *\n * ## Reliability by Provider\n *\n * **Reliable detection (returns error with detectable message):**\n * - Anthropic: \"prompt is too long: X tokens > Y maximum\" or \"request_too_large\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\", \"exceeds the model's maximum context length of X tokens\", or \"exceeds model's maximum context length (X)\"\n * - Google Gemini: \"input token count exceeds the maximum\"\n * - xAI (Grok): \"maximum prompt length is X but request contains Y\"\n * - Groq: \"reduce the length of the messages\"\n * - Cerebras: 400/413 status code (no body)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - OpenRouter (most backends): \"maximum context length is X tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n * - Kimi For Coding: \"exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - DashScope/Qwen: \"Range of input length should be [1, X]\"\n *\n * **Unreliable detection:**\n * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),\n *   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.\n * - Xiaomi MiMo: Truncates input to fit contextWindow then returns stopReason \"length\" with\n *   output=0. Pass contextWindow param to detect via the \"filled context + zero output\" signal.\n * - Ollama: May truncate input silently for some setups, but may also return explicit\n *   overflow errors that match the patterns above. Silent truncation still cannot be\n *   detected here because we do not know the expected token count.\n *\n * ## Custom Providers\n *\n * If you've added custom models via settings.json, this function may not detect\n * overflow errors from those providers. To add support:\n *\n * 1. Send a request that exceeds the model's context window\n * 2. Check the errorMessage in the response\n * 3. Create a regex pattern that matches the error\n * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or\n *    check the errorMessage yourself before calling this function\n *\n * @param message - The assistant message to check\n * @param contextWindow - Optional context window size for detecting silent overflow (z.ai)\n * @returns true if the message indicates a context overflow\n */\nexport function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean {\n\t// Case 1: Check error message patterns\n\tif (message.stopReason === \"error\" && message.errorMessage) {\n\t\t// Skip messages matching known non-overflow patterns (e.g. throttling / rate-limit)\n\t\tconst isNonOverflow = NON_OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!));\n\t\tif (!isNonOverflow && OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 2: Silent overflow (z.ai style) - successful but usage exceeds context\n\tif (contextWindow && message.stopReason === \"stop\") {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens > contextWindow) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 3: Length-stop overflow (Xiaomi MiMo style) - server truncates oversized input\n\t// to fit the context window, leaving no room for output. Returns stopReason \"length\"\n\t// with output=0 and input+cacheRead filling the context window.\n\tif (contextWindow && message.stopReason === \"length\" && message.usage.output === 0) {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens >= contextWindow * 0.99) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n"]}
+{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AACH,MAAM,iBAAiB,GAAG;IACzB,qBAAqB,EAAE,2BAA2B;IAClD,oBAAoB,EAAE,kDAAkD;IACxE,wCAAwC,EAAE,iBAAiB;IAC3D,6BAA6B,EAAE,uCAAuC;IACtE,4FAA4F,EAAE,sCAAsC;IACpI,yCAAyC,EAAE,kBAAkB;IAC7D,+BAA+B,EAAE,aAAa;IAC9C,oCAAoC,EAAE,OAAO;IAC7C,uCAAuC,EAAE,6BAA6B;IACtE,kEAAkE,EAAE,sBAAsB;IAC1F,iFAAiF,EAAE,cAAc;IACjG,2BAA2B,EAAE,iBAAiB;IAC9C,qCAAqC,EAAE,mBAAmB;IAC1D,kCAAkC,EAAE,YAAY;IAChD,+BAA+B,EAAE,UAAU;IAC3C,6BAA6B,EAAE,kBAAkB;IACjD,sDAAsD,EAAE,UAAU;IAClE,+EAA+E,EAAE,aAAa;IAC9F,gCAAgC,EAAE,yDAAyD;IAC3F,oDAAoD,EAAE,iCAAiC;IACvF,kCAAkC,EAAE,8BAA8B;IAClE,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;IAC5C,+CAA+C,EAAE,iCAAiC;CAClF,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,qBAAqB,GAAG;IAC7B,2CAA2C,EAAE,oFAAoF;IACjI,aAAa,EAAE,wBAAwB;IACvC,oBAAoB,EAAE,yBAAyB;CAC/C,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqDG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB;IAClF,uCAAuC;IACvC,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QAC5D,oFAAoF;QACpF,MAAM,aAAa,GAAG,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,CAAC;QACvF,IAAI,CAAC,aAAa,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,EAAE,CAAC;YACpF,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,8EAA8E;IAC9E,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,MAAM,EAAE,CAAC;QACpD,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,GAAG,aAAa,EAAE,CAAC;YACjC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,sFAAsF;IACtF,qFAAqF;IACrF,gEAAgE;IAChE,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpF,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,IAAI,aAAa,GAAG,IAAI,EAAE,CAAC;YACzC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AACd,CAAC;AACD;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,OAAyB,EAAE,gBAAwB;IACtF,OAAO,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,gBAAgB,GAAG,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,gBAAgB,CAAC;AAC3G,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB;IAClC,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAC/B,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\n/**\n * Regex patterns to detect context overflow errors from different providers.\n *\n * These patterns match error messages returned when the input exceeds\n * the model's context window.\n *\n * Provider-specific patterns (with example error messages):\n *\n * - Anthropic: \"prompt is too long: 213462 tokens > 200000 maximum\"\n * - Anthropic: \"413 {\\\"error\\\":{\\\"type\\\":\\\"request_too_large\\\",\\\"message\\\":\\\"Request exceeds the maximum size\\\"}}\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - OpenAI/LiteLLM: \"Requested token count exceeds the model's maximum context length of 131072 tokens\"\n * - OpenAI-compatible: \"Input length (265330) exceeds model's maximum context length (262144).\"\n * - Google: \"The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)\"\n * - xAI: \"This model's maximum prompt length is 131072 but the request contains 537812 tokens\"\n * - Groq: \"Please reduce the length of the messages or completion\"\n * - OpenRouter: \"This endpoint's maximum context length is X tokens. However, you requested about Y tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"the request exceeds the available context size, try increasing it\"\n * - LM Studio: \"tokens to keep from the initial prompt is greater than the context length\"\n * - GitHub Copilot: \"prompt token count of X exceeds the limit of Y\"\n * - MiniMax: \"invalid params, context window exceeds limit\"\n * - Kimi For Coding: \"Your request exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - Cerebras: \"400/413 status code (no body)\"\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Xiaomi MiMo: Truncates input to fill contextWindow exactly, then returns finish_reason \"length\"\n *   with output=0 (no room left to generate). Detected via stopReason \"length\" + zero output +\n *   input filling the context window.\n * - DashScope/Qwen: \"Range of input length should be [1, X]\" (HTTP 400 invalid_parameter_error)\n * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n */\nconst OVERFLOW_PATTERNS = [\n\t/prompt is too long/i, // Anthropic token overflow\n\t/request_too_large/i, // Anthropic request byte-size overflow (HTTP 413)\n\t/input is too long for requested model/i, // Amazon Bedrock\n\t/exceeds the context window/i, // OpenAI (Completions & Responses API)\n\t/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\\d,]+ tokens?|\\s*\\([\\d,]+\\))/i, // OpenAI-compatible proxies (LiteLLM)\n\t/input token count.*exceeds the maximum/i, // Google (Gemini)\n\t/maximum prompt length is \\d+/i, // xAI (Grok)\n\t/reduce the length of the messages/i, // Groq\n\t/maximum context length is \\d+ tokens/i, // OpenRouter (most backends)\n\t/exceeds (?:the )?maximum allowed input length of [\\d,]+ tokens?/i, // OpenRouter/Poolside\n\t/input \\(\\d+ tokens\\) is longer than the model'?s context length \\(\\d+ tokens\\)/i, // Together AI\n\t/exceeds the limit of \\d+/i, // GitHub Copilot\n\t/exceeds the available context size/i, // llama.cpp server\n\t/greater than the context length/i, // LM Studio\n\t/context window exceeds limit/i, // MiniMax\n\t/exceeded model token limit/i, // Kimi For Coding\n\t/too large for model with \\d+ maximum context length/i, // Mistral\n\t/prompt has [\\d,]+ tokens?, but the configured context size is [\\d,]+ tokens?/i, // DS4 server\n\t/model_context_window_exceeded/i, // z.ai non-standard finish_reason surfaced as error text\n\t/prompt too long; exceeded (?:max )?context length/i, // Ollama explicit overflow error\n\t/range of input length should be/i, // DashScope / Qwen Token Plan\n\t/context[_ ]length[_ ]exceeded/i, // Generic fallback\n\t/too many tokens/i, // Generic fallback\n\t/token limit exceeded/i, // Generic fallback\n\t/^4(?:00|13)\\s*(?:status code)?\\s*\\(no body\\)/i, // Cerebras: 400/413 with no body\n];\n\n/**\n * Patterns that indicate non-overflow errors (e.g. rate limiting, server errors).\n * Error messages matching any of these are excluded from overflow detection\n * even if they also match an OVERFLOW_PATTERN.\n *\n * Example: Bedrock formats throttling errors as \"ThrottlingException: Too many tokens,\n * please wait before trying again.\" which would match the /too many tokens/i overflow\n * pattern without this exclusion.\n */\nconst NON_OVERFLOW_PATTERNS = [\n\t/^(Throttling error|Service unavailable):/i, // AWS Bedrock non-overflow errors (human-readable prefixes from formatBedrockError)\n\t/rate limit/i, // Generic rate limiting\n\t/too many requests/i, // Generic HTTP 429 style\n];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles three cases:\n * 1. Error-based overflow: Most providers return stopReason \"error\" with a\n *    specific error message pattern.\n * 2. Silent overflow: Some providers accept overflow requests and return\n *    successfully. For these, we check if usage.input exceeds the context window.\n * 3. Length-stop overflow: Xiaomi MiMo can return \"length\" with zero output when\n *    the input fills the context window.\n *\n * ## Reliability by Provider\n *\n * **Reliable detection (returns error with detectable message):**\n * - Anthropic: \"prompt is too long: X tokens > Y maximum\" or \"request_too_large\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\", \"exceeds the model's maximum context length of X tokens\", or \"exceeds model's maximum context length (X)\"\n * - Google Gemini: \"input token count exceeds the maximum\"\n * - xAI (Grok): \"maximum prompt length is X but request contains Y\"\n * - Groq: \"reduce the length of the messages\"\n * - Cerebras: 400/413 status code (no body)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - OpenRouter (most backends): \"maximum context length is X tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n * - Kimi For Coding: \"exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - DashScope/Qwen: \"Range of input length should be [1, X]\"\n *\n * **Unreliable detection:**\n * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),\n *   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.\n * - Xiaomi MiMo: Truncates input to fit contextWindow then returns stopReason \"length\" with\n *   output=0. Pass contextWindow param to detect via the \"filled context + zero output\" signal.\n * - Ollama: May truncate input silently for some setups, but may also return explicit\n *   overflow errors that match the patterns above. Silent truncation still cannot be\n *   detected here because we do not know the expected token count.\n *\n * ## Custom Providers\n *\n * If you've added custom models via settings.json, this function may not detect\n * overflow errors from those providers. To add support:\n *\n * 1. Send a request that exceeds the model's context window\n * 2. Check the errorMessage in the response\n * 3. Create a regex pattern that matches the error\n * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or\n *    check the errorMessage yourself before calling this function\n *\n * @param message - The assistant message to check\n * @param contextWindow - Optional context window size for detecting silent overflow (z.ai)\n * @returns true if the message indicates a context overflow\n */\nexport function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean {\n\t// Case 1: Check error message patterns\n\tif (message.stopReason === \"error\" && message.errorMessage) {\n\t\t// Skip messages matching known non-overflow patterns (e.g. throttling / rate-limit)\n\t\tconst isNonOverflow = NON_OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!));\n\t\tif (!isNonOverflow && OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 2: Silent overflow (z.ai style) - successful but usage exceeds context\n\tif (contextWindow && message.stopReason === \"stop\") {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens > contextWindow) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 3: Length-stop overflow (Xiaomi MiMo style) - server truncates oversized input\n\t// to fit the context window, leaving no room for output. Returns stopReason \"length\"\n\t// with output=0 and input+cacheRead filling the context window.\n\tif (contextWindow && message.stopReason === \"length\" && message.usage.output === 0) {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens >= contextWindow * 0.99) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n/**\n * Check whether a length stop ended below the caller or model's intended output limit.\n * Such responses may be caused by context pressure or provider-side truncation, so callers\n * can make one bounded compact-and-retry attempt. `desiredMaxOutput` must be the original\n * limit before any context-based clamping.\n */\nexport function isRecoverableLength(message: AssistantMessage, desiredMaxOutput: number): boolean {\n\treturn message.stopReason === \"length\" && desiredMaxOutput > 0 && message.usage.output < desiredMaxOutput;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n"]}

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

@@ -3,12 +3,14 @@ export declare const PROMPT_CACHE_TTL_SHORT_SECONDS = 300;
 export declare const PROMPT_CACHE_TTL_LONG_SECONDS = 3600;
 export declare function isAnthropicApiBaseUrl(baseUrl: string): boolean;
 export declare function getAnthropicCompat(model: Model<"anthropic-messages">): Required<Omit<AnthropicMessagesCompat, "forceAdaptiveThinking">>;
-export type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "cacheControlFormat" | "toolCallFormat" | "deferredToolsMode" | "toolSchemaFlavor" | "supportsPromptCacheKey"> & {
+export type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "cacheControlFormat" | "toolCallFormat" | "deferredToolsMode" | "toolSchemaFlavor" | "supportsPromptCacheKey" | "chatTemplateArgs" | "supportsThinkingTokenBudget"> & {
     cacheControlFormat?: OpenAICompletionsCompat["cacheControlFormat"];
     supportsPromptCacheKey?: OpenAICompletionsCompat["supportsPromptCacheKey"];
     toolCallFormat?: OpenAICompletionsCompat["toolCallFormat"];
     deferredToolsMode?: OpenAICompletionsCompat["deferredToolsMode"];
     toolSchemaFlavor?: OpenAICompletionsCompat["toolSchemaFlavor"];
+    chatTemplateArgs?: OpenAICompletionsCompat["chatTemplateArgs"];
+    supportsThinkingTokenBudget?: OpenAICompletionsCompat["supportsThinkingTokenBudget"];
 };
 /**
  * Get resolved compatibility settings for a model.

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

@@ -1 +1 @@
-{"version":3,"file":"prompt-cache-ttl.d.ts","sourceRoot":"","sources":["../../src/utils/prompt-cache-ttl.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,uBAAuB,EACvB,GAAG,EAEH,KAAK,EACL,uBAAuB,EACvB,WAAW,EACX,MAAM,aAAa,CAAC;AAGrB,eAAO,MAAM,8BAA8B,MAAM,CAAC;AAClD,eAAO,MAAM,6BAA6B,OAAO,CAAC;AAElD,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAM9D;AAkBD,wBAAgB,kBAAkB,CACjC,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAChC,QAAQ,CAAC,IAAI,CAAC,uBAAuB,EAAE,uBAAuB,CAAC,CAAC,CA6BlE;AAED,MAAM,MAAM,+BAA+B,GAAG,IAAI,CACjD,QAAQ,CAAC,uBAAuB,CAAC,EACjC,oBAAoB,GAAG,gBAAgB,GAAG,mBAAmB,GAAG,kBAAkB,GAAG,wBAAwB,CAC7G,GAAG;IACH,kBAAkB,CAAC,EAAE,uBAAuB,CAAC,oBAAoB,CAAC,CAAC;IACnE,sBAAsB,CAAC,EAAE,uBAAuB,CAAC,wBAAwB,CAAC,CAAC;IAC3E,cAAc,CAAC,EAAE,uBAAuB,CAAC,gBAAgB,CAAC,CAAC;IAC3D,iBAAiB,CAAC,EAAE,uBAAuB,CAAC,mBAAmB,CAAC,CAAC;IACjE,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;CAC/D,CAAC;AA6GF;;;GAGG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAAG,+BAA+B,CAmC9G;AAED,wBAAgB,8BAA8B,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAM5F;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,GAAG,OAAO,CAKxF;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,OAAO,CAmBzG;AAiDD,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CA4CrG"}
+{"version":3,"file":"prompt-cache-ttl.d.ts","sourceRoot":"","sources":["../../src/utils/prompt-cache-ttl.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,uBAAuB,EACvB,GAAG,EAEH,KAAK,EACL,uBAAuB,EACvB,WAAW,EACX,MAAM,aAAa,CAAC;AAGrB,eAAO,MAAM,8BAA8B,MAAM,CAAC;AAClD,eAAO,MAAM,6BAA6B,OAAO,CAAC;AAElD,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAM9D;AAkBD,wBAAgB,kBAAkB,CACjC,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAChC,QAAQ,CAAC,IAAI,CAAC,uBAAuB,EAAE,uBAAuB,CAAC,CAAC,CA6BlE;AAED,MAAM,MAAM,+BAA+B,GAAG,IAAI,CACjD,QAAQ,CAAC,uBAAuB,CAAC,EAC/B,oBAAoB,GACpB,gBAAgB,GAChB,mBAAmB,GACnB,kBAAkB,GAClB,wBAAwB,GACxB,kBAAkB,GAClB,6BAA6B,CAC/B,GAAG;IACH,kBAAkB,CAAC,EAAE,uBAAuB,CAAC,oBAAoB,CAAC,CAAC;IACnE,sBAAsB,CAAC,EAAE,uBAAuB,CAAC,wBAAwB,CAAC,CAAC;IAC3E,cAAc,CAAC,EAAE,uBAAuB,CAAC,gBAAgB,CAAC,CAAC;IAC3D,iBAAiB,CAAC,EAAE,uBAAuB,CAAC,mBAAmB,CAAC,CAAC;IACjE,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;IAC/D,gBAAgB,CAAC,EAAE,uBAAuB,CAAC,kBAAkB,CAAC,CAAC;IAC/D,2BAA2B,CAAC,EAAE,uBAAuB,CAAC,6BAA6B,CAAC,CAAC;CACrF,CAAC;AA6GF;;;GAGG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,KAAK,CAAC,oBAAoB,CAAC,GAAG,+BAA+B,CAqC9G;AAED,wBAAgB,8BAA8B,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAM5F;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,GAAG,OAAO,CAKxF;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,OAAO,CAmBzG;AAiDD,wBAAgB,4BAA4B,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CA4CrG"}

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

@@ -167,7 +167,9 @@ export function getOpenAICompletionsCompat(model) {
         openRouterRouting: model.compat.openRouterRouting ?? detected.openRouterRouting,
         vercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,
         chatTemplateKwargs: model.compat.chatTemplateKwargs ?? detected.chatTemplateKwargs,
+        chatTemplateArgs: model.compat.chatTemplateArgs ?? detected.chatTemplateArgs,
         zaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,
+        supportsThinkingTokenBudget: model.compat.supportsThinkingTokenBudget ?? detected.supportsThinkingTokenBudget,
         supportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,
         toolSchemaFlavor: model.compat.toolSchemaFlavor ?? detected.toolSchemaFlavor,
         toolCallFormat: model.compat.toolCallFormat ?? detected.toolCallFormat,

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

@@ -1 +1 @@
-{"version":3,"file":"prompt-cache-ttl.js","sourceRoot":"","sources":["../../src/utils/prompt-cache-ttl.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AAExD,MAAM,CAAC,MAAM,8BAA8B,GAAG,GAAG,CAAC;AAClD,MAAM,CAAC,MAAM,6BAA6B,GAAG,IAAI,CAAC;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;AAaD;;;;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;QAC7B,UAAU;QACV,qBAAqB;QACrB,UAAU;QACV,QAAQ;QACR,SAAS;QACT,KAAK,CAAC;IAEP,MAAM,MAAM,GAAG,QAAQ,KAAK,KAAK,IAAI,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IAClE,MAAM,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,8BAA8B,GAAG,CAAC,YAAY,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IAC1E,MAAM,mBAAmB,GAAG,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC;IACpF,MAAM,8BAA8B,GAAG,8BAA8B,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CACrF,mBAAmB,CAAC,UAAU,CAAC,MAAM,CAAC,CACtC,CAAC;IACF,MAAM,kBAAkB,GAAG,QAAQ,KAAK,YAAY,IAAI,8BAA8B,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IAEjH,OAAO;QACN,aAAa,EAAE,CAAC,aAAa;QAC7B,qBAAqB,EAAE,8BAA8B,IAAI,CAAC,CAAC,aAAa,IAAI,CAAC,YAAY,CAAC;QAC1F,uBAAuB,EACtB,CAAC,MAAM,IAAI,CAAC,KAAK,IAAI,CAAC,UAAU,IAAI,CAAC,UAAU,IAAI,CAAC,qBAAqB,IAAI,CAAC,QAAQ,IAAI,CAAC,SAAS;QACrG,wBAAwB,EAAE,IAAI;QAC9B,oBAAoB,EAAE,IAAI;QAC1B,cAAc,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,uBAAuB;QACrE,sBAAsB,EAAE,KAAK;QAC7B,gCAAgC,EAAE,KAAK;QACvC,sBAAsB,EAAE,KAAK;QAC7B,2CAA2C,EAAE,UAAU;QACvD,cAAc,EAAE,UAAU;YACzB,CAAC,CAAC,UAAU;YACZ,CAAC,CAAC,KAAK;gBACN,CAAC,CAAC,KAAK;gBACP,CAAC,CAAC,UAAU;oBACX,CAAC,CAAC,UAAU;oBACZ,CAAC,CAAC,SAAS;wBACV,CAAC,CAAC,UAAU;wBACZ,CAAC,CAAC,YAAY;4BACb,CAAC,CAAC,YAAY;4BACd,CAAC,CAAC,QAAQ;QACf,iBAAiB,EAAE,EAAE;QACrB,oBAAoB,EAAE,EAAE;QACxB,kBAAkB,EAAE,EAAE;QACtB,aAAa,EAAE,KAAK;QACpB,kBAAkB,EAAE,CAAC,UAAU,IAAI,CAAC,UAAU,IAAI,CAAC,qBAAqB,IAAI,CAAC,QAAQ;QACrF,gBAAgB,EAAE,UAAU,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,SAAS;QAC1D,wBAAwB,EAAE,IAAI;QAC9B,cAAc,EAAE,SAAS;QACzB,0BAA0B,EAAE,KAAK;QACjC,kBAAkB;QAClB,0BAA0B,EAAE,YAAY;QACxC,iBAAiB,EAAE,SAAS;QAC5B,qBAAqB,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ;QAC7D,sBAAsB,EAAE,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC;QACxE,0BAA0B,EAAE,CAAC,CAC5B,UAAU;YACV,qBAAqB;YACrB,qBAAqB;YACrB,QAAQ;YACR,SAAS,CACT;KACD,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,0BAA0B,CAAC,KAAkC;IAC5E,MAAM,QAAQ,GAAG,6BAA6B,CAAC,KAAK,CAAC,CAAC;IACtD,IAAI,CAAC,KAAK,CAAC,MAAM;QAAE,OAAO,QAAQ,CAAC;IAEnC,OAAO;QACN,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,qBAAqB,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,qBAAqB;QAC3F,uBAAuB,EAAE,KAAK,CAAC,MAAM,CAAC,uBAAuB,IAAI,QAAQ,CAAC,uBAAuB;QACjG,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,oBAAoB,EAAE,KAAK,CAAC,MAAM,CAAC,oBAAoB,IAAI,QAAQ,CAAC,oBAAoB;QACxF,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,gCAAgC,EAC/B,KAAK,CAAC,MAAM,CAAC,gCAAgC,IAAI,QAAQ,CAAC,gCAAgC;QAC3F,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,2CAA2C,EAC1C,KAAK,CAAC,MAAM,CAAC,2CAA2C;YACxD,QAAQ,CAAC,2CAA2C;QACrD,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,iBAAiB,IAAI,QAAQ,CAAC,iBAAiB;QAC/E,oBAAoB,EAAE,KAAK,CAAC,MAAM,CAAC,oBAAoB,IAAI,QAAQ,CAAC,oBAAoB;QACxF,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAC5E,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;QAC1G,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;QAC1G,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,iBAAiB,IAAI,QAAQ,CAAC,iBAAiB;QAC/E,qBAAqB,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,qBAAqB;QAC3F,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;KAC1G,CAAC;AACH,CAAC;AAED,MAAM,UAAU,8BAA8B,CAAC,OAAe,EAAE,SAAkB;IACjF,MAAM,MAAM,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;IAC5D,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE;QAC/B,MAAM,KAAK,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;QAClC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,CAAC;IACjD,CAAC,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAuC;IAC9E,MAAM,UAAU,GAAG,8BAA8B,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IACxE,OAAO,UAAU,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CACpC,CAAC,UAAU,EAAE,YAAY,EAAE,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,YAAY,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,YAAY,CAAC,CAAC,CAChG,CAAC;AACH,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,qBAAqB,CAAC,KAAuC,EAAE,GAAiB;IAC/F,MAAM,UAAU,GAAG,8BAA8B,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IAExE,MAAM,YAAY,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClE,IAAI,CAAC,YAAY,EAAE,CAAC;QACnB,0EAA0E;QAC1E,8DAA8D;QAC9D,IAAI,mBAAmB,CAAC,yBAAyB,EAAE,GAAG,CAAC,KAAK,GAAG;YAAE,OAAO,IAAI,CAAC;QAC7E,OAAO,KAAK,CAAC;IACd,CAAC;IACD,8CAA8C;IAC9C,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACjH,gDAAgD;IAChD,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IAC3D,oBAAoB;IACpB,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,mBAAmB,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACzE,mBAAmB;IACnB,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACxE,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,8BAA8B,CACtC,cAA+B,EAC/B,GAAiB,EACjB,QAAQ,GAAmB,OAAO;IAElC,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,OAAO,CAAC,GAAG,CAAC,kBAAkB,KAAK,SAAS,EAAE,CAAC;QACpF,OAAO,OAAO,CAAC;IAChB,CAAC;IACD,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED,SAAS,sCAAsC,CAAC,cAA+B,EAAE,GAAiB;IACjG,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,4BAA4B,CAAC,cAA+B,EAAE,GAAiB;IACvF,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,oCAAoC,CAAC,cAA+B,EAAE,GAAiB;IAC/F,IAAI,cAAc,EAAE,CAAC;QACpB,OAAO,cAAc,CAAC;IACvB,CAAC;IACD,IAAI,mBAAmB,CAAC,oBAAoB,EAAE,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC;QAC/D,OAAO,MAAM,CAAC;IACf,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,4BAA4B,CAAC,KAAiB,EAAE,GAAiB;IAChF,QAAQ,KAAK,CAAC,GAAG,EAAE,CAAC;QACnB,KAAK,kBAAkB;YACtB,wFAAwF;YACxF,OAAO,8BAA8B,CAAC;QACvC,KAAK,oBAAoB,EAAE,CAAC;YAC3B,MAAM,cAAc,GAAG,KAAoC,CAAC;YAC5D,MAAM,SAAS,GAAG,8BAA8B,CAAC,cAAc,CAAC,cAAc,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;YAC9F,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,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,IAAI,uBAAuB,CAAC,YAAY,CAAC;gBACnE,CAAC,CAAC,6BAA6B;gBAC/B,CAAC,CAAC,8BAA8B,CAAC;QACnC,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\" | \"supportsPromptCacheKey\"\n> & {\n\tcacheControlFormat?: OpenAICompletionsCompat[\"cacheControlFormat\"];\n\tsupportsPromptCacheKey?: OpenAICompletionsCompat[\"supportsPromptCacheKey\"];\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\") ||\n\t\tisMoonshot ||\n\t\tisCloudflareAiGateway ||\n\t\tisTogether ||\n\t\tisNvidia ||\n\t\tisAntLing ||\n\t\tisZai;\n\n\tconst isGrok = provider === \"xai\" || baseUrl.includes(\"api.x.ai\");\n\tconst isDeepSeek = provider === \"deepseek\" || baseUrl.includes(\"deepseek.com\");\n\tconst isOpenRouterDeveloperRoleModel =\n\t\tisOpenRouter && (model.id.startsWith(\"anthropic/\") || model.id.startsWith(\"openai/\"));\n\tconst openRouterCacheControlPrefixes = [\"anthropic/\", \"qwen/\", \"google/\"];\n\tconst cacheControlModelId = model.id.startsWith(\"~\") ? model.id.slice(1) : model.id;\n\tconst supportsOpenRouterCacheControl = openRouterCacheControlPrefixes.some((prefix) =>\n\t\tcacheControlModelId.startsWith(prefix),\n\t);\n\tconst cacheControlFormat = provider === \"openrouter\" && supportsOpenRouterCacheControl ? \"anthropic\" : undefined;\n\n\treturn {\n\t\tsupportsStore: !isNonStandard,\n\t\tsupportsDeveloperRole: isOpenRouterDeveloperRoleModel || (!isNonStandard && !isOpenRouter),\n\t\tsupportsReasoningEffort:\n\t\t\t!isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia && !isAntLing,\n\t\tsupportsUsageInStreaming: true,\n\t\tsupportsFinishReason: true,\n\t\tmaxTokensField: useMaxTokens ? \"max_tokens\" : \"max_completion_tokens\",\n\t\trequiresToolResultName: false,\n\t\trequiresAssistantAfterToolResult: false,\n\t\trequiresThinkingAsText: false,\n\t\trequiresReasoningContentOnAssistantMessages: isDeepSeek,\n\t\tthinkingFormat: isDeepSeek\n\t\t\t? \"deepseek\"\n\t\t\t: isZai\n\t\t\t\t? \"zai\"\n\t\t\t\t: isTogether\n\t\t\t\t\t? \"together\"\n\t\t\t\t\t: isAntLing\n\t\t\t\t\t\t? \"ant-ling\"\n\t\t\t\t\t\t: isOpenRouter\n\t\t\t\t\t\t\t? \"openrouter\"\n\t\t\t\t\t\t\t: \"openai\",\n\t\topenRouterRouting: {},\n\t\tvercelGatewayRouting: {},\n\t\tchatTemplateKwargs: {},\n\t\tzaiToolStream: false,\n\t\tsupportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,\n\t\ttoolSchemaFlavor: isMoonshot ? \"moonshot-mfjs\" : undefined,\n\t\tsupportsDisabledThinking: true,\n\t\ttoolCallFormat: undefined,\n\t\tsupportsOpenAIGrammarTools: false,\n\t\tcacheControlFormat,\n\t\tsendSessionAffinityHeaders: isOpenRouter,\n\t\tdeferredToolsMode: undefined,\n\t\tsessionAffinityFormat: isOpenRouter ? \"openrouter\" : \"openai\",\n\t\tsupportsPromptCacheKey: isMoonshot || baseUrl.includes(\"api.openai.com\"),\n\t\tsupportsLongCacheRetention: !(\n\t\t\tisTogether ||\n\t\t\tisCloudflareWorkersAI ||\n\t\t\tisCloudflareAiGateway ||\n\t\t\tisNvidia ||\n\t\t\tisAntLing\n\t\t),\n\t};\n}\n\n/**\n * Get resolved compatibility settings for a model.\n * Auto-detects from provider/URL then overrides with explicit model.compat.\n */\nexport function getOpenAICompletionsCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst detected = detectOpenAICompletionsCompat(model);\n\tif (!model.compat) return detected;\n\n\treturn {\n\t\tsupportsStore: model.compat.supportsStore ?? detected.supportsStore,\n\t\tsupportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,\n\t\tsupportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,\n\t\tsupportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? detected.supportsUsageInStreaming,\n\t\tsupportsFinishReason: model.compat.supportsFinishReason ?? detected.supportsFinishReason,\n\t\tmaxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,\n\t\trequiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,\n\t\trequiresAssistantAfterToolResult:\n\t\t\tmodel.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,\n\t\trequiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,\n\t\trequiresReasoningContentOnAssistantMessages:\n\t\t\tmodel.compat.requiresReasoningContentOnAssistantMessages ??\n\t\t\tdetected.requiresReasoningContentOnAssistantMessages,\n\t\tthinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,\n\t\tsupportsDisabledThinking: model.compat.supportsDisabledThinking ?? detected.supportsDisabledThinking,\n\t\topenRouterRouting: model.compat.openRouterRouting ?? detected.openRouterRouting,\n\t\tvercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,\n\t\tchatTemplateKwargs: model.compat.chatTemplateKwargs ?? detected.chatTemplateKwargs,\n\t\tzaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,\n\t\tsupportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,\n\t\ttoolSchemaFlavor: model.compat.toolSchemaFlavor ?? detected.toolSchemaFlavor,\n\t\ttoolCallFormat: model.compat.toolCallFormat ?? detected.toolCallFormat,\n\t\tsupportsOpenAIGrammarTools: model.compat.supportsOpenAIGrammarTools ?? detected.supportsOpenAIGrammarTools,\n\t\tcacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,\n\t\tsendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,\n\t\tdeferredToolsMode: model.compat.deferredToolsMode ?? detected.deferredToolsMode,\n\t\tsessionAffinityFormat: model.compat.sessionAffinityFormat ?? detected.sessionAffinityFormat,\n\t\tsupportsPromptCacheKey: model.compat.supportsPromptCacheKey ?? detected.supportsPromptCacheKey,\n\t\tsupportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,\n\t};\n}\n\nexport function getBedrockModelMatchCandidates(modelId: string, modelName?: string): string[] {\n\tconst values = modelName ? [modelId, modelName] : [modelId];\n\treturn values.flatMap((value) => {\n\t\tconst lower = value.toLowerCase();\n\t\treturn [lower, lower.replace(/[\\s_.:]+/g, \"-\")];\n\t});\n}\n\nexport function supportsOneHourCacheTtl(model: Model<\"bedrock-converse-stream\">): boolean {\n\tconst candidates = getBedrockModelMatchCandidates(model.id, model.name);\n\treturn candidates.some((candidate) =>\n\t\t[\"opus-4-5\", \"sonnet-4-5\", \"haiku-4-5\"].some((modelVersion) => candidate.includes(modelVersion)),\n\t);\n}\n\n/**\n * Check if the model supports prompt caching.\n * Supported: Claude 3.5 Haiku, Claude 3.7 Sonnet, Claude 4.x models, Claude 5 models\n *\n * For base models and system-defined inference profiles the model ID / ARN\n * contains the model name, so we can decide locally.\n *\n * For application inference profiles (whose ARNs don't contain the model name),\n * also checks model.name which is user-controlled via models.json or registerProvider.\n * As a last resort, set AWS_BEDROCK_FORCE_CACHE=1 to enable cache points.\n * Amazon Nova models have automatic caching and don't need explicit cache points.\n */\nexport function supportsPromptCaching(model: Model<\"bedrock-converse-stream\">, env?: ProviderEnv): boolean {\n\tconst candidates = getBedrockModelMatchCandidates(model.id, model.name);\n\n\tconst hasClaudeRef = candidates.some((s) => s.includes(\"claude\"));\n\tif (!hasClaudeRef) {\n\t\t// Application inference profiles don't contain the model name in the ARN.\n\t\t// Allow users to force cache points via environment variable.\n\t\tif (getProviderEnvValue(\"AWS_BEDROCK_FORCE_CACHE\", env) === \"1\") return true;\n\t\treturn false;\n\t}\n\t// Claude 5 models (fable-5, opus-5, sonnet-5)\n\tif (candidates.some((s) => s.includes(\"fable-5\") || s.includes(\"opus-5\") || s.includes(\"sonnet-5\"))) return true;\n\t// Claude 4.x models (opus-4, sonnet-4, haiku-4)\n\tif (candidates.some((s) => s.includes(\"-4-\"))) return true;\n\t// Claude 3.7 Sonnet\n\tif (candidates.some((s) => s.includes(\"claude-3-7-sonnet\"))) return true;\n\t// Claude 3.5 Haiku\n\tif (candidates.some((s) => s.includes(\"claude-3-5-haiku\"))) return true;\n\treturn false;\n}\n\nfunction resolveAnthropicCacheRetention(\n\tcacheRetention?: CacheRetention,\n\tenv?: ProviderEnv,\n\tfallback: CacheRetention = \"short\",\n): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\tif (typeof process !== \"undefined\" && process.env.PI_CACHE_RETENTION !== undefined) {\n\t\treturn \"short\";\n\t}\n\treturn fallback;\n}\n\nfunction resolveOpenAICompletionsCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction resolveBedrockCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction resolveOpenAIResponsesCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nexport function resolvePromptCacheTtlSeconds(model: Model<Api>, env?: ProviderEnv): number | undefined {\n\tswitch (model.api) {\n\t\tcase \"claude-sdk-oauth\":\n\t\t\t// The Claude SDK owns prompt caching for this lane and uses Anthropic's default 5m TTL.\n\t\t\treturn PROMPT_CACHE_TTL_SHORT_SECONDS;\n\t\tcase \"anthropic-messages\": {\n\t\t\tconst anthropicModel = model as Model<\"anthropic-messages\">;\n\t\t\tconst retention = resolveAnthropicCacheRetention(anthropicModel.cacheRetention, env, \"short\");\n\t\t\tif (retention === \"none\") return 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\" && supportsOneHourCacheTtl(bedrockModel)\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 \"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"]}
+{"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;AAqBD;;;;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;QAC7B,UAAU;QACV,qBAAqB;QACrB,UAAU;QACV,QAAQ;QACR,SAAS;QACT,KAAK,CAAC;IAEP,MAAM,MAAM,GAAG,QAAQ,KAAK,KAAK,IAAI,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;IAClE,MAAM,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,8BAA8B,GAAG,CAAC,YAAY,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;IAC1E,MAAM,mBAAmB,GAAG,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC;IACpF,MAAM,8BAA8B,GAAG,8BAA8B,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CACrF,mBAAmB,CAAC,UAAU,CAAC,MAAM,CAAC,CACtC,CAAC;IACF,MAAM,kBAAkB,GAAG,QAAQ,KAAK,YAAY,IAAI,8BAA8B,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IAEjH,OAAO;QACN,aAAa,EAAE,CAAC,aAAa;QAC7B,qBAAqB,EAAE,8BAA8B,IAAI,CAAC,CAAC,aAAa,IAAI,CAAC,YAAY,CAAC;QAC1F,uBAAuB,EACtB,CAAC,MAAM,IAAI,CAAC,KAAK,IAAI,CAAC,UAAU,IAAI,CAAC,UAAU,IAAI,CAAC,qBAAqB,IAAI,CAAC,QAAQ,IAAI,CAAC,SAAS;QACrG,wBAAwB,EAAE,IAAI;QAC9B,oBAAoB,EAAE,IAAI;QAC1B,cAAc,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,uBAAuB;QACrE,sBAAsB,EAAE,KAAK;QAC7B,gCAAgC,EAAE,KAAK;QACvC,sBAAsB,EAAE,KAAK;QAC7B,2CAA2C,EAAE,UAAU;QACvD,cAAc,EAAE,UAAU;YACzB,CAAC,CAAC,UAAU;YACZ,CAAC,CAAC,KAAK;gBACN,CAAC,CAAC,KAAK;gBACP,CAAC,CAAC,UAAU;oBACX,CAAC,CAAC,UAAU;oBACZ,CAAC,CAAC,SAAS;wBACV,CAAC,CAAC,UAAU;wBACZ,CAAC,CAAC,YAAY;4BACb,CAAC,CAAC,YAAY;4BACd,CAAC,CAAC,QAAQ;QACf,iBAAiB,EAAE,EAAE;QACrB,oBAAoB,EAAE,EAAE;QACxB,kBAAkB,EAAE,EAAE;QACtB,aAAa,EAAE,KAAK;QACpB,kBAAkB,EAAE,CAAC,UAAU,IAAI,CAAC,UAAU,IAAI,CAAC,qBAAqB,IAAI,CAAC,QAAQ;QACrF,gBAAgB,EAAE,UAAU,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,SAAS;QAC1D,wBAAwB,EAAE,IAAI;QAC9B,cAAc,EAAE,SAAS;QACzB,0BAA0B,EAAE,KAAK;QACjC,kBAAkB;QAClB,0BAA0B,EAAE,YAAY;QACxC,iBAAiB,EAAE,SAAS;QAC5B,qBAAqB,EAAE,YAAY,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,QAAQ;QAC7D,sBAAsB,EAAE,UAAU,IAAI,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC;QACxE,0BAA0B,EAAE,CAAC,CAC5B,UAAU;YACV,qBAAqB;YACrB,qBAAqB;YACrB,QAAQ;YACR,SAAS,CACT;KACD,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,0BAA0B,CAAC,KAAkC;IAC5E,MAAM,QAAQ,GAAG,6BAA6B,CAAC,KAAK,CAAC,CAAC;IACtD,IAAI,CAAC,KAAK,CAAC,MAAM;QAAE,OAAO,QAAQ,CAAC;IAEnC,OAAO;QACN,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,qBAAqB,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,qBAAqB;QAC3F,uBAAuB,EAAE,KAAK,CAAC,MAAM,CAAC,uBAAuB,IAAI,QAAQ,CAAC,uBAAuB;QACjG,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,oBAAoB,EAAE,KAAK,CAAC,MAAM,CAAC,oBAAoB,IAAI,QAAQ,CAAC,oBAAoB;QACxF,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,gCAAgC,EAC/B,KAAK,CAAC,MAAM,CAAC,gCAAgC,IAAI,QAAQ,CAAC,gCAAgC;QAC3F,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,2CAA2C,EAC1C,KAAK,CAAC,MAAM,CAAC,2CAA2C;YACxD,QAAQ,CAAC,2CAA2C;QACrD,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,wBAAwB,EAAE,KAAK,CAAC,MAAM,CAAC,wBAAwB,IAAI,QAAQ,CAAC,wBAAwB;QACpG,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,iBAAiB,IAAI,QAAQ,CAAC,iBAAiB;QAC/E,oBAAoB,EAAE,KAAK,CAAC,MAAM,CAAC,oBAAoB,IAAI,QAAQ,CAAC,oBAAoB;QACxF,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAC5E,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QACnE,2BAA2B,EAAE,KAAK,CAAC,MAAM,CAAC,2BAA2B,IAAI,QAAQ,CAAC,2BAA2B;QAC7G,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,gBAAgB,EAAE,KAAK,CAAC,MAAM,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAC5E,cAAc,EAAE,KAAK,CAAC,MAAM,CAAC,cAAc,IAAI,QAAQ,CAAC,cAAc;QACtE,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;QAC1G,kBAAkB,EAAE,KAAK,CAAC,MAAM,CAAC,kBAAkB,IAAI,QAAQ,CAAC,kBAAkB;QAClF,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;QAC1G,iBAAiB,EAAE,KAAK,CAAC,MAAM,CAAC,iBAAiB,IAAI,QAAQ,CAAC,iBAAiB;QAC/E,qBAAqB,EAAE,KAAK,CAAC,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,qBAAqB;QAC3F,sBAAsB,EAAE,KAAK,CAAC,MAAM,CAAC,sBAAsB,IAAI,QAAQ,CAAC,sBAAsB;QAC9F,0BAA0B,EAAE,KAAK,CAAC,MAAM,CAAC,0BAA0B,IAAI,QAAQ,CAAC,0BAA0B;KAC1G,CAAC;AACH,CAAC;AAED,MAAM,UAAU,8BAA8B,CAAC,OAAe,EAAE,SAAkB;IACjF,MAAM,MAAM,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;IAC5D,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE;QAC/B,MAAM,KAAK,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;QAClC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,OAAO,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,CAAC;IACjD,CAAC,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAuC;IAC9E,MAAM,UAAU,GAAG,8BAA8B,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IACxE,OAAO,UAAU,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CACpC,CAAC,UAAU,EAAE,YAAY,EAAE,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,YAAY,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,YAAY,CAAC,CAAC,CAChG,CAAC;AACH,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,qBAAqB,CAAC,KAAuC,EAAE,GAAiB;IAC/F,MAAM,UAAU,GAAG,8BAA8B,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IAExE,MAAM,YAAY,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;IAClE,IAAI,CAAC,YAAY,EAAE,CAAC;QACnB,0EAA0E;QAC1E,8DAA8D;QAC9D,IAAI,mBAAmB,CAAC,yBAAyB,EAAE,GAAG,CAAC,KAAK,GAAG;YAAE,OAAO,IAAI,CAAC;QAC7E,OAAO,KAAK,CAAC;IACd,CAAC;IACD,8CAA8C;IAC9C,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACjH,gDAAgD;IAChD,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IAC3D,oBAAoB;IACpB,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,mBAAmB,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACzE,mBAAmB;IACnB,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACxE,OAAO,KAAK,CAAC;AACd,CAAC;AAED,SAAS,8BAA8B,CACtC,cAA+B,EAC/B,GAAiB,EACjB,WAA2B,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,kBAAkB;YACtB,wFAAwF;YACxF,OAAO,8BAA8B,CAAC;QACvC,KAAK,oBAAoB,CAAC,CAAC,CAAC;YAC3B,MAAM,cAAc,GAAG,KAAoC,CAAC;YAC5D,MAAM,SAAS,GAAG,8BAA8B,CAAC,cAAc,CAAC,cAAc,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;YAC9F,IAAI,SAAS,KAAK,MAAM;gBAAE,OAAO,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,CAAC,CAAC,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,IAAI,uBAAuB,CAAC,YAAY,CAAC;gBACnE,CAAC,CAAC,6BAA6B;gBAC/B,CAAC,CAAC,8BAA8B,CAAC;QACnC,CAAC;QACD,KAAK,oBAAoB,CAAC,CAAC,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,CAAC,CAAC,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\"\n\t| \"toolCallFormat\"\n\t| \"deferredToolsMode\"\n\t| \"toolSchemaFlavor\"\n\t| \"supportsPromptCacheKey\"\n\t| \"chatTemplateArgs\"\n\t| \"supportsThinkingTokenBudget\"\n> & {\n\tcacheControlFormat?: OpenAICompletionsCompat[\"cacheControlFormat\"];\n\tsupportsPromptCacheKey?: OpenAICompletionsCompat[\"supportsPromptCacheKey\"];\n\ttoolCallFormat?: OpenAICompletionsCompat[\"toolCallFormat\"];\n\tdeferredToolsMode?: OpenAICompletionsCompat[\"deferredToolsMode\"];\n\ttoolSchemaFlavor?: OpenAICompletionsCompat[\"toolSchemaFlavor\"];\n\tchatTemplateArgs?: OpenAICompletionsCompat[\"chatTemplateArgs\"];\n\tsupportsThinkingTokenBudget?: OpenAICompletionsCompat[\"supportsThinkingTokenBudget\"];\n};\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\") ||\n\t\tisMoonshot ||\n\t\tisCloudflareAiGateway ||\n\t\tisTogether ||\n\t\tisNvidia ||\n\t\tisAntLing ||\n\t\tisZai;\n\n\tconst isGrok = provider === \"xai\" || baseUrl.includes(\"api.x.ai\");\n\tconst isDeepSeek = provider === \"deepseek\" || baseUrl.includes(\"deepseek.com\");\n\tconst isOpenRouterDeveloperRoleModel =\n\t\tisOpenRouter && (model.id.startsWith(\"anthropic/\") || model.id.startsWith(\"openai/\"));\n\tconst openRouterCacheControlPrefixes = [\"anthropic/\", \"qwen/\", \"google/\"];\n\tconst cacheControlModelId = model.id.startsWith(\"~\") ? model.id.slice(1) : model.id;\n\tconst supportsOpenRouterCacheControl = openRouterCacheControlPrefixes.some((prefix) =>\n\t\tcacheControlModelId.startsWith(prefix),\n\t);\n\tconst cacheControlFormat = provider === \"openrouter\" && supportsOpenRouterCacheControl ? \"anthropic\" : undefined;\n\n\treturn {\n\t\tsupportsStore: !isNonStandard,\n\t\tsupportsDeveloperRole: isOpenRouterDeveloperRoleModel || (!isNonStandard && !isOpenRouter),\n\t\tsupportsReasoningEffort:\n\t\t\t!isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia && !isAntLing,\n\t\tsupportsUsageInStreaming: true,\n\t\tsupportsFinishReason: true,\n\t\tmaxTokensField: useMaxTokens ? \"max_tokens\" : \"max_completion_tokens\",\n\t\trequiresToolResultName: false,\n\t\trequiresAssistantAfterToolResult: false,\n\t\trequiresThinkingAsText: false,\n\t\trequiresReasoningContentOnAssistantMessages: isDeepSeek,\n\t\tthinkingFormat: isDeepSeek\n\t\t\t? \"deepseek\"\n\t\t\t: isZai\n\t\t\t\t? \"zai\"\n\t\t\t\t: isTogether\n\t\t\t\t\t? \"together\"\n\t\t\t\t\t: isAntLing\n\t\t\t\t\t\t? \"ant-ling\"\n\t\t\t\t\t\t: isOpenRouter\n\t\t\t\t\t\t\t? \"openrouter\"\n\t\t\t\t\t\t\t: \"openai\",\n\t\topenRouterRouting: {},\n\t\tvercelGatewayRouting: {},\n\t\tchatTemplateKwargs: {},\n\t\tzaiToolStream: false,\n\t\tsupportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,\n\t\ttoolSchemaFlavor: isMoonshot ? \"moonshot-mfjs\" : undefined,\n\t\tsupportsDisabledThinking: true,\n\t\ttoolCallFormat: undefined,\n\t\tsupportsOpenAIGrammarTools: false,\n\t\tcacheControlFormat,\n\t\tsendSessionAffinityHeaders: isOpenRouter,\n\t\tdeferredToolsMode: undefined,\n\t\tsessionAffinityFormat: isOpenRouter ? \"openrouter\" : \"openai\",\n\t\tsupportsPromptCacheKey: isMoonshot || baseUrl.includes(\"api.openai.com\"),\n\t\tsupportsLongCacheRetention: !(\n\t\t\tisTogether ||\n\t\t\tisCloudflareWorkersAI ||\n\t\t\tisCloudflareAiGateway ||\n\t\t\tisNvidia ||\n\t\t\tisAntLing\n\t\t),\n\t};\n}\n\n/**\n * Get resolved compatibility settings for a model.\n * Auto-detects from provider/URL then overrides with explicit model.compat.\n */\nexport function getOpenAICompletionsCompat(model: Model<\"openai-completions\">): ResolvedOpenAICompletionsCompat {\n\tconst detected = detectOpenAICompletionsCompat(model);\n\tif (!model.compat) return detected;\n\n\treturn {\n\t\tsupportsStore: model.compat.supportsStore ?? detected.supportsStore,\n\t\tsupportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,\n\t\tsupportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,\n\t\tsupportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? detected.supportsUsageInStreaming,\n\t\tsupportsFinishReason: model.compat.supportsFinishReason ?? detected.supportsFinishReason,\n\t\tmaxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,\n\t\trequiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,\n\t\trequiresAssistantAfterToolResult:\n\t\t\tmodel.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,\n\t\trequiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,\n\t\trequiresReasoningContentOnAssistantMessages:\n\t\t\tmodel.compat.requiresReasoningContentOnAssistantMessages ??\n\t\t\tdetected.requiresReasoningContentOnAssistantMessages,\n\t\tthinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,\n\t\tsupportsDisabledThinking: model.compat.supportsDisabledThinking ?? detected.supportsDisabledThinking,\n\t\topenRouterRouting: model.compat.openRouterRouting ?? detected.openRouterRouting,\n\t\tvercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,\n\t\tchatTemplateKwargs: model.compat.chatTemplateKwargs ?? detected.chatTemplateKwargs,\n\t\tchatTemplateArgs: model.compat.chatTemplateArgs ?? detected.chatTemplateArgs,\n\t\tzaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,\n\t\tsupportsThinkingTokenBudget: model.compat.supportsThinkingTokenBudget ?? detected.supportsThinkingTokenBudget,\n\t\tsupportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,\n\t\ttoolSchemaFlavor: model.compat.toolSchemaFlavor ?? detected.toolSchemaFlavor,\n\t\ttoolCallFormat: model.compat.toolCallFormat ?? detected.toolCallFormat,\n\t\tsupportsOpenAIGrammarTools: model.compat.supportsOpenAIGrammarTools ?? detected.supportsOpenAIGrammarTools,\n\t\tcacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,\n\t\tsendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,\n\t\tdeferredToolsMode: model.compat.deferredToolsMode ?? detected.deferredToolsMode,\n\t\tsessionAffinityFormat: model.compat.sessionAffinityFormat ?? detected.sessionAffinityFormat,\n\t\tsupportsPromptCacheKey: model.compat.supportsPromptCacheKey ?? detected.supportsPromptCacheKey,\n\t\tsupportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,\n\t};\n}\n\nexport function getBedrockModelMatchCandidates(modelId: string, modelName?: string): string[] {\n\tconst values = modelName ? [modelId, modelName] : [modelId];\n\treturn values.flatMap((value) => {\n\t\tconst lower = value.toLowerCase();\n\t\treturn [lower, lower.replace(/[\\s_.:]+/g, \"-\")];\n\t});\n}\n\nexport function supportsOneHourCacheTtl(model: Model<\"bedrock-converse-stream\">): boolean {\n\tconst candidates = getBedrockModelMatchCandidates(model.id, model.name);\n\treturn candidates.some((candidate) =>\n\t\t[\"opus-4-5\", \"sonnet-4-5\", \"haiku-4-5\"].some((modelVersion) => candidate.includes(modelVersion)),\n\t);\n}\n\n/**\n * Check if the model supports prompt caching.\n * Supported: Claude 3.5 Haiku, Claude 3.7 Sonnet, Claude 4.x models, Claude 5 models\n *\n * For base models and system-defined inference profiles the model ID / ARN\n * contains the model name, so we can decide locally.\n *\n * For application inference profiles (whose ARNs don't contain the model name),\n * also checks model.name which is user-controlled via models.json or registerProvider.\n * As a last resort, set AWS_BEDROCK_FORCE_CACHE=1 to enable cache points.\n * Amazon Nova models have automatic caching and don't need explicit cache points.\n */\nexport function supportsPromptCaching(model: Model<\"bedrock-converse-stream\">, env?: ProviderEnv): boolean {\n\tconst candidates = getBedrockModelMatchCandidates(model.id, model.name);\n\n\tconst hasClaudeRef = candidates.some((s) => s.includes(\"claude\"));\n\tif (!hasClaudeRef) {\n\t\t// Application inference profiles don't contain the model name in the ARN.\n\t\t// Allow users to force cache points via environment variable.\n\t\tif (getProviderEnvValue(\"AWS_BEDROCK_FORCE_CACHE\", env) === \"1\") return true;\n\t\treturn false;\n\t}\n\t// Claude 5 models (fable-5, opus-5, sonnet-5)\n\tif (candidates.some((s) => s.includes(\"fable-5\") || s.includes(\"opus-5\") || s.includes(\"sonnet-5\"))) return true;\n\t// Claude 4.x models (opus-4, sonnet-4, haiku-4)\n\tif (candidates.some((s) => s.includes(\"-4-\"))) return true;\n\t// Claude 3.7 Sonnet\n\tif (candidates.some((s) => s.includes(\"claude-3-7-sonnet\"))) return true;\n\t// Claude 3.5 Haiku\n\tif (candidates.some((s) => s.includes(\"claude-3-5-haiku\"))) return true;\n\treturn false;\n}\n\nfunction resolveAnthropicCacheRetention(\n\tcacheRetention?: CacheRetention,\n\tenv?: ProviderEnv,\n\tfallback: CacheRetention = \"short\",\n): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\tif (typeof process !== \"undefined\" && process.env.PI_CACHE_RETENTION !== undefined) {\n\t\treturn \"short\";\n\t}\n\treturn fallback;\n}\n\nfunction resolveOpenAICompletionsCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction resolveBedrockCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction resolveOpenAIResponsesCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (getProviderEnvValue(\"PI_CACHE_RETENTION\", env) === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nexport function resolvePromptCacheTtlSeconds(model: Model<Api>, env?: ProviderEnv): number | undefined {\n\tswitch (model.api) {\n\t\tcase \"claude-sdk-oauth\":\n\t\t\t// The Claude SDK owns prompt caching for this lane and uses Anthropic's default 5m TTL.\n\t\t\treturn PROMPT_CACHE_TTL_SHORT_SECONDS;\n\t\tcase \"anthropic-messages\": {\n\t\t\tconst anthropicModel = model as Model<\"anthropic-messages\">;\n\t\t\tconst retention = resolveAnthropicCacheRetention(anthropicModel.cacheRetention, env, \"short\");\n\t\t\tif (retention === \"none\") return 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\" && supportsOneHourCacheTtl(bedrockModel)\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 \"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.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider-retry.js","sourceRoot":"","sources":["../../src/utils/provider-retry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,wBAAwB,EAAE,sBAAsB,EAAE,MAAM,iBAAiB,CAAC;AAEnF,MAAM,0BAA0B,GAAG,MAAM,CAAC;AAC1C,MAAM,mCAAmC,GAAG,iDAAiD,CAAC;AAiB9F,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,OAAO,GAAG,oBAAoB,IAAI,CAAC,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,uBAAuB,IAAI,CAAC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,OAAO,oBAAoB,EAAE,CAAC;QAC9I,MAAM,KAAK,GAA4B,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,wBAAwB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,EAAE;YAC3G,YAAY,EAAE,OAAO;SACrB,CAAC,CAAC;QACH,MAAM,KAAK,CAAC;IACb,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,eAAe,CAAC,KAAoB,EAAE,UAAkB,EAAE,eAAmC;IACrG,MAAM,MAAM,GAAG,sBAAsB,CAAC;QACrC,MAAM,EAAE,GAAG;QACX,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,QAAQ,EAAE,EAAE;KACZ,CAAC,CAAC;IACH,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;QAC1B,OAAO,0BAA0B,CAAC,MAAM,EAAE,eAAe,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IAC3E,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":["import { appendRetryAfterMsMarker, extract429RetryAfterMs } from \"./retry-hint.ts\";\n\nconst DEFAULT_MAX_RETRY_DELAY_MS = 60_000;\nconst DIGITALOCEAN_STREAM_FAILURE_MESSAGE = \"Upstream error from DigitalOcean: stream failed\";\n\nexport interface ProviderRetryDelayError extends Error {\n\treadonly retryAfterMs: number;\n}\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\tconst message = `Server requested ${Math.ceil(delayMs / 1000)}s retry delay (max: ${Math.ceil(maxDelayMs / 1000)}s). ${providerErrorMessage}`;\n\t\tconst error: ProviderRetryDelayError = Object.assign(new Error(appendRetryAfterMsMarker(message, delayMs)), {\n\t\t\tretryAfterMs: delayMs,\n\t\t});\n\t\tthrow error;\n\t}\n\treturn delayMs;\n}\n\nfunction getRetryDelayMs(error: ProviderError, retryIndex: number, maxRetryDelayMs: number | undefined): number {\n\tconst hintMs = extract429RetryAfterMs({\n\t\tstatus: 429,\n\t\theaders: error.headers,\n\t\tbodyText: \"\",\n\t});\n\tif (hintMs !== undefined) {\n\t\treturn validateServerRetryDelayMs(hintMs, 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"]}
+{"version":3,"file":"provider-retry.js","sourceRoot":"","sources":["../../src/utils/provider-retry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,wBAAwB,EAAE,sBAAsB,EAAE,MAAM,iBAAiB,CAAC;AAEnF,MAAM,0BAA0B,GAAG,MAAM,CAAC;AAC1C,MAAM,mCAAmC,GAAG,iDAAiD,CAAC;AAiB9F,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,OAAO,GAAG,oBAAoB,IAAI,CAAC,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,uBAAuB,IAAI,CAAC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,OAAO,oBAAoB,EAAE,CAAC;QAC9I,MAAM,KAAK,GAA4B,MAAM,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,wBAAwB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,EAAE;YAC3G,YAAY,EAAE,OAAO;SACrB,CAAC,CAAC;QACH,MAAM,KAAK,CAAC;IACb,CAAC;IACD,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,SAAS,eAAe,CAAC,KAAoB,EAAE,UAAkB,EAAE,eAAmC;IACrG,MAAM,MAAM,GAAG,sBAAsB,CAAC;QACrC,MAAM,EAAE,GAAG;QACX,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,QAAQ,EAAE,EAAE;KACZ,CAAC,CAAC;IACH,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;QAC1B,OAAO,0BAA0B,CAAC,MAAM,EAAE,eAAe,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IAC3E,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,UAAgC,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,UAAgC,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":["import { appendRetryAfterMsMarker, extract429RetryAfterMs } from \"./retry-hint.ts\";\n\nconst DEFAULT_MAX_RETRY_DELAY_MS = 60_000;\nconst DIGITALOCEAN_STREAM_FAILURE_MESSAGE = \"Upstream error from DigitalOcean: stream failed\";\n\nexport interface ProviderRetryDelayError extends Error {\n\treadonly retryAfterMs: number;\n}\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\tconst message = `Server requested ${Math.ceil(delayMs / 1000)}s retry delay (max: ${Math.ceil(maxDelayMs / 1000)}s). ${providerErrorMessage}`;\n\t\tconst error: ProviderRetryDelayError = Object.assign(new Error(appendRetryAfterMsMarker(message, delayMs)), {\n\t\t\tretryAfterMs: delayMs,\n\t\t});\n\t\tthrow error;\n\t}\n\treturn delayMs;\n}\n\nfunction getRetryDelayMs(error: ProviderError, retryIndex: number, maxRetryDelayMs: number | undefined): number {\n\tconst hintMs = extract429RetryAfterMs({\n\t\tstatus: 429,\n\t\theaders: error.headers,\n\t\tbodyText: \"\",\n\t});\n\tif (hintMs !== undefined) {\n\t\treturn validateServerRetryDelayMs(hintMs, 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

@@ -21,6 +21,31 @@ export interface RetryCallbacks {
     /** Emitted once when the loop ends: success if a later call completed normally. */
     onRetryFinished?: (success: boolean, attempt: number, finalError?: string) => void | Promise<void>;
 }
+/**
+ * Run a producer that signals failure by THROWING, with the same bounded
+ * backoff, abort, and callback semantics as {@link retryAssistantCall}.
+ *
+ * `retryAssistantCall` is value-based: its producer reports failure by resolving
+ * an `AssistantMessage` with `stopReason: "error"`. Callers that instead reject
+ * (compaction summarization, and any other request pipeline that throws) need
+ * the identical retry policy without reshaping their failures into assistant
+ * messages first, so both live here and share one implementation of the delay,
+ * abort, and callback contract.
+ *
+ * Behavior:
+ * - A resolved value is returned immediately.
+ * - A throw that `isRetryable` rejects is rethrown at once: deterministic
+ *   failures never spend a retry.
+ * - Otherwise the call is retried up to `policy.maxRetries` times with the same
+ *   exponential backoff (`baseDelayMs * 2^(attempt-1)`), rethrowing the final
+ *   error when the budget is exhausted.
+ * - An abort during the backoff sleep stops the loop and rethrows; unlike the
+ *   assistant-message path there is no aborted value to normalize to.
+ *
+ * When `policy` is undefined or disabled the producer runs exactly once and its
+ * error propagates unchanged.
+ */
+export declare function retryTransientCall<T>(produce: () => Promise<T>, isRetryable: (error: unknown) => boolean, policy: RetryPolicy | undefined, signal: AbortSignal | undefined, callbacks?: RetryCallbacks): Promise<T>;
 /**
  * Run a single assistant-producing call with bounded retry on transient errors.
  *

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;AAqIpD;;;;;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;AAYD,wBAAgB,0BAA0B,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAE7E;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAIzE;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,YAAY,EAAE,MAAM,GAAG,OAAO,CAIrE"}
+{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAqIpD;;;;;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;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,wBAAsB,kBAAkB,CAAC,CAAC,EACzC,OAAO,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACzB,WAAW,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,CAAC,CAAC,CAiCZ;AAOD;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAsB,kBAAkB,CACvC,OAAO,EAAE,MAAM,OAAO,CAAC,gBAAgB,CAAC,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,gBAAgB,CAAC,CA4C3B;AAED;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAU5E;AAYD,wBAAgB,0BAA0B,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAE7E;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAIzE;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,YAAY,EAAE,MAAM,GAAG,OAAO,CAIrE"}

package/dist/utils/retry.js

@@ -131,6 +131,70 @@ function sleep(ms, signal) {
         }, { once: true });
     });
 }
+/**
+ * Run a producer that signals failure by THROWING, with the same bounded
+ * backoff, abort, and callback semantics as {@link retryAssistantCall}.
+ *
+ * `retryAssistantCall` is value-based: its producer reports failure by resolving
+ * an `AssistantMessage` with `stopReason: "error"`. Callers that instead reject
+ * (compaction summarization, and any other request pipeline that throws) need
+ * the identical retry policy without reshaping their failures into assistant
+ * messages first, so both live here and share one implementation of the delay,
+ * abort, and callback contract.
+ *
+ * Behavior:
+ * - A resolved value is returned immediately.
+ * - A throw that `isRetryable` rejects is rethrown at once: deterministic
+ *   failures never spend a retry.
+ * - Otherwise the call is retried up to `policy.maxRetries` times with the same
+ *   exponential backoff (`baseDelayMs * 2^(attempt-1)`), rethrowing the final
+ *   error when the budget is exhausted.
+ * - An abort during the backoff sleep stops the loop and rethrows; unlike the
+ *   assistant-message path there is no aborted value to normalize to.
+ *
+ * When `policy` is undefined or disabled the producer runs exactly once and its
+ * error propagates unchanged.
+ */
+export async function retryTransientCall(produce, isRetryable, policy, signal, callbacks) {
+    const maxAttempts = policy?.enabled ? policy.maxRetries : 0;
+    let attempt = 0;
+    let lastRetry;
+    for (;;) {
+        try {
+            const value = await produce();
+            if (lastRetry)
+                await callbacks?.onRetryFinished?.(true, lastRetry.attempt);
+            return value;
+        }
+        catch (error) {
+            if (error instanceof RetrySleepAbortError)
+                throw error;
+            if (attempt >= maxAttempts || !isRetryable(error)) {
+                if (lastRetry) {
+                    await callbacks?.onRetryFinished?.(false, lastRetry.attempt, errorMessageOf(error));
+                }
+                throw error;
+            }
+            attempt++;
+            lastRetry = { attempt, errorMessage: errorMessageOf(error) };
+            const delayMs = policy.baseDelayMs * 2 ** (attempt - 1);
+            await callbacks?.onRetryScheduled?.(attempt, maxAttempts, delayMs, lastRetry.errorMessage);
+            try {
+                await sleep(delayMs, signal);
+            }
+            catch (sleepError) {
+                await callbacks?.onRetryFinished?.(false, attempt, lastRetry.errorMessage);
+                throw sleepError instanceof RetrySleepAbortError ? error : sleepError;
+            }
+            await callbacks?.onRetryAttemptStart?.();
+        }
+    }
+}
+function errorMessageOf(error) {
+    if (error instanceof Error)
+        return error.message || "Unknown error";
+    return String(error) || "Unknown error";
+}
 /**
  * Run a single assistant-producing call with bounded retry on transient errors.
  *