langchain 1.0.6 → 1.1.1

package/dist/agents/middleware/modelRetry.d.cts

@@ -0,0 +1,134 @@
+import { AgentMiddleware } from "./types.cjs";
+import { z } from "zod/v3";
+
+//#region src/agents/middleware/modelRetry.d.ts
+
+/**
+ * Configuration options for the Model Retry Middleware.
+ */
+declare const ModelRetryMiddlewareOptionsSchema: z.ZodObject<{
+  /**
+   * Behavior when all retries are exhausted. Options:
+   * - `"continue"` (default): Return an AIMessage with error details, allowing
+   *   the agent to potentially handle the failure gracefully.
+   * - `"error"`: Re-raise the exception, stopping agent execution.
+   * - Custom function: Function that takes the exception and returns a string
+   *   for the AIMessage content, allowing custom error formatting.
+   */
+  onFailure: z.ZodDefault<z.ZodUnion<[z.ZodLiteral<"error">, z.ZodLiteral<"continue">, z.ZodFunction<z.ZodTuple<[z.ZodType<Error, z.ZodTypeDef, Error>], z.ZodUnknown>, z.ZodString>]>>;
+} & {
+  maxRetries: z.ZodDefault<z.ZodNumber>;
+  retryOn: z.ZodDefault<z.ZodUnion<[z.ZodFunction<z.ZodTuple<[z.ZodType<Error, z.ZodTypeDef, Error>], z.ZodUnknown>, z.ZodBoolean>, z.ZodArray<z.ZodType<new (...args: any[]) => Error, z.ZodTypeDef, new (...args: any[]) => Error>, "many">]>>;
+  backoffFactor: z.ZodDefault<z.ZodNumber>;
+  initialDelayMs: z.ZodDefault<z.ZodNumber>;
+  maxDelayMs: z.ZodDefault<z.ZodNumber>;
+  jitter: z.ZodDefault<z.ZodBoolean>;
+}, "strip", z.ZodTypeAny, {
+  maxRetries: number;
+  retryOn: (new (...args: any[]) => Error)[] | ((args_0: Error, ...args: unknown[]) => boolean);
+  backoffFactor: number;
+  initialDelayMs: number;
+  maxDelayMs: number;
+  jitter: boolean;
+  onFailure: "continue" | "error" | ((args_0: Error, ...args: unknown[]) => string);
+}, {
+  maxRetries?: number | undefined;
+  retryOn?: (new (...args: any[]) => Error)[] | ((args_0: Error, ...args: unknown[]) => boolean) | undefined;
+  backoffFactor?: number | undefined;
+  initialDelayMs?: number | undefined;
+  maxDelayMs?: number | undefined;
+  jitter?: boolean | undefined;
+  onFailure?: "continue" | "error" | ((args_0: Error, ...args: unknown[]) => string) | undefined;
+}>;
+type ModelRetryMiddlewareConfig = z.input<typeof ModelRetryMiddlewareOptionsSchema>;
+/**
+ * Middleware that automatically retries failed model calls with configurable backoff.
+ *
+ * Supports retrying on specific exceptions and exponential backoff.
+ *
+ * @example Basic usage with default settings (2 retries, exponential backoff)
+ * ```ts
+ * import { createAgent, modelRetryMiddleware } from "langchain";
+ *
+ * const agent = createAgent({
+ *   model: "openai:gpt-4o",
+ *   tools: [searchTool],
+ *   middleware: [modelRetryMiddleware()],
+ * });
+ * ```
+ *
+ * @example Retry specific exceptions only
+ * ```ts
+ * import { modelRetryMiddleware } from "langchain";
+ *
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 4,
+ *   retryOn: [TimeoutError, NetworkError],
+ *   backoffFactor: 1.5,
+ * });
+ * ```
+ *
+ * @example Custom exception filtering
+ * ```ts
+ * function shouldRetry(error: Error): boolean {
+ *   // Only retry on rate limit errors
+ *   if (error.name === "RateLimitError") {
+ *     return true;
+ *   }
+ *   // Or check for specific HTTP status codes
+ *   if (error.name === "HTTPError" && "statusCode" in error) {
+ *     const statusCode = (error as any).statusCode;
+ *     return statusCode === 429 || statusCode === 503;
+ *   }
+ *   return false;
+ * }
+ *
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 3,
+ *   retryOn: shouldRetry,
+ * });
+ * ```
+ *
+ * @example Return error message instead of raising
+ * ```ts
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 4,
+ *   onFailure: "continue", // Return AIMessage with error instead of throwing
+ * });
+ * ```
+ *
+ * @example Custom error message formatting
+ * ```ts
+ * const formatError = (error: Error) =>
+ *   `Model call failed: ${error.message}. Please try again later.`;
+ *
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 4,
+ *   onFailure: formatError,
+ * });
+ * ```
+ *
+ * @example Constant backoff (no exponential growth)
+ * ```ts
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 5,
+ *   backoffFactor: 0.0, // No exponential growth
+ *   initialDelayMs: 2000, // Always wait 2 seconds
+ * });
+ * ```
+ *
+ * @example Raise exception on failure
+ * ```ts
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 2,
+ *   onFailure: "error", // Re-raise exception instead of returning message
+ * });
+ * ```
+ *
+ * @param config - Configuration options for the retry middleware
+ * @returns A middleware instance that handles model failures with retries
+ */
+declare function modelRetryMiddleware(config?: ModelRetryMiddlewareConfig): AgentMiddleware;
+//#endregion
+export { ModelRetryMiddlewareConfig, modelRetryMiddleware };
+//# sourceMappingURL=modelRetry.d.cts.map

package/dist/agents/middleware/modelRetry.d.ts

@@ -0,0 +1,134 @@
+import { AgentMiddleware } from "./types.js";
+import { z } from "zod/v3";
+
+//#region src/agents/middleware/modelRetry.d.ts
+
+/**
+ * Configuration options for the Model Retry Middleware.
+ */
+declare const ModelRetryMiddlewareOptionsSchema: z.ZodObject<{
+  /**
+   * Behavior when all retries are exhausted. Options:
+   * - `"continue"` (default): Return an AIMessage with error details, allowing
+   *   the agent to potentially handle the failure gracefully.
+   * - `"error"`: Re-raise the exception, stopping agent execution.
+   * - Custom function: Function that takes the exception and returns a string
+   *   for the AIMessage content, allowing custom error formatting.
+   */
+  onFailure: z.ZodDefault<z.ZodUnion<[z.ZodLiteral<"error">, z.ZodLiteral<"continue">, z.ZodFunction<z.ZodTuple<[z.ZodType<Error, z.ZodTypeDef, Error>], z.ZodUnknown>, z.ZodString>]>>;
+} & {
+  maxRetries: z.ZodDefault<z.ZodNumber>;
+  retryOn: z.ZodDefault<z.ZodUnion<[z.ZodFunction<z.ZodTuple<[z.ZodType<Error, z.ZodTypeDef, Error>], z.ZodUnknown>, z.ZodBoolean>, z.ZodArray<z.ZodType<new (...args: any[]) => Error, z.ZodTypeDef, new (...args: any[]) => Error>, "many">]>>;
+  backoffFactor: z.ZodDefault<z.ZodNumber>;
+  initialDelayMs: z.ZodDefault<z.ZodNumber>;
+  maxDelayMs: z.ZodDefault<z.ZodNumber>;
+  jitter: z.ZodDefault<z.ZodBoolean>;
+}, "strip", z.ZodTypeAny, {
+  maxRetries: number;
+  retryOn: (new (...args: any[]) => Error)[] | ((args_0: Error, ...args: unknown[]) => boolean);
+  backoffFactor: number;
+  initialDelayMs: number;
+  maxDelayMs: number;
+  jitter: boolean;
+  onFailure: "continue" | "error" | ((args_0: Error, ...args: unknown[]) => string);
+}, {
+  maxRetries?: number | undefined;
+  retryOn?: (new (...args: any[]) => Error)[] | ((args_0: Error, ...args: unknown[]) => boolean) | undefined;
+  backoffFactor?: number | undefined;
+  initialDelayMs?: number | undefined;
+  maxDelayMs?: number | undefined;
+  jitter?: boolean | undefined;
+  onFailure?: "continue" | "error" | ((args_0: Error, ...args: unknown[]) => string) | undefined;
+}>;
+type ModelRetryMiddlewareConfig = z.input<typeof ModelRetryMiddlewareOptionsSchema>;
+/**
+ * Middleware that automatically retries failed model calls with configurable backoff.
+ *
+ * Supports retrying on specific exceptions and exponential backoff.
+ *
+ * @example Basic usage with default settings (2 retries, exponential backoff)
+ * ```ts
+ * import { createAgent, modelRetryMiddleware } from "langchain";
+ *
+ * const agent = createAgent({
+ *   model: "openai:gpt-4o",
+ *   tools: [searchTool],
+ *   middleware: [modelRetryMiddleware()],
+ * });
+ * ```
+ *
+ * @example Retry specific exceptions only
+ * ```ts
+ * import { modelRetryMiddleware } from "langchain";
+ *
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 4,
+ *   retryOn: [TimeoutError, NetworkError],
+ *   backoffFactor: 1.5,
+ * });
+ * ```
+ *
+ * @example Custom exception filtering
+ * ```ts
+ * function shouldRetry(error: Error): boolean {
+ *   // Only retry on rate limit errors
+ *   if (error.name === "RateLimitError") {
+ *     return true;
+ *   }
+ *   // Or check for specific HTTP status codes
+ *   if (error.name === "HTTPError" && "statusCode" in error) {
+ *     const statusCode = (error as any).statusCode;
+ *     return statusCode === 429 || statusCode === 503;
+ *   }
+ *   return false;
+ * }
+ *
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 3,
+ *   retryOn: shouldRetry,
+ * });
+ * ```
+ *
+ * @example Return error message instead of raising
+ * ```ts
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 4,
+ *   onFailure: "continue", // Return AIMessage with error instead of throwing
+ * });
+ * ```
+ *
+ * @example Custom error message formatting
+ * ```ts
+ * const formatError = (error: Error) =>
+ *   `Model call failed: ${error.message}. Please try again later.`;
+ *
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 4,
+ *   onFailure: formatError,
+ * });
+ * ```
+ *
+ * @example Constant backoff (no exponential growth)
+ * ```ts
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 5,
+ *   backoffFactor: 0.0, // No exponential growth
+ *   initialDelayMs: 2000, // Always wait 2 seconds
+ * });
+ * ```
+ *
+ * @example Raise exception on failure
+ * ```ts
+ * const retry = modelRetryMiddleware({
+ *   maxRetries: 2,
+ *   onFailure: "error", // Re-raise exception instead of returning message
+ * });
+ * ```
+ *
+ * @param config - Configuration options for the retry middleware
+ * @returns A middleware instance that handles model failures with retries
+ */
+declare function modelRetryMiddleware(config?: ModelRetryMiddlewareConfig): AgentMiddleware;
+//#endregion
+export { ModelRetryMiddlewareConfig, modelRetryMiddleware };
+//# sourceMappingURL=modelRetry.d.ts.map

package/dist/agents/middleware/modelRetry.js

@@ -0,0 +1,161 @@
+import { calculateRetryDelay, sleep } from "./utils.js";
+import { createMiddleware } from "../middleware.js";
+import { RetrySchema } from "./constants.js";
+import { InvalidRetryConfigError } from "./error.js";
+import { AIMessage } from "@langchain/core/messages";
+import { z } from "zod/v3";
+
+//#region src/agents/middleware/modelRetry.ts
+/**
+* Configuration options for the Model Retry Middleware.
+*/
+const ModelRetryMiddlewareOptionsSchema = z.object({ onFailure: z.union([
+	z.literal("error"),
+	z.literal("continue"),
+	z.function().args(z.instanceof(Error)).returns(z.string())
+]).default("continue") }).merge(RetrySchema);
+/**
+* Middleware that automatically retries failed model calls with configurable backoff.
+*
+* Supports retrying on specific exceptions and exponential backoff.
+*
+* @example Basic usage with default settings (2 retries, exponential backoff)
+* ```ts
+* import { createAgent, modelRetryMiddleware } from "langchain";
+*
+* const agent = createAgent({
+*   model: "openai:gpt-4o",
+*   tools: [searchTool],
+*   middleware: [modelRetryMiddleware()],
+* });
+* ```
+*
+* @example Retry specific exceptions only
+* ```ts
+* import { modelRetryMiddleware } from "langchain";
+*
+* const retry = modelRetryMiddleware({
+*   maxRetries: 4,
+*   retryOn: [TimeoutError, NetworkError],
+*   backoffFactor: 1.5,
+* });
+* ```
+*
+* @example Custom exception filtering
+* ```ts
+* function shouldRetry(error: Error): boolean {
+*   // Only retry on rate limit errors
+*   if (error.name === "RateLimitError") {
+*     return true;
+*   }
+*   // Or check for specific HTTP status codes
+*   if (error.name === "HTTPError" && "statusCode" in error) {
+*     const statusCode = (error as any).statusCode;
+*     return statusCode === 429 || statusCode === 503;
+*   }
+*   return false;
+* }
+*
+* const retry = modelRetryMiddleware({
+*   maxRetries: 3,
+*   retryOn: shouldRetry,
+* });
+* ```
+*
+* @example Return error message instead of raising
+* ```ts
+* const retry = modelRetryMiddleware({
+*   maxRetries: 4,
+*   onFailure: "continue", // Return AIMessage with error instead of throwing
+* });
+* ```
+*
+* @example Custom error message formatting
+* ```ts
+* const formatError = (error: Error) =>
+*   `Model call failed: ${error.message}. Please try again later.`;
+*
+* const retry = modelRetryMiddleware({
+*   maxRetries: 4,
+*   onFailure: formatError,
+* });
+* ```
+*
+* @example Constant backoff (no exponential growth)
+* ```ts
+* const retry = modelRetryMiddleware({
+*   maxRetries: 5,
+*   backoffFactor: 0.0, // No exponential growth
+*   initialDelayMs: 2000, // Always wait 2 seconds
+* });
+* ```
+*
+* @example Raise exception on failure
+* ```ts
+* const retry = modelRetryMiddleware({
+*   maxRetries: 2,
+*   onFailure: "error", // Re-raise exception instead of returning message
+* });
+* ```
+*
+* @param config - Configuration options for the retry middleware
+* @returns A middleware instance that handles model failures with retries
+*/
+function modelRetryMiddleware(config = {}) {
+	const { success, error, data } = ModelRetryMiddlewareOptionsSchema.safeParse(config);
+	if (!success) throw new InvalidRetryConfigError(error);
+	const { maxRetries, retryOn, onFailure, backoffFactor, initialDelayMs, maxDelayMs, jitter } = data;
+	/**
+	* Check if the exception should trigger a retry.
+	*/
+	const shouldRetryException = (error$1) => {
+		if (typeof retryOn === "function") return retryOn(error$1);
+		return retryOn.some((ErrorConstructor) => error$1.constructor === ErrorConstructor);
+	};
+	const delayConfig = {
+		backoffFactor,
+		initialDelayMs,
+		maxDelayMs,
+		jitter
+	};
+	/**
+	* Format the failure message when retries are exhausted.
+	*/
+	const formatFailureMessage = (error$1, attemptsMade) => {
+		const errorType = error$1.constructor.name;
+		const attemptWord = attemptsMade === 1 ? "attempt" : "attempts";
+		return `Model call failed after ${attemptsMade} ${attemptWord} with ${errorType}: ${error$1.message}`;
+	};
+	/**
+	* Handle failure when all retries are exhausted.
+	*/
+	const handleFailure = (error$1, attemptsMade) => {
+		if (onFailure === "error") throw error$1;
+		let content;
+		if (typeof onFailure === "function") content = onFailure(error$1);
+		else content = formatFailureMessage(error$1, attemptsMade);
+		return new AIMessage({ content });
+	};
+	return createMiddleware({
+		name: "modelRetryMiddleware",
+		contextSchema: ModelRetryMiddlewareOptionsSchema,
+		wrapModelCall: async (request, handler) => {
+			for (let attempt = 0; attempt <= maxRetries; attempt++) try {
+				return await handler(request);
+			} catch (error$1) {
+				const attemptsMade = attempt + 1;
+				const err = error$1 && typeof error$1 === "object" && "message" in error$1 ? error$1 : new Error(String(error$1));
+				if (!shouldRetryException(err)) return handleFailure(err, attemptsMade);
+				if (attempt < maxRetries) {
+					const delay = calculateRetryDelay(delayConfig, attempt);
+					if (delay > 0) await sleep(delay);
+				} else return handleFailure(err, attemptsMade);
+			}
+			throw new Error("Unexpected: retry loop completed without returning");
+		}
+	});
+}
+
+//#endregion
+export { modelRetryMiddleware };
+//# sourceMappingURL=modelRetry.js.map

package/dist/agents/middleware/modelRetry.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"modelRetry.js","names":["config: ModelRetryMiddlewareConfig","error: Error","error","attemptsMade: number","content: string"],"sources":["../../../src/agents/middleware/modelRetry.ts"],"sourcesContent":["/**\n * Model retry middleware for agents.\n */\nimport { z } from \"zod/v3\";\nimport { AIMessage } from \"@langchain/core/messages\";\n\nimport { createMiddleware } from \"../middleware.js\";\nimport type { AgentMiddleware } from \"./types.js\";\nimport { sleep, calculateRetryDelay } from \"./utils.js\";\nimport { RetrySchema } from \"./constants.js\";\nimport { InvalidRetryConfigError } from \"./error.js\";\n\n/**\n * Configuration options for the Model Retry Middleware.\n */\nexport const ModelRetryMiddlewareOptionsSchema = z\n  .object({\n    /**\n     * Behavior when all retries are exhausted. Options:\n     * - `\"continue\"` (default): Return an AIMessage with error details, allowing\n     *   the agent to potentially handle the failure gracefully.\n     * - `\"error\"`: Re-raise the exception, stopping agent execution.\n     * - Custom function: Function that takes the exception and returns a string\n     *   for the AIMessage content, allowing custom error formatting.\n     */\n    onFailure: z\n      .union([\n        z.literal(\"error\"),\n        z.literal(\"continue\"),\n        z.function().args(z.instanceof(Error)).returns(z.string()),\n      ])\n      .default(\"continue\"),\n  })\n  .merge(RetrySchema);\n\nexport type ModelRetryMiddlewareConfig = z.input<\n  typeof ModelRetryMiddlewareOptionsSchema\n>;\n\n/**\n * Middleware that automatically retries failed model calls with configurable backoff.\n *\n * Supports retrying on specific exceptions and exponential backoff.\n *\n * @example Basic usage with default settings (2 retries, exponential backoff)\n * ```ts\n * import { createAgent, modelRetryMiddleware } from \"langchain\";\n *\n * const agent = createAgent({\n *   model: \"openai:gpt-4o\",\n *   tools: [searchTool],\n *   middleware: [modelRetryMiddleware()],\n * });\n * ```\n *\n * @example Retry specific exceptions only\n * ```ts\n * import { modelRetryMiddleware } from \"langchain\";\n *\n * const retry = modelRetryMiddleware({\n *   maxRetries: 4,\n *   retryOn: [TimeoutError, NetworkError],\n *   backoffFactor: 1.5,\n * });\n * ```\n *\n * @example Custom exception filtering\n * ```ts\n * function shouldRetry(error: Error): boolean {\n *   // Only retry on rate limit errors\n *   if (error.name === \"RateLimitError\") {\n *     return true;\n *   }\n *   // Or check for specific HTTP status codes\n *   if (error.name === \"HTTPError\" && \"statusCode\" in error) {\n *     const statusCode = (error as any).statusCode;\n *     return statusCode === 429 || statusCode === 503;\n *   }\n *   return false;\n * }\n *\n * const retry = modelRetryMiddleware({\n *   maxRetries: 3,\n *   retryOn: shouldRetry,\n * });\n * ```\n *\n * @example Return error message instead of raising\n * ```ts\n * const retry = modelRetryMiddleware({\n *   maxRetries: 4,\n *   onFailure: \"continue\", // Return AIMessage with error instead of throwing\n * });\n * ```\n *\n * @example Custom error message formatting\n * ```ts\n * const formatError = (error: Error) =>\n *   `Model call failed: ${error.message}. Please try again later.`;\n *\n * const retry = modelRetryMiddleware({\n *   maxRetries: 4,\n *   onFailure: formatError,\n * });\n * ```\n *\n * @example Constant backoff (no exponential growth)\n * ```ts\n * const retry = modelRetryMiddleware({\n *   maxRetries: 5,\n *   backoffFactor: 0.0, // No exponential growth\n *   initialDelayMs: 2000, // Always wait 2 seconds\n * });\n * ```\n *\n * @example Raise exception on failure\n * ```ts\n * const retry = modelRetryMiddleware({\n *   maxRetries: 2,\n *   onFailure: \"error\", // Re-raise exception instead of returning message\n * });\n * ```\n *\n * @param config - Configuration options for the retry middleware\n * @returns A middleware instance that handles model failures with retries\n */\nexport function modelRetryMiddleware(\n  config: ModelRetryMiddlewareConfig = {}\n): AgentMiddleware {\n  const { success, error, data } =\n    ModelRetryMiddlewareOptionsSchema.safeParse(config);\n  if (!success) {\n    throw new InvalidRetryConfigError(error);\n  }\n  const {\n    maxRetries,\n    retryOn,\n    onFailure,\n    backoffFactor,\n    initialDelayMs,\n    maxDelayMs,\n    jitter,\n  } = data;\n\n  /**\n   * Check if the exception should trigger a retry.\n   */\n  const shouldRetryException = (error: Error): boolean => {\n    if (typeof retryOn === \"function\") {\n      return retryOn(error);\n    }\n    // retryOn is an array of error constructors\n    return retryOn.some(\n      (ErrorConstructor) => error.constructor === ErrorConstructor\n    );\n  };\n\n  // Use the exported calculateRetryDelay function with our config\n  const delayConfig = { backoffFactor, initialDelayMs, maxDelayMs, jitter };\n\n  /**\n   * Format the failure message when retries are exhausted.\n   */\n  const formatFailureMessage = (error: Error, attemptsMade: number): string => {\n    const errorType = error.constructor.name;\n    const attemptWord = attemptsMade === 1 ? \"attempt\" : \"attempts\";\n    return `Model call failed after ${attemptsMade} ${attemptWord} with ${errorType}: ${error.message}`;\n  };\n\n  /**\n   * Handle failure when all retries are exhausted.\n   */\n  const handleFailure = (error: Error, attemptsMade: number): AIMessage => {\n    if (onFailure === \"error\") {\n      throw error;\n    }\n\n    let content: string;\n    if (typeof onFailure === \"function\") {\n      content = onFailure(error);\n    } else {\n      content = formatFailureMessage(error, attemptsMade);\n    }\n\n    return new AIMessage({\n      content,\n    });\n  };\n\n  return createMiddleware({\n    name: \"modelRetryMiddleware\",\n    contextSchema: ModelRetryMiddlewareOptionsSchema,\n    wrapModelCall: async (request, handler) => {\n      // Initial attempt + retries\n      for (let attempt = 0; attempt <= maxRetries; attempt++) {\n        try {\n          return await handler(request);\n        } catch (error) {\n          const attemptsMade = attempt + 1; // attempt is 0-indexed\n\n          // Ensure error is an Error instance\n          const err =\n            error && typeof error === \"object\" && \"message\" in error\n              ? (error as Error)\n              : new Error(String(error));\n\n          // Check if we should retry this exception\n          if (!shouldRetryException(err)) {\n            // Exception is not retryable, handle failure immediately\n            return handleFailure(err, attemptsMade);\n          }\n\n          // Check if we have more retries left\n          if (attempt < maxRetries) {\n            // Calculate and apply backoff delay\n            const delay = calculateRetryDelay(delayConfig, attempt);\n            if (delay > 0) {\n              await sleep(delay);\n            }\n            // Continue to next retry\n          } else {\n            // No more retries, handle failure\n            return handleFailure(err, attemptsMade);\n          }\n        }\n      }\n\n      // Unreachable: loop always returns via handler success or handleFailure\n      throw new Error(\"Unexpected: retry loop completed without returning\");\n    },\n  });\n}\n"],"mappings":";;;;;;;;;;;AAeA,MAAa,oCAAoC,EAC9C,OAAO,EASN,WAAW,EACR,MAAM;CACL,EAAE,QAAQ,QAAQ;CAClB,EAAE,QAAQ,WAAW;CACrB,EAAE,UAAU,CAAC,KAAK,EAAE,WAAW,MAAM,CAAC,CAAC,QAAQ,EAAE,QAAQ,CAAC;AAC3D,EAAC,CACD,QAAQ,WAAW,CACvB,EAAC,CACD,MAAM,YAAY;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6FrB,SAAgB,qBACdA,SAAqC,CAAE,GACtB;CACjB,MAAM,EAAE,SAAS,OAAO,MAAM,GAC5B,kCAAkC,UAAU,OAAO;AACrD,KAAI,CAAC,QACH,OAAM,IAAI,wBAAwB;CAEpC,MAAM,EACJ,YACA,SACA,WACA,eACA,gBACA,YACA,QACD,GAAG;;;;CAKJ,MAAM,uBAAuB,CAACC,YAA0B;AACtD,MAAI,OAAO,YAAY,WACrB,QAAO,QAAQC,QAAM;AAGvB,SAAO,QAAQ,KACb,CAAC,qBAAqBA,QAAM,gBAAgB,iBAC7C;CACF;CAGD,MAAM,cAAc;EAAE;EAAe;EAAgB;EAAY;CAAQ;;;;CAKzE,MAAM,uBAAuB,CAACD,SAAcE,iBAAiC;EAC3E,MAAM,YAAYD,QAAM,YAAY;EACpC,MAAM,cAAc,iBAAiB,IAAI,YAAY;AACrD,SAAO,CAAC,wBAAwB,EAAE,aAAa,CAAC,EAAE,YAAY,MAAM,EAAE,UAAU,EAAE,EAAEA,QAAM,SAAS;CACpG;;;;CAKD,MAAM,gBAAgB,CAACD,SAAcE,iBAAoC;AACvE,MAAI,cAAc,QAChB,OAAMD;EAGR,IAAIE;AACJ,MAAI,OAAO,cAAc,YACvB,UAAU,UAAUF,QAAM;OAE1B,UAAU,qBAAqBA,SAAO,aAAa;AAGrD,SAAO,IAAI,UAAU,EACnB,QACD;CACF;AAED,QAAO,iBAAiB;EACtB,MAAM;EACN,eAAe;EACf,eAAe,OAAO,SAAS,YAAY;AAEzC,QAAK,IAAI,UAAU,GAAG,WAAW,YAAY,UAC3C,KAAI;AACF,WAAO,MAAM,QAAQ,QAAQ;GAC9B,SAAQA,SAAO;IACd,MAAM,eAAe,UAAU;IAG/B,MAAM,MACJA,WAAS,OAAOA,YAAU,YAAY,aAAaA,UAC9CA,UACD,IAAI,MAAM,OAAOA,QAAM;AAG7B,QAAI,CAAC,qBAAqB,IAAI,CAE5B,QAAO,cAAc,KAAK,aAAa;AAIzC,QAAI,UAAU,YAAY;KAExB,MAAM,QAAQ,oBAAoB,aAAa,QAAQ;AACvD,SAAI,QAAQ,GACV,MAAM,MAAM,MAAM;IAGrB,MAEC,QAAO,cAAc,KAAK,aAAa;GAE1C;AAIH,SAAM,IAAI,MAAM;EACjB;CACF,EAAC;AACH"}

package/dist/agents/middleware/{promptCaching.cjs → provider/anthropic/promptCaching.cjs}

@@ -1,8 +1,8 @@
-const require_rolldown_runtime = require('../../_virtual/rolldown_runtime.cjs');
-const require_middleware = require('../middleware.cjs');
+const require_rolldown_runtime = require('../../../../_virtual/rolldown_runtime.cjs');
+const require_middleware = require('../../../middleware.cjs');
 const zod_v3 = require_rolldown_runtime.__toESM(require("zod/v3"));
 
-//#region src/agents/middleware/promptCaching.ts
+//#region src/agents/middleware/provider/anthropic/promptCaching.ts
 const DEFAULT_ENABLE_CACHING = true;
 const DEFAULT_TTL = "5m";
 const DEFAULT_MIN_MESSAGES_TO_CACHE = 3;

package/dist/agents/middleware/provider/anthropic/promptCaching.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"promptCaching.cjs","names":["z","message: string","middlewareOptions?: PromptCachingMiddlewareConfig","createMiddleware"],"sources":["../../../../../src/agents/middleware/provider/anthropic/promptCaching.ts"],"sourcesContent":["import { z } from \"zod/v3\";\nimport { ContentBlock } from \"@langchain/core/messages\";\nimport { InferInteropZodInput } from \"@langchain/core/utils/types\";\n\nimport { ConfigurableModel } from \"../../../../chat_models/universal.js\";\nimport { createMiddleware } from \"../../../middleware.js\";\n\nconst DEFAULT_ENABLE_CACHING = true;\nconst DEFAULT_TTL = \"5m\";\nconst DEFAULT_MIN_MESSAGES_TO_CACHE = 3;\nconst DEFAULT_UNSUPPORTED_MODEL_BEHAVIOR = \"warn\";\n\nconst contextSchema = z.object({\n  /**\n   * Whether to enable prompt caching.\n   * @default true\n   */\n  enableCaching: z.boolean().optional(),\n  /**\n   * The time-to-live for the cached prompt.\n   * @default \"5m\"\n   */\n  ttl: z.enum([\"5m\", \"1h\"]).optional(),\n  /**\n   * The minimum number of messages required before caching is applied.\n   * @default 3\n   */\n  minMessagesToCache: z.number().optional(),\n  /**\n   * The behavior to take when an unsupported model is used.\n   * - \"ignore\" will ignore the unsupported model and continue without caching.\n   * - \"warn\" will warn the user and continue without caching.\n   * - \"raise\" will raise an error and stop the agent.\n   * @default \"warn\"\n   */\n  unsupportedModelBehavior: z.enum([\"ignore\", \"warn\", \"raise\"]).optional(),\n});\nexport type PromptCachingMiddlewareConfig = Partial<\n  InferInteropZodInput<typeof contextSchema>\n>;\n\nclass PromptCachingMiddlewareError extends Error {\n  constructor(message: string) {\n    super(message);\n    this.name = \"PromptCachingMiddlewareError\";\n  }\n}\n\n/**\n * Creates a prompt caching middleware for Anthropic models to optimize API usage.\n *\n * This middleware automatically adds cache control headers to the last messages when using Anthropic models,\n * enabling their prompt caching feature. This can significantly reduce costs for applications with repetitive\n * prompts, long system messages, or extensive conversation histories.\n *\n * ## How It Works\n *\n * The middleware intercepts model requests and adds cache control metadata that tells Anthropic's\n * API to cache processed prompt prefixes. On subsequent requests with matching prefixes, the\n * cached representations are reused, skipping redundant token processing.\n *\n * ## Benefits\n *\n * - **Cost Reduction**: Avoid reprocessing the same tokens repeatedly (up to 90% savings on cached portions)\n * - **Lower Latency**: Cached prompts are processed faster as embeddings are pre-computed\n * - **Better Scalability**: Reduced computational load enables handling more requests\n * - **Consistent Performance**: Stable response times for repetitive queries\n *\n * @param middlewareOptions - Configuration options for the caching behavior\n * @param middlewareOptions.enableCaching - Whether to enable prompt caching (default: `true`)\n * @param middlewareOptions.ttl - Cache time-to-live: `\"5m\"` for 5 minutes or `\"1h\"` for 1 hour (default: `\"5m\"`)\n * @param middlewareOptions.minMessagesToCache - Minimum number of messages required before caching is applied (default: `3`)\n * @param middlewareOptions.unsupportedModelBehavior - The behavior to take when an unsupported model is used (default: `\"warn\"`)\n *\n * @returns A middleware instance that can be passed to `createAgent`\n *\n * @throws {Error} If used with non-Anthropic models\n *\n * @example\n * Basic usage with default settings\n * ```typescript\n * import { createAgent } from \"langchain\";\n * import { anthropicPromptCachingMiddleware } from \"langchain\";\n *\n * const agent = createAgent({\n *   model: \"anthropic:claude-3-5-sonnet\",\n *   middleware: [\n *     anthropicPromptCachingMiddleware()\n *   ]\n * });\n * ```\n *\n * @example\n * Custom configuration for longer conversations\n * ```typescript\n * const cachingMiddleware = anthropicPromptCachingMiddleware({\n *   ttl: \"1h\",  // Cache for 1 hour instead of default 5 minutes\n *   minMessagesToCache: 5  // Only cache after 5 messages\n * });\n *\n * const agent = createAgent({\n *   model: \"anthropic:claude-3-5-sonnet\",\n *   systemPrompt: \"You are a helpful assistant with deep knowledge of...\", // Long system prompt\n *   middleware: [cachingMiddleware]\n * });\n * ```\n *\n * @example\n * Conditional caching based on runtime context\n * ```typescript\n * const agent = createAgent({\n *   model: \"anthropic:claude-3-5-sonnet\",\n *   middleware: [\n *     anthropicPromptCachingMiddleware({\n *       enableCaching: true,\n *       ttl: \"5m\"\n *     })\n *   ]\n * });\n *\n * // Disable caching for specific requests\n * await agent.invoke(\n *   { messages: [new HumanMessage(\"Process this without caching\")] },\n *   {\n *     configurable: {\n *       middleware_context: { enableCaching: false }\n *     }\n *   }\n * );\n * ```\n *\n * @example\n * Optimal setup for customer support chatbot\n * ```typescript\n * const supportAgent = createAgent({\n *   model: \"anthropic:claude-3-5-sonnet\",\n *   systemPrompt: `You are a customer support agent for ACME Corp.\n *\n *     Company policies:\n *     - Always be polite and professional\n *     - Refer to knowledge base for product information\n *     - Escalate billing issues to human agents\n *     ... (extensive policies and guidelines)\n *   `,\n *   tools: [searchKnowledgeBase, createTicket, checkOrderStatus],\n *   middleware: [\n *     anthropicPromptCachingMiddleware({\n *       ttl: \"1h\",  // Long TTL for stable system prompt\n *       minMessagesToCache: 1  // Cache immediately due to large system prompt\n *     })\n *   ]\n * });\n * ```\n *\n * @remarks\n * - **Anthropic Only**: This middleware only works with Anthropic models and will throw an error if used with other providers\n * - **Automatic Application**: Caching is applied automatically when message count exceeds `minMessagesToCache`\n * - **Cache Scope**: Caches are isolated per API key and cannot be shared across different keys\n * - **TTL Options**: Only supports \"5m\" (5 minutes) and \"1h\" (1 hour) as TTL values per Anthropic's API\n * - **Best Use Cases**: Long system prompts, multi-turn conversations, repetitive queries, RAG applications\n * - **Cost Impact**: Cached tokens are billed at 10% of the base input token price, cache writes are billed at 25% of the base\n *\n * @see {@link createAgent} for agent creation\n * @see {@link https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching} Anthropic's prompt caching documentation\n * @public\n */\nexport function anthropicPromptCachingMiddleware(\n  middlewareOptions?: PromptCachingMiddlewareConfig\n) {\n  return createMiddleware({\n    name: \"PromptCachingMiddleware\",\n    contextSchema,\n    wrapModelCall: (request, handler) => {\n      /**\n       * Prefer runtime context values over middleware options values over defaults\n       */\n      const enableCaching =\n        request.runtime.context.enableCaching ??\n        middlewareOptions?.enableCaching ??\n        DEFAULT_ENABLE_CACHING;\n      const ttl =\n        request.runtime.context.ttl ?? middlewareOptions?.ttl ?? DEFAULT_TTL;\n      const minMessagesToCache =\n        request.runtime.context.minMessagesToCache ??\n        middlewareOptions?.minMessagesToCache ??\n        DEFAULT_MIN_MESSAGES_TO_CACHE;\n      const unsupportedModelBehavior =\n        request.runtime.context.unsupportedModelBehavior ??\n        middlewareOptions?.unsupportedModelBehavior ??\n        DEFAULT_UNSUPPORTED_MODEL_BEHAVIOR;\n\n      // Skip if caching is disabled\n      if (!enableCaching || !request.model) {\n        return handler(request);\n      }\n\n      const isAnthropicModel =\n        request.model.getName() === \"ChatAnthropic\" ||\n        (request.model.getName() === \"ConfigurableModel\" &&\n          (request.model as ConfigurableModel)._defaultConfig?.modelProvider ===\n            \"anthropic\");\n      if (!isAnthropicModel) {\n        // Get model name for better error context\n        const modelName = request.model.getName();\n        const modelInfo =\n          request.model.getName() === \"ConfigurableModel\"\n            ? `${modelName} (${\n                (request.model as ConfigurableModel)._defaultConfig\n                  ?.modelProvider\n              })`\n            : modelName;\n\n        const baseMessage = `Unsupported model '${modelInfo}'. Prompt caching requires an Anthropic model`;\n\n        if (unsupportedModelBehavior === \"raise\") {\n          throw new PromptCachingMiddlewareError(\n            `${baseMessage} (e.g., 'anthropic:claude-4-0-sonnet').`\n          );\n        } else if (unsupportedModelBehavior === \"warn\") {\n          console.warn(\n            `PromptCachingMiddleware: Skipping caching for ${modelName}. Consider switching to an Anthropic model for caching benefits.`\n          );\n        }\n        return handler(request);\n      }\n\n      const messagesCount =\n        request.state.messages.length + (request.systemPrompt ? 1 : 0);\n\n      if (messagesCount < minMessagesToCache) {\n        return handler(request);\n      }\n\n      /**\n       * Add cache_control to the last message\n       */\n      const lastMessage = request.messages.at(-1);\n      if (!lastMessage) {\n        return handler(request);\n      }\n\n      const NewMessageConstructor =\n        Object.getPrototypeOf(lastMessage).constructor;\n      if (Array.isArray(lastMessage.content)) {\n        const newMessage = new NewMessageConstructor({\n          ...lastMessage,\n          content: [\n            ...lastMessage.content.slice(0, -1),\n            {\n              ...lastMessage.content.at(-1),\n              cache_control: {\n                type: \"ephemeral\",\n                ttl,\n              },\n            } as ContentBlock,\n          ],\n        });\n        return handler({\n          ...request,\n          messages: [...request.messages.slice(0, -1), newMessage],\n        });\n      } else if (typeof lastMessage.content === \"string\") {\n        const newMessage = new NewMessageConstructor({\n          ...lastMessage,\n          content: [\n            {\n              type: \"text\",\n              text: lastMessage.content,\n              cache_control: {\n                type: \"ephemeral\",\n                ttl,\n              },\n            },\n          ],\n        });\n        return handler({\n          ...request,\n          messages: [...request.messages.slice(0, -1), newMessage],\n        });\n      }\n\n      throw new PromptCachingMiddlewareError(\n        \"Last message content is not a string or array\"\n      );\n    },\n  });\n}\n"],"mappings":";;;;;AAOA,MAAM,yBAAyB;AAC/B,MAAM,cAAc;AACpB,MAAM,gCAAgC;AACtC,MAAM,qCAAqC;AAE3C,MAAM,gBAAgBA,SAAE,OAAO;CAK7B,eAAeA,SAAE,SAAS,CAAC,UAAU;CAKrC,KAAKA,SAAE,KAAK,CAAC,MAAM,IAAK,EAAC,CAAC,UAAU;CAKpC,oBAAoBA,SAAE,QAAQ,CAAC,UAAU;CAQzC,0BAA0BA,SAAE,KAAK;EAAC;EAAU;EAAQ;CAAQ,EAAC,CAAC,UAAU;AACzE,EAAC;AAKF,IAAM,+BAAN,cAA2C,MAAM;CAC/C,YAAYC,SAAiB;EAC3B,MAAM,QAAQ;EACd,KAAK,OAAO;CACb;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwHD,SAAgB,iCACdC,mBACA;AACA,QAAOC,oCAAiB;EACtB,MAAM;EACN;EACA,eAAe,CAAC,SAAS,YAAY;;;;GAInC,MAAM,gBACJ,QAAQ,QAAQ,QAAQ,iBACxB,mBAAmB,iBACnB;GACF,MAAM,MACJ,QAAQ,QAAQ,QAAQ,OAAO,mBAAmB,OAAO;GAC3D,MAAM,qBACJ,QAAQ,QAAQ,QAAQ,sBACxB,mBAAmB,sBACnB;GACF,MAAM,2BACJ,QAAQ,QAAQ,QAAQ,4BACxB,mBAAmB,4BACnB;AAGF,OAAI,CAAC,iBAAiB,CAAC,QAAQ,MAC7B,QAAO,QAAQ,QAAQ;GAGzB,MAAM,mBACJ,QAAQ,MAAM,SAAS,KAAK,mBAC3B,QAAQ,MAAM,SAAS,KAAK,uBAC1B,QAAQ,MAA4B,gBAAgB,kBACnD;AACN,OAAI,CAAC,kBAAkB;IAErB,MAAM,YAAY,QAAQ,MAAM,SAAS;IACzC,MAAM,YACJ,QAAQ,MAAM,SAAS,KAAK,sBACxB,GAAG,UAAU,EAAE,EACZ,QAAQ,MAA4B,gBACjC,cACL,CAAC,CAAC,GACH;IAEN,MAAM,cAAc,CAAC,mBAAmB,EAAE,UAAU,6CAA6C,CAAC;AAElG,QAAI,6BAA6B,QAC/B,OAAM,IAAI,6BACR,GAAG,YAAY,uCAAuC,CAAC;aAEhD,6BAA6B,QACtC,QAAQ,KACN,CAAC,8CAA8C,EAAE,UAAU,gEAAgE,CAAC,CAC7H;AAEH,WAAO,QAAQ,QAAQ;GACxB;GAED,MAAM,gBACJ,QAAQ,MAAM,SAAS,UAAU,QAAQ,eAAe,IAAI;AAE9D,OAAI,gBAAgB,mBAClB,QAAO,QAAQ,QAAQ;;;;GAMzB,MAAM,cAAc,QAAQ,SAAS,GAAG,GAAG;AAC3C,OAAI,CAAC,YACH,QAAO,QAAQ,QAAQ;GAGzB,MAAM,wBACJ,OAAO,eAAe,YAAY,CAAC;AACrC,OAAI,MAAM,QAAQ,YAAY,QAAQ,EAAE;IACtC,MAAM,aAAa,IAAI,sBAAsB;KAC3C,GAAG;KACH,SAAS,CACP,GAAG,YAAY,QAAQ,MAAM,GAAG,GAAG,EACnC;MACE,GAAG,YAAY,QAAQ,GAAG,GAAG;MAC7B,eAAe;OACb,MAAM;OACN;MACD;KACF,CACF;IACF;AACD,WAAO,QAAQ;KACb,GAAG;KACH,UAAU,CAAC,GAAG,QAAQ,SAAS,MAAM,GAAG,GAAG,EAAE,UAAW;IACzD,EAAC;GACH,WAAU,OAAO,YAAY,YAAY,UAAU;IAClD,MAAM,aAAa,IAAI,sBAAsB;KAC3C,GAAG;KACH,SAAS,CACP;MACE,MAAM;MACN,MAAM,YAAY;MAClB,eAAe;OACb,MAAM;OACN;MACD;KACF,CACF;IACF;AACD,WAAO,QAAQ;KACb,GAAG;KACH,UAAU,CAAC,GAAG,QAAQ,SAAS,MAAM,GAAG,GAAG,EAAE,UAAW;IACzD,EAAC;GACH;AAED,SAAM,IAAI,6BACR;EAEH;CACF,EAAC;AACH"}

package/dist/agents/middleware/{promptCaching.d.cts → provider/anthropic/promptCaching.d.cts}

@@ -1,8 +1,8 @@
-import { AgentMiddleware } from "./types.cjs";
+import { AgentMiddleware } from "../../types.cjs";
 import { InferInteropZodInput } from "@langchain/core/utils/types";
 import { z } from "zod/v3";
 
-//#region src/agents/middleware/promptCaching.d.ts
+//#region src/agents/middleware/provider/anthropic/promptCaching.d.ts
 declare const contextSchema: z.ZodObject<{
   /**
    * Whether to enable prompt caching.

package/dist/agents/middleware/{promptCaching.d.ts → provider/anthropic/promptCaching.d.ts}

@@ -1,8 +1,8 @@
-import { AgentMiddleware } from "./types.js";
+import { AgentMiddleware } from "../../types.js";
 import { z } from "zod/v3";
 import { InferInteropZodInput } from "@langchain/core/utils/types";
 
-//#region src/agents/middleware/promptCaching.d.ts
+//#region src/agents/middleware/provider/anthropic/promptCaching.d.ts
 declare const contextSchema: z.ZodObject<{
   /**
    * Whether to enable prompt caching.

package/dist/agents/middleware/{promptCaching.js → provider/anthropic/promptCaching.js}

@@ -1,7 +1,7 @@
-import { createMiddleware } from "../middleware.js";
+import { createMiddleware } from "../../../middleware.js";
 import { z } from "zod/v3";
 
-//#region src/agents/middleware/promptCaching.ts
+//#region src/agents/middleware/provider/anthropic/promptCaching.ts
 const DEFAULT_ENABLE_CACHING = true;
 const DEFAULT_TTL = "5m";
 const DEFAULT_MIN_MESSAGES_TO_CACHE = 3;

package/dist/agents/middleware/provider/anthropic/promptCaching.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"promptCaching.js","names":["message: string","middlewareOptions?: PromptCachingMiddlewareConfig"],"sources":["../../../../../src/agents/middleware/provider/anthropic/promptCaching.ts"],"sourcesContent":["import { z } from \"zod/v3\";\nimport { ContentBlock } from \"@langchain/core/messages\";\nimport { InferInteropZodInput } from \"@langchain/core/utils/types\";\n\nimport { ConfigurableModel } from \"../../../../chat_models/universal.js\";\nimport { createMiddleware } from \"../../../middleware.js\";\n\nconst DEFAULT_ENABLE_CACHING = true;\nconst DEFAULT_TTL = \"5m\";\nconst DEFAULT_MIN_MESSAGES_TO_CACHE = 3;\nconst DEFAULT_UNSUPPORTED_MODEL_BEHAVIOR = \"warn\";\n\nconst contextSchema = z.object({\n  /**\n   * Whether to enable prompt caching.\n   * @default true\n   */\n  enableCaching: z.boolean().optional(),\n  /**\n   * The time-to-live for the cached prompt.\n   * @default \"5m\"\n   */\n  ttl: z.enum([\"5m\", \"1h\"]).optional(),\n  /**\n   * The minimum number of messages required before caching is applied.\n   * @default 3\n   */\n  minMessagesToCache: z.number().optional(),\n  /**\n   * The behavior to take when an unsupported model is used.\n   * - \"ignore\" will ignore the unsupported model and continue without caching.\n   * - \"warn\" will warn the user and continue without caching.\n   * - \"raise\" will raise an error and stop the agent.\n   * @default \"warn\"\n   */\n  unsupportedModelBehavior: z.enum([\"ignore\", \"warn\", \"raise\"]).optional(),\n});\nexport type PromptCachingMiddlewareConfig = Partial<\n  InferInteropZodInput<typeof contextSchema>\n>;\n\nclass PromptCachingMiddlewareError extends Error {\n  constructor(message: string) {\n    super(message);\n    this.name = \"PromptCachingMiddlewareError\";\n  }\n}\n\n/**\n * Creates a prompt caching middleware for Anthropic models to optimize API usage.\n *\n * This middleware automatically adds cache control headers to the last messages when using Anthropic models,\n * enabling their prompt caching feature. This can significantly reduce costs for applications with repetitive\n * prompts, long system messages, or extensive conversation histories.\n *\n * ## How It Works\n *\n * The middleware intercepts model requests and adds cache control metadata that tells Anthropic's\n * API to cache processed prompt prefixes. On subsequent requests with matching prefixes, the\n * cached representations are reused, skipping redundant token processing.\n *\n * ## Benefits\n *\n * - **Cost Reduction**: Avoid reprocessing the same tokens repeatedly (up to 90% savings on cached portions)\n * - **Lower Latency**: Cached prompts are processed faster as embeddings are pre-computed\n * - **Better Scalability**: Reduced computational load enables handling more requests\n * - **Consistent Performance**: Stable response times for repetitive queries\n *\n * @param middlewareOptions - Configuration options for the caching behavior\n * @param middlewareOptions.enableCaching - Whether to enable prompt caching (default: `true`)\n * @param middlewareOptions.ttl - Cache time-to-live: `\"5m\"` for 5 minutes or `\"1h\"` for 1 hour (default: `\"5m\"`)\n * @param middlewareOptions.minMessagesToCache - Minimum number of messages required before caching is applied (default: `3`)\n * @param middlewareOptions.unsupportedModelBehavior - The behavior to take when an unsupported model is used (default: `\"warn\"`)\n *\n * @returns A middleware instance that can be passed to `createAgent`\n *\n * @throws {Error} If used with non-Anthropic models\n *\n * @example\n * Basic usage with default settings\n * ```typescript\n * import { createAgent } from \"langchain\";\n * import { anthropicPromptCachingMiddleware } from \"langchain\";\n *\n * const agent = createAgent({\n *   model: \"anthropic:claude-3-5-sonnet\",\n *   middleware: [\n *     anthropicPromptCachingMiddleware()\n *   ]\n * });\n * ```\n *\n * @example\n * Custom configuration for longer conversations\n * ```typescript\n * const cachingMiddleware = anthropicPromptCachingMiddleware({\n *   ttl: \"1h\",  // Cache for 1 hour instead of default 5 minutes\n *   minMessagesToCache: 5  // Only cache after 5 messages\n * });\n *\n * const agent = createAgent({\n *   model: \"anthropic:claude-3-5-sonnet\",\n *   systemPrompt: \"You are a helpful assistant with deep knowledge of...\", // Long system prompt\n *   middleware: [cachingMiddleware]\n * });\n * ```\n *\n * @example\n * Conditional caching based on runtime context\n * ```typescript\n * const agent = createAgent({\n *   model: \"anthropic:claude-3-5-sonnet\",\n *   middleware: [\n *     anthropicPromptCachingMiddleware({\n *       enableCaching: true,\n *       ttl: \"5m\"\n *     })\n *   ]\n * });\n *\n * // Disable caching for specific requests\n * await agent.invoke(\n *   { messages: [new HumanMessage(\"Process this without caching\")] },\n *   {\n *     configurable: {\n *       middleware_context: { enableCaching: false }\n *     }\n *   }\n * );\n * ```\n *\n * @example\n * Optimal setup for customer support chatbot\n * ```typescript\n * const supportAgent = createAgent({\n *   model: \"anthropic:claude-3-5-sonnet\",\n *   systemPrompt: `You are a customer support agent for ACME Corp.\n *\n *     Company policies:\n *     - Always be polite and professional\n *     - Refer to knowledge base for product information\n *     - Escalate billing issues to human agents\n *     ... (extensive policies and guidelines)\n *   `,\n *   tools: [searchKnowledgeBase, createTicket, checkOrderStatus],\n *   middleware: [\n *     anthropicPromptCachingMiddleware({\n *       ttl: \"1h\",  // Long TTL for stable system prompt\n *       minMessagesToCache: 1  // Cache immediately due to large system prompt\n *     })\n *   ]\n * });\n * ```\n *\n * @remarks\n * - **Anthropic Only**: This middleware only works with Anthropic models and will throw an error if used with other providers\n * - **Automatic Application**: Caching is applied automatically when message count exceeds `minMessagesToCache`\n * - **Cache Scope**: Caches are isolated per API key and cannot be shared across different keys\n * - **TTL Options**: Only supports \"5m\" (5 minutes) and \"1h\" (1 hour) as TTL values per Anthropic's API\n * - **Best Use Cases**: Long system prompts, multi-turn conversations, repetitive queries, RAG applications\n * - **Cost Impact**: Cached tokens are billed at 10% of the base input token price, cache writes are billed at 25% of the base\n *\n * @see {@link createAgent} for agent creation\n * @see {@link https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching} Anthropic's prompt caching documentation\n * @public\n */\nexport function anthropicPromptCachingMiddleware(\n  middlewareOptions?: PromptCachingMiddlewareConfig\n) {\n  return createMiddleware({\n    name: \"PromptCachingMiddleware\",\n    contextSchema,\n    wrapModelCall: (request, handler) => {\n      /**\n       * Prefer runtime context values over middleware options values over defaults\n       */\n      const enableCaching =\n        request.runtime.context.enableCaching ??\n        middlewareOptions?.enableCaching ??\n        DEFAULT_ENABLE_CACHING;\n      const ttl =\n        request.runtime.context.ttl ?? middlewareOptions?.ttl ?? DEFAULT_TTL;\n      const minMessagesToCache =\n        request.runtime.context.minMessagesToCache ??\n        middlewareOptions?.minMessagesToCache ??\n        DEFAULT_MIN_MESSAGES_TO_CACHE;\n      const unsupportedModelBehavior =\n        request.runtime.context.unsupportedModelBehavior ??\n        middlewareOptions?.unsupportedModelBehavior ??\n        DEFAULT_UNSUPPORTED_MODEL_BEHAVIOR;\n\n      // Skip if caching is disabled\n      if (!enableCaching || !request.model) {\n        return handler(request);\n      }\n\n      const isAnthropicModel =\n        request.model.getName() === \"ChatAnthropic\" ||\n        (request.model.getName() === \"ConfigurableModel\" &&\n          (request.model as ConfigurableModel)._defaultConfig?.modelProvider ===\n            \"anthropic\");\n      if (!isAnthropicModel) {\n        // Get model name for better error context\n        const modelName = request.model.getName();\n        const modelInfo =\n          request.model.getName() === \"ConfigurableModel\"\n            ? `${modelName} (${\n                (request.model as ConfigurableModel)._defaultConfig\n                  ?.modelProvider\n              })`\n            : modelName;\n\n        const baseMessage = `Unsupported model '${modelInfo}'. Prompt caching requires an Anthropic model`;\n\n        if (unsupportedModelBehavior === \"raise\") {\n          throw new PromptCachingMiddlewareError(\n            `${baseMessage} (e.g., 'anthropic:claude-4-0-sonnet').`\n          );\n        } else if (unsupportedModelBehavior === \"warn\") {\n          console.warn(\n            `PromptCachingMiddleware: Skipping caching for ${modelName}. Consider switching to an Anthropic model for caching benefits.`\n          );\n        }\n        return handler(request);\n      }\n\n      const messagesCount =\n        request.state.messages.length + (request.systemPrompt ? 1 : 0);\n\n      if (messagesCount < minMessagesToCache) {\n        return handler(request);\n      }\n\n      /**\n       * Add cache_control to the last message\n       */\n      const lastMessage = request.messages.at(-1);\n      if (!lastMessage) {\n        return handler(request);\n      }\n\n      const NewMessageConstructor =\n        Object.getPrototypeOf(lastMessage).constructor;\n      if (Array.isArray(lastMessage.content)) {\n        const newMessage = new NewMessageConstructor({\n          ...lastMessage,\n          content: [\n            ...lastMessage.content.slice(0, -1),\n            {\n              ...lastMessage.content.at(-1),\n              cache_control: {\n                type: \"ephemeral\",\n                ttl,\n              },\n            } as ContentBlock,\n          ],\n        });\n        return handler({\n          ...request,\n          messages: [...request.messages.slice(0, -1), newMessage],\n        });\n      } else if (typeof lastMessage.content === \"string\") {\n        const newMessage = new NewMessageConstructor({\n          ...lastMessage,\n          content: [\n            {\n              type: \"text\",\n              text: lastMessage.content,\n              cache_control: {\n                type: \"ephemeral\",\n                ttl,\n              },\n            },\n          ],\n        });\n        return handler({\n          ...request,\n          messages: [...request.messages.slice(0, -1), newMessage],\n        });\n      }\n\n      throw new PromptCachingMiddlewareError(\n        \"Last message content is not a string or array\"\n      );\n    },\n  });\n}\n"],"mappings":";;;;AAOA,MAAM,yBAAyB;AAC/B,MAAM,cAAc;AACpB,MAAM,gCAAgC;AACtC,MAAM,qCAAqC;AAE3C,MAAM,gBAAgB,EAAE,OAAO;CAK7B,eAAe,EAAE,SAAS,CAAC,UAAU;CAKrC,KAAK,EAAE,KAAK,CAAC,MAAM,IAAK,EAAC,CAAC,UAAU;CAKpC,oBAAoB,EAAE,QAAQ,CAAC,UAAU;CAQzC,0BAA0B,EAAE,KAAK;EAAC;EAAU;EAAQ;CAAQ,EAAC,CAAC,UAAU;AACzE,EAAC;AAKF,IAAM,+BAAN,cAA2C,MAAM;CAC/C,YAAYA,SAAiB;EAC3B,MAAM,QAAQ;EACd,KAAK,OAAO;CACb;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwHD,SAAgB,iCACdC,mBACA;AACA,QAAO,iBAAiB;EACtB,MAAM;EACN;EACA,eAAe,CAAC,SAAS,YAAY;;;;GAInC,MAAM,gBACJ,QAAQ,QAAQ,QAAQ,iBACxB,mBAAmB,iBACnB;GACF,MAAM,MACJ,QAAQ,QAAQ,QAAQ,OAAO,mBAAmB,OAAO;GAC3D,MAAM,qBACJ,QAAQ,QAAQ,QAAQ,sBACxB,mBAAmB,sBACnB;GACF,MAAM,2BACJ,QAAQ,QAAQ,QAAQ,4BACxB,mBAAmB,4BACnB;AAGF,OAAI,CAAC,iBAAiB,CAAC,QAAQ,MAC7B,QAAO,QAAQ,QAAQ;GAGzB,MAAM,mBACJ,QAAQ,MAAM,SAAS,KAAK,mBAC3B,QAAQ,MAAM,SAAS,KAAK,uBAC1B,QAAQ,MAA4B,gBAAgB,kBACnD;AACN,OAAI,CAAC,kBAAkB;IAErB,MAAM,YAAY,QAAQ,MAAM,SAAS;IACzC,MAAM,YACJ,QAAQ,MAAM,SAAS,KAAK,sBACxB,GAAG,UAAU,EAAE,EACZ,QAAQ,MAA4B,gBACjC,cACL,CAAC,CAAC,GACH;IAEN,MAAM,cAAc,CAAC,mBAAmB,EAAE,UAAU,6CAA6C,CAAC;AAElG,QAAI,6BAA6B,QAC/B,OAAM,IAAI,6BACR,GAAG,YAAY,uCAAuC,CAAC;aAEhD,6BAA6B,QACtC,QAAQ,KACN,CAAC,8CAA8C,EAAE,UAAU,gEAAgE,CAAC,CAC7H;AAEH,WAAO,QAAQ,QAAQ;GACxB;GAED,MAAM,gBACJ,QAAQ,MAAM,SAAS,UAAU,QAAQ,eAAe,IAAI;AAE9D,OAAI,gBAAgB,mBAClB,QAAO,QAAQ,QAAQ;;;;GAMzB,MAAM,cAAc,QAAQ,SAAS,GAAG,GAAG;AAC3C,OAAI,CAAC,YACH,QAAO,QAAQ,QAAQ;GAGzB,MAAM,wBACJ,OAAO,eAAe,YAAY,CAAC;AACrC,OAAI,MAAM,QAAQ,YAAY,QAAQ,EAAE;IACtC,MAAM,aAAa,IAAI,sBAAsB;KAC3C,GAAG;KACH,SAAS,CACP,GAAG,YAAY,QAAQ,MAAM,GAAG,GAAG,EACnC;MACE,GAAG,YAAY,QAAQ,GAAG,GAAG;MAC7B,eAAe;OACb,MAAM;OACN;MACD;KACF,CACF;IACF;AACD,WAAO,QAAQ;KACb,GAAG;KACH,UAAU,CAAC,GAAG,QAAQ,SAAS,MAAM,GAAG,GAAG,EAAE,UAAW;IACzD,EAAC;GACH,WAAU,OAAO,YAAY,YAAY,UAAU;IAClD,MAAM,aAAa,IAAI,sBAAsB;KAC3C,GAAG;KACH,SAAS,CACP;MACE,MAAM;MACN,MAAM,YAAY;MAClB,eAAe;OACb,MAAM;OACN;MACD;KACF,CACF;IACF;AACD,WAAO,QAAQ;KACb,GAAG;KACH,UAAU,CAAC,GAAG,QAAQ,SAAS,MAAM,GAAG,GAAG,EAAE,UAAW;IACzD,EAAC;GACH;AAED,SAAM,IAAI,6BACR;EAEH;CACF,EAAC;AACH"}