@dvina/agents 0.3.7 → 0.5.0

package/dist/chunk-NHWEEBN2.mjs

@@ -0,0 +1,159 @@
+var __create = Object.create;
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __getProtoOf = Object.getPrototypeOf;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __require = /* @__PURE__ */ ((x) => typeof require !== "undefined" ? require : typeof Proxy !== "undefined" ? new Proxy(x, {
+  get: (a, b) => (typeof require !== "undefined" ? require : a)[b]
+}) : x)(function(x) {
+  if (typeof require !== "undefined") return require.apply(this, arguments);
+  throw Error('Dynamic require of "' + x + '" is not supported');
+});
+var __commonJS = (cb, mod) => function __require2() {
+  return mod || (0, cb[__getOwnPropNames(cb)[0]])((mod = { exports: {} }).exports, mod), mod.exports;
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
+  // If the importer is in node compatibility mode or this is not an ESM
+  // file that has been converted to a CommonJS file using a Babel-
+  // compatible transform (i.e. "__esModule" has not been set), then set
+  // "default" to the CommonJS "module.exports" for node compatibility.
+  isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
+  mod
+));
+
+// src/runtime/langchain/model-resolver.ts
+import { AzureChatOpenAI, ChatOpenAI } from "@langchain/openai";
+var LangchainModelResolver = class {
+  constructor(config) {
+    this.config = config;
+  }
+  resolve(modelString, tags) {
+    const parts = modelString.split(":");
+    if (parts.length === 2) {
+      const [provider, modelName] = parts;
+      return this.resolveByProvider(provider, "default", modelName, tags);
+    }
+    if (parts.length === 3) {
+      const [provider, configName, modelName] = parts;
+      return this.resolveByProvider(provider, configName, modelName, tags);
+    }
+    throw new Error(
+      'Model string must follow format "provider:modelName" (uses "default" config) or "provider:configName:modelName"'
+    );
+  }
+  resolveByProvider(provider, configName, modelName, tags) {
+    switch (provider) {
+      case "openai":
+        return this.resolveOpenAI(configName, modelName, tags);
+      case "azure":
+        return this.resolveAzure(configName, modelName, tags);
+      default:
+        throw new Error(`Unsupported model provider: ${provider}`);
+    }
+  }
+  resolveOpenAI(configName, modelName, tags) {
+    const providerConfig = this.config.openai?.[configName];
+    if (!providerConfig) {
+      throw new Error(`Configuration "${configName}" for provider "openai" is missing`);
+    }
+    return new ChatOpenAI({
+      apiKey: providerConfig.apiKey,
+      modelName,
+      tags
+    });
+  }
+  resolveAzure(resourceName, modelName, tags) {
+    const resource = this.config.azure?.[resourceName];
+    if (!resource) {
+      throw new Error(`Resource "${resourceName}" for provider "azure" is missing`);
+    }
+    const modelEntry = resource.models.find((m) => m.model === modelName);
+    if (!modelEntry) {
+      throw new Error(`Model "${modelName}" not found in Azure resource "${resourceName}"`);
+    }
+    return new AzureChatOpenAI({
+      model: modelEntry.model,
+      azureOpenAIApiKey: resource.apiKey,
+      azureOpenAIApiInstanceName: this.extractInstanceName(resource.endpoint),
+      azureOpenAIApiDeploymentName: modelEntry.deploymentName,
+      azureOpenAIApiVersion: modelEntry.apiVersion,
+      tags
+    });
+  }
+  extractInstanceName(endpoint) {
+    try {
+      const url = new URL(endpoint);
+      return url.hostname.split(".")[0];
+    } catch (e) {
+      return endpoint;
+    }
+  }
+};
+
+// src/runtime/langchain/utils.ts
+import { AIMessage, HumanMessage, ToolMessage } from "langchain";
+function convertToLangchainMessages(messages) {
+  const result = [];
+  let tcIdx = 0;
+  let pendingToolCallIds = [];
+  for (const msg of messages) {
+    if (msg.role === "human") {
+      result.push(
+        new HumanMessage({
+          content: msg.content.map((c) => {
+            if (c.type === "image") {
+              return { type: "image_url", image_url: { url: c.url } };
+            }
+            return c;
+          })
+        })
+      );
+    } else if (msg.role === "ai") {
+      if (msg.toolCalls && msg.toolCalls.length > 0) {
+        pendingToolCallIds = msg.toolCalls.map(() => `tc_${++tcIdx}`);
+        result.push(
+          new AIMessage({
+            content: msg.content,
+            tool_calls: msg.toolCalls.map((tc, i) => ({
+              id: pendingToolCallIds[i],
+              name: tc.name,
+              args: tc.input ? JSON.parse(tc.input) : {}
+            }))
+          })
+        );
+      } else {
+        result.push(new AIMessage(msg.content));
+      }
+    } else if (msg.role === "tool") {
+      const toolCallId = pendingToolCallIds.shift();
+      if (!toolCallId)
+        throw new Error(`ToolMessage for "${msg.name}" without a preceding AiMessage with toolCalls`);
+      result.push(
+        new ToolMessage({
+          content: msg.output,
+          tool_call_id: toolCallId,
+          name: msg.name
+        })
+      );
+    }
+  }
+  return result;
+}
+
+export {
+  __require,
+  __commonJS,
+  __toESM,
+  convertToLangchainMessages,
+  LangchainModelResolver
+};
+//# sourceMappingURL=chunk-NHWEEBN2.mjs.map

package/dist/chunk-NHWEEBN2.mjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/runtime/langchain/model-resolver.ts","../src/runtime/langchain/utils.ts"],"sourcesContent":["import { BaseLanguageModel } from '@langchain/core/language_models/base';\nimport { AzureChatOpenAI, ChatOpenAI } from '@langchain/openai';\n\nexport type LangchainOpenAIConfig = {\n\tapiKey: string;\n};\n\nexport type LangchainAzureResourceConfig = {\n\tapiKey: string;\n\tendpoint: string;\n\tmodels: {\n\t\tmodel: string;\n\t\tapiVersion: string;\n\t\tdeploymentName: string;\n\t}[];\n};\n\nexport type ResourceName = string;\n\nexport type LangchainModelConfig = {\n\topenai?: Record<string, LangchainOpenAIConfig>;\n\tazure?: Record<ResourceName, LangchainAzureResourceConfig>;\n};\n\nexport class LangchainModelResolver {\n\tconstructor(private config: LangchainModelConfig) {}\n\n\tresolve(modelString: string, tags?: string[]): BaseLanguageModel {\n\t\tconst parts = modelString.split(':');\n\n\t\tif (parts.length === 2) {\n\t\t\tconst [provider, modelName] = parts;\n\t\t\treturn this.resolveByProvider(provider, 'default', modelName, tags);\n\t\t}\n\n\t\tif (parts.length === 3) {\n\t\t\tconst [provider, configName, modelName] = parts;\n\t\t\treturn this.resolveByProvider(provider, configName, modelName, tags);\n\t\t}\n\n\t\tthrow new Error(\n\t\t\t'Model string must follow format \"provider:modelName\" (uses \"default\" config) or \"provider:configName:modelName\"',\n\t\t);\n\t}\n\n\tprivate resolveByProvider(\n\t\tprovider: string,\n\t\tconfigName: string,\n\t\tmodelName: string,\n\t\ttags?: string[],\n\t): BaseLanguageModel {\n\t\tswitch (provider) {\n\t\t\tcase 'openai':\n\t\t\t\treturn this.resolveOpenAI(configName, modelName, tags);\n\t\t\tcase 'azure':\n\t\t\t\treturn this.resolveAzure(configName, modelName, tags);\n\t\t\tdefault:\n\t\t\t\tthrow new Error(`Unsupported model provider: ${provider}`);\n\t\t}\n\t}\n\n\tprivate resolveOpenAI(configName: string, modelName: string, tags?: string[]): ChatOpenAI {\n\t\tconst providerConfig = this.config.openai?.[configName];\n\t\tif (!providerConfig) {\n\t\t\tthrow new Error(`Configuration \"${configName}\" for provider \"openai\" is missing`);\n\t\t}\n\n\t\treturn new ChatOpenAI({\n\t\t\tapiKey: providerConfig.apiKey,\n\t\t\tmodelName: modelName,\n\t\t\ttags: tags,\n\t\t});\n\t}\n\n\tprivate resolveAzure(resourceName: string, modelName: string, tags?: string[]): AzureChatOpenAI {\n\t\tconst resource = this.config.azure?.[resourceName];\n\t\tif (!resource) {\n\t\t\tthrow new Error(`Resource \"${resourceName}\" for provider \"azure\" is missing`);\n\t\t}\n\n\t\tconst modelEntry = resource.models.find((m) => m.model === modelName);\n\t\tif (!modelEntry) {\n\t\t\tthrow new Error(`Model \"${modelName}\" not found in Azure resource \"${resourceName}\"`);\n\t\t}\n\n\t\treturn new AzureChatOpenAI({\n\t\t\tmodel: modelEntry.model,\n\t\t\tazureOpenAIApiKey: resource.apiKey,\n\t\t\tazureOpenAIApiInstanceName: this.extractInstanceName(resource.endpoint),\n\t\t\tazureOpenAIApiDeploymentName: modelEntry.deploymentName,\n\t\t\tazureOpenAIApiVersion: modelEntry.apiVersion,\n\t\t\ttags: tags,\n\t\t});\n\t}\n\n\tprivate extractInstanceName(endpoint: string): string {\n\t\ttry {\n\t\t\tconst url = new URL(endpoint);\n\t\t\treturn url.hostname.split('.')[0];\n\t\t} catch (e) {\n\t\t\treturn endpoint;\n\t\t}\n\t}\n}\n","import { Message } from '@core/agent.interface';\nimport { AIMessage, BaseMessage, HumanMessage, ToolMessage } from 'langchain';\n\nexport function convertToLangchainMessages(messages: Message[]): BaseMessage[] {\n\tconst result: BaseMessage[] = [];\n\tlet tcIdx = 0;\n\tlet pendingToolCallIds: string[] = [];\n\n\tfor (const msg of messages) {\n\t\tif (msg.role === 'human') {\n\t\t\tresult.push(\n\t\t\t\tnew HumanMessage({\n\t\t\t\t\tcontent: msg.content.map((c) => {\n\t\t\t\t\t\tif (c.type === 'image') {\n\t\t\t\t\t\t\treturn { type: 'image_url', image_url: { url: c.url } };\n\t\t\t\t\t\t}\n\t\t\t\t\t\treturn c;\n\t\t\t\t\t}) as any,\n\t\t\t\t}),\n\t\t\t);\n\t\t} else if (msg.role === 'ai') {\n\t\t\tif (msg.toolCalls && msg.toolCalls.length > 0) {\n\t\t\t\tpendingToolCallIds = msg.toolCalls.map(() => `tc_${++tcIdx}`);\n\t\t\t\tresult.push(\n\t\t\t\t\tnew AIMessage({\n\t\t\t\t\t\tcontent: msg.content,\n\t\t\t\t\t\ttool_calls: msg.toolCalls.map((tc, i) => ({\n\t\t\t\t\t\t\tid: pendingToolCallIds[i],\n\t\t\t\t\t\t\tname: tc.name,\n\t\t\t\t\t\t\targs: tc.input ? JSON.parse(tc.input) : {},\n\t\t\t\t\t\t})),\n\t\t\t\t\t}),\n\t\t\t\t);\n\t\t\t} else {\n\t\t\t\tresult.push(new AIMessage(msg.content));\n\t\t\t}\n\t\t} else if (msg.role === 'tool') {\n\t\t\tconst toolCallId = pendingToolCallIds.shift();\n\t\t\tif (!toolCallId)\n\t\t\t\tthrow new Error(`ToolMessage for \"${msg.name}\" without a preceding AiMessage with toolCalls`);\n\t\t\tresult.push(\n\t\t\t\tnew ToolMessage({\n\t\t\t\t\tcontent: msg.output,\n\t\t\t\t\ttool_call_id: toolCallId,\n\t\t\t\t\tname: msg.name,\n\t\t\t\t}),\n\t\t\t);\n\t\t}\n\t}\n\n\treturn result;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AACA,SAAS,iBAAiB,kBAAkB;AAuBrC,IAAM,yBAAN,MAA6B;AAAA,EACnC,YAAoB,QAA8B;AAA9B;AAAA,EAA+B;AAAA,EAEnD,QAAQ,aAAqB,MAAoC;AAChE,UAAM,QAAQ,YAAY,MAAM,GAAG;AAEnC,QAAI,MAAM,WAAW,GAAG;AACvB,YAAM,CAAC,UAAU,SAAS,IAAI;AAC9B,aAAO,KAAK,kBAAkB,UAAU,WAAW,WAAW,IAAI;AAAA,IACnE;AAEA,QAAI,MAAM,WAAW,GAAG;AACvB,YAAM,CAAC,UAAU,YAAY,SAAS,IAAI;AAC1C,aAAO,KAAK,kBAAkB,UAAU,YAAY,WAAW,IAAI;AAAA,IACpE;AAEA,UAAM,IAAI;AAAA,MACT;AAAA,IACD;AAAA,EACD;AAAA,EAEQ,kBACP,UACA,YACA,WACA,MACoB;AACpB,YAAQ,UAAU;AAAA,MACjB,KAAK;AACJ,eAAO,KAAK,cAAc,YAAY,WAAW,IAAI;AAAA,MACtD,KAAK;AACJ,eAAO,KAAK,aAAa,YAAY,WAAW,IAAI;AAAA,MACrD;AACC,cAAM,IAAI,MAAM,+BAA+B,QAAQ,EAAE;AAAA,IAC3D;AAAA,EACD;AAAA,EAEQ,cAAc,YAAoB,WAAmB,MAA6B;AACzF,UAAM,iBAAiB,KAAK,OAAO,SAAS,UAAU;AACtD,QAAI,CAAC,gBAAgB;AACpB,YAAM,IAAI,MAAM,kBAAkB,UAAU,oCAAoC;AAAA,IACjF;AAEA,WAAO,IAAI,WAAW;AAAA,MACrB,QAAQ,eAAe;AAAA,MACvB;AAAA,MACA;AAAA,IACD,CAAC;AAAA,EACF;AAAA,EAEQ,aAAa,cAAsB,WAAmB,MAAkC;AAC/F,UAAM,WAAW,KAAK,OAAO,QAAQ,YAAY;AACjD,QAAI,CAAC,UAAU;AACd,YAAM,IAAI,MAAM,aAAa,YAAY,mCAAmC;AAAA,IAC7E;AAEA,UAAM,aAAa,SAAS,OAAO,KAAK,CAAC,MAAM,EAAE,UAAU,SAAS;AACpE,QAAI,CAAC,YAAY;AAChB,YAAM,IAAI,MAAM,UAAU,SAAS,kCAAkC,YAAY,GAAG;AAAA,IACrF;AAEA,WAAO,IAAI,gBAAgB;AAAA,MAC1B,OAAO,WAAW;AAAA,MAClB,mBAAmB,SAAS;AAAA,MAC5B,4BAA4B,KAAK,oBAAoB,SAAS,QAAQ;AAAA,MACtE,8BAA8B,WAAW;AAAA,MACzC,uBAAuB,WAAW;AAAA,MAClC;AAAA,IACD,CAAC;AAAA,EACF;AAAA,EAEQ,oBAAoB,UAA0B;AACrD,QAAI;AACH,YAAM,MAAM,IAAI,IAAI,QAAQ;AAC5B,aAAO,IAAI,SAAS,MAAM,GAAG,EAAE,CAAC;AAAA,IACjC,SAAS,GAAG;AACX,aAAO;AAAA,IACR;AAAA,EACD;AACD;;;ACtGA,SAAS,WAAwB,cAAc,mBAAmB;AAE3D,SAAS,2BAA2B,UAAoC;AAC9E,QAAM,SAAwB,CAAC;AAC/B,MAAI,QAAQ;AACZ,MAAI,qBAA+B,CAAC;AAEpC,aAAW,OAAO,UAAU;AAC3B,QAAI,IAAI,SAAS,SAAS;AACzB,aAAO;AAAA,QACN,IAAI,aAAa;AAAA,UAChB,SAAS,IAAI,QAAQ,IAAI,CAAC,MAAM;AAC/B,gBAAI,EAAE,SAAS,SAAS;AACvB,qBAAO,EAAE,MAAM,aAAa,WAAW,EAAE,KAAK,EAAE,IAAI,EAAE;AAAA,YACvD;AACA,mBAAO;AAAA,UACR,CAAC;AAAA,QACF,CAAC;AAAA,MACF;AAAA,IACD,WAAW,IAAI,SAAS,MAAM;AAC7B,UAAI,IAAI,aAAa,IAAI,UAAU,SAAS,GAAG;AAC9C,6BAAqB,IAAI,UAAU,IAAI,MAAM,MAAM,EAAE,KAAK,EAAE;AAC5D,eAAO;AAAA,UACN,IAAI,UAAU;AAAA,YACb,SAAS,IAAI;AAAA,YACb,YAAY,IAAI,UAAU,IAAI,CAAC,IAAI,OAAO;AAAA,cACzC,IAAI,mBAAmB,CAAC;AAAA,cACxB,MAAM,GAAG;AAAA,cACT,MAAM,GAAG,QAAQ,KAAK,MAAM,GAAG,KAAK,IAAI,CAAC;AAAA,YAC1C,EAAE;AAAA,UACH,CAAC;AAAA,QACF;AAAA,MACD,OAAO;AACN,eAAO,KAAK,IAAI,UAAU,IAAI,OAAO,CAAC;AAAA,MACvC;AAAA,IACD,WAAW,IAAI,SAAS,QAAQ;AAC/B,YAAM,aAAa,mBAAmB,MAAM;AAC5C,UAAI,CAAC;AACJ,cAAM,IAAI,MAAM,oBAAoB,IAAI,IAAI,gDAAgD;AAC7F,aAAO;AAAA,QACN,IAAI,YAAY;AAAA,UACf,SAAS,IAAI;AAAA,UACb,cAAc;AAAA,UACd,MAAM,IAAI;AAAA,QACX,CAAC;AAAA,MACF;AAAA,IACD;AAAA,EACD;AAEA,SAAO;AACR;","names":[]}

package/dist/eval/index.d.mts

@@ -0,0 +1,117 @@
+import { T as ToolDefinition, A as Agent, L as LangchainModelConfig, M as Message, d as AiMessage, a as ToolSpec, H as HumanMessage, f as ToolMessage } from '../model-resolver-U0J9x1a6.mjs';
+import * as zod from 'zod';
+import { z } from 'zod';
+import { BaseMessage } from '@langchain/core/messages';
+
+/** Factory that creates a fresh Agent per test case. Receives extra suite-level tools as ToolDefinition[]. */
+type CreateTargetFn = (extraTools: ToolDefinition[]) => Agent | Promise<Agent>;
+interface EvalConfig {
+    /** Required for model-based target and LLM evaluators (respondsInLanguage, llmJudge). */
+    modelConfig: LangchainModelConfig;
+    /** Required for model-based target. Also used as fallback for evaluatorModel. */
+    model?: string;
+    /** Model for evaluators needing LLM calls (language detection, LLM-as-judge). */
+    evaluatorModel: string;
+    /** System prompt for model-based target. Ignored when createTarget is used. Can be overridden per-suite or per-case. */
+    systemPrompt?: string;
+    /** Factory that creates a fresh Agent per test case. When set, this is the default target. */
+    createTarget?: CreateTargetFn;
+}
+declare function configureEvals(config: EvalConfig): void;
+
+interface MockToolDef {
+    name: string;
+    description: string;
+    schema: z.ZodObject<any> | Record<string, unknown>;
+    /**
+     * Canned response the mock tool returns.
+     * Can be a static string, or a function that receives input and returns a response.
+     * If a function is provided, it receives the full invocation count as a second arg
+     * to support scenarios like "first call fails, second call succeeds".
+     */
+    response: string | ((input: Record<string, unknown>, callCount: number) => string);
+}
+
+type EvaluatorFn = (args: {
+    outputs: Record<string, any>;
+    referenceOutputs: Record<string, any>;
+}) => Promise<any>;
+interface ResolvedExpectation {
+    evaluator: EvaluatorFn;
+    referenceOutputs: Record<string, unknown>;
+}
+/** A factory that receives test context and returns an evaluator + its referenceOutputs. */
+type Expectation = (ctx: {
+    message: string;
+}) => ResolvedExpectation;
+/**
+ * Expect the agent to call tools in order (superset trajectory match).
+ * Empty `[]` means the agent should answer directly without calling any tools.
+ */
+declare function toolsCalled(tools: string[]): Expectation;
+/**
+ * Run an LLM-as-judge evaluator on the trajectory.
+ * Requires `toolsCalled` in the same expect array.
+ * Uses the globally configured evaluator model.
+ */
+declare function llmJudge(): Expectation;
+/** Assert the agent made zero tool calls. */
+declare function noTools(): Expectation;
+/**
+ * Assert the response is in the given language (ISO 639-1 code).
+ * Uses the globally configured evaluator model for language detection.
+ * @param code - ISO 639-1 language code (e.g. 'en', 'tr', 'de').
+ */
+declare function respondsInLanguage(code: string): Expectation;
+/** Assert the response contains all given strings. */
+declare function contains(strings: string[]): Expectation;
+/** Assert the response does not contain any of the given strings. */
+declare function notContains(strings: string[]): Expectation;
+
+declare function human(content: string): HumanMessage;
+declare function ai(content: string, toolCalls?: string[]): AiMessage;
+declare function toolResult(name: string, output: string): ToolMessage;
+interface ToolDef {
+    description: string;
+    /** A plain key→description record, or a ZodObject passed through from a ToolSpec. */
+    schema?: Record<string, string> | zod.ZodObject<any>;
+    /** Auto-stringified if not a string or function. */
+    response: unknown | ((input: Record<string, unknown>, callCount: number) => string);
+}
+interface TestCase {
+    /** Test name. Defaults to the last human message content if omitted. */
+    name?: string;
+    messages: Message[];
+    systemPrompt?: string;
+    /** Override suite-level tools for this case. */
+    tools?: Record<string, ToolDef>;
+    expect: Expectation[];
+}
+type TargetFn = (inputs: {
+    systemPrompt?: string;
+    messages: Message[];
+    tools: MockToolDef[];
+}) => Promise<{
+    messages: BaseMessage[];
+}>;
+interface SuiteConfig {
+    /** Custom target function, or model string override. Auto-created from global config if omitted. */
+    target?: TargetFn | string;
+    /** Factory that creates a fresh Agent per test case. Overrides global createTarget. */
+    createTarget?: CreateTargetFn;
+    /** System prompt for all cases in this suite. Overrides the global prompt; can be overridden per-case. */
+    systemPrompt?: string;
+    tools?: Record<string, ToolDef>;
+    cases: TestCase[];
+}
+/**
+ * Converts a `ToolSpec[]` (from a real tool provider) into the
+ * `Record<string, ToolDef>` that `defineSuite` expects.
+ *
+ * `responses` maps tool names to canned mock responses. Tools without an
+ * entry in `responses` default to `''`.
+ */
+declare function fromToolSpecs(specs: ToolSpec[], responses?: Record<string, ToolDef['response']>): Record<string, ToolDef>;
+declare function defineSuite(name: string, config: SuiteConfig): void;
+
+export { type CreateTargetFn, type EvalConfig, type Expectation, type SuiteConfig, type TestCase, type ToolDef, ai, configureEvals, contains, defineSuite, fromToolSpecs, human, llmJudge, noTools, notContains, respondsInLanguage, toolResult, toolsCalled };

package/dist/eval/index.d.ts

@@ -0,0 +1,117 @@
+import { T as ToolDefinition, A as Agent, L as LangchainModelConfig, M as Message, d as AiMessage, a as ToolSpec, H as HumanMessage, f as ToolMessage } from '../model-resolver-U0J9x1a6.js';
+import * as zod from 'zod';
+import { z } from 'zod';
+import { BaseMessage } from '@langchain/core/messages';
+
+/** Factory that creates a fresh Agent per test case. Receives extra suite-level tools as ToolDefinition[]. */
+type CreateTargetFn = (extraTools: ToolDefinition[]) => Agent | Promise<Agent>;
+interface EvalConfig {
+    /** Required for model-based target and LLM evaluators (respondsInLanguage, llmJudge). */
+    modelConfig: LangchainModelConfig;
+    /** Required for model-based target. Also used as fallback for evaluatorModel. */
+    model?: string;
+    /** Model for evaluators needing LLM calls (language detection, LLM-as-judge). */
+    evaluatorModel: string;
+    /** System prompt for model-based target. Ignored when createTarget is used. Can be overridden per-suite or per-case. */
+    systemPrompt?: string;
+    /** Factory that creates a fresh Agent per test case. When set, this is the default target. */
+    createTarget?: CreateTargetFn;
+}
+declare function configureEvals(config: EvalConfig): void;
+
+interface MockToolDef {
+    name: string;
+    description: string;
+    schema: z.ZodObject<any> | Record<string, unknown>;
+    /**
+     * Canned response the mock tool returns.
+     * Can be a static string, or a function that receives input and returns a response.
+     * If a function is provided, it receives the full invocation count as a second arg
+     * to support scenarios like "first call fails, second call succeeds".
+     */
+    response: string | ((input: Record<string, unknown>, callCount: number) => string);
+}
+
+type EvaluatorFn = (args: {
+    outputs: Record<string, any>;
+    referenceOutputs: Record<string, any>;
+}) => Promise<any>;
+interface ResolvedExpectation {
+    evaluator: EvaluatorFn;
+    referenceOutputs: Record<string, unknown>;
+}
+/** A factory that receives test context and returns an evaluator + its referenceOutputs. */
+type Expectation = (ctx: {
+    message: string;
+}) => ResolvedExpectation;
+/**
+ * Expect the agent to call tools in order (superset trajectory match).
+ * Empty `[]` means the agent should answer directly without calling any tools.
+ */
+declare function toolsCalled(tools: string[]): Expectation;
+/**
+ * Run an LLM-as-judge evaluator on the trajectory.
+ * Requires `toolsCalled` in the same expect array.
+ * Uses the globally configured evaluator model.
+ */
+declare function llmJudge(): Expectation;
+/** Assert the agent made zero tool calls. */
+declare function noTools(): Expectation;
+/**
+ * Assert the response is in the given language (ISO 639-1 code).
+ * Uses the globally configured evaluator model for language detection.
+ * @param code - ISO 639-1 language code (e.g. 'en', 'tr', 'de').
+ */
+declare function respondsInLanguage(code: string): Expectation;
+/** Assert the response contains all given strings. */
+declare function contains(strings: string[]): Expectation;
+/** Assert the response does not contain any of the given strings. */
+declare function notContains(strings: string[]): Expectation;
+
+declare function human(content: string): HumanMessage;
+declare function ai(content: string, toolCalls?: string[]): AiMessage;
+declare function toolResult(name: string, output: string): ToolMessage;
+interface ToolDef {
+    description: string;
+    /** A plain key→description record, or a ZodObject passed through from a ToolSpec. */
+    schema?: Record<string, string> | zod.ZodObject<any>;
+    /** Auto-stringified if not a string or function. */
+    response: unknown | ((input: Record<string, unknown>, callCount: number) => string);
+}
+interface TestCase {
+    /** Test name. Defaults to the last human message content if omitted. */
+    name?: string;
+    messages: Message[];
+    systemPrompt?: string;
+    /** Override suite-level tools for this case. */
+    tools?: Record<string, ToolDef>;
+    expect: Expectation[];
+}
+type TargetFn = (inputs: {
+    systemPrompt?: string;
+    messages: Message[];
+    tools: MockToolDef[];
+}) => Promise<{
+    messages: BaseMessage[];
+}>;
+interface SuiteConfig {
+    /** Custom target function, or model string override. Auto-created from global config if omitted. */
+    target?: TargetFn | string;
+    /** Factory that creates a fresh Agent per test case. Overrides global createTarget. */
+    createTarget?: CreateTargetFn;
+    /** System prompt for all cases in this suite. Overrides the global prompt; can be overridden per-case. */
+    systemPrompt?: string;
+    tools?: Record<string, ToolDef>;
+    cases: TestCase[];
+}
+/**
+ * Converts a `ToolSpec[]` (from a real tool provider) into the
+ * `Record<string, ToolDef>` that `defineSuite` expects.
+ *
+ * `responses` maps tool names to canned mock responses. Tools without an
+ * entry in `responses` default to `''`.
+ */
+declare function fromToolSpecs(specs: ToolSpec[], responses?: Record<string, ToolDef['response']>): Record<string, ToolDef>;
+declare function defineSuite(name: string, config: SuiteConfig): void;
+
+export { type CreateTargetFn, type EvalConfig, type Expectation, type SuiteConfig, type TestCase, type ToolDef, ai, configureEvals, contains, defineSuite, fromToolSpecs, human, llmJudge, noTools, notContains, respondsInLanguage, toolResult, toolsCalled };