@almadar/llm 1.0.16 → 2.0.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@almadar/llm",
-  "version": "1.0.16",
+  "version": "2.0.0",
   "description": "Multi-provider LLM client with rate limiting, token tracking, structured outputs, and continuation handling",
   "type": "module",
   "main": "./dist/index.js",
@@ -35,7 +35,7 @@
     "zod": "^3.22.0"
   },
   "peerDependencies": {
-    "@almadar/core": "1.0.17"
+    "@almadar/core": ">=2.0.0"
   },
   "peerDependenciesMeta": {
     "@almadar/core": {
@@ -46,6 +46,15 @@
     "tsup": "^8.0.0",
     "typescript": "^5.3.0"
   },
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/almadar-io/almadar.git",
+    "directory": "docs/packages/llm"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "homepage": "https://github.com/almadar-io/almadar#readme",
   "scripts": {
     "build": "tsup",
     "dev": "tsup --watch",

package/src/client.ts

@@ -12,10 +12,8 @@
  */
 
 import { ChatOpenAI } from '@langchain/openai';
-import {
-  ChatAnthropic,
-  type ChatAnthropicCallOptions,
-} from '@langchain/anthropic';
+import { ChatAnthropic } from '@langchain/anthropic';
+import type { BaseMessageLike } from '@langchain/core/messages';
 import Anthropic from '@anthropic-ai/sdk';
 import { z } from 'zod';
 import {
@@ -26,81 +24,38 @@ import {
 import { TokenTracker, getGlobalTokenTracker } from './token-tracker.js';
 import { parseJsonResponse } from './json-parser.js';
 
-// eslint-disable-next-line @typescript-eslint/no-explicit-any
-type MessageLike = any;
-
 // ============================================================================
-// Caching Chat Anthropic Wrapper
+// Anthropic Cache Control Helper
 // ============================================================================
 
-class CachingChatAnthropic extends ChatAnthropic {
-  async invoke(
-    input: MessageLike[] | string,
-    options?: Partial<ChatAnthropicCallOptions>,
-  ): Promise<MessageLike> {
-    let messages: MessageLike[];
-    if (typeof input === 'string') {
-      messages = [{ role: 'user', content: input }];
-    } else {
-      messages = input;
+function addCacheControlToSystemMessages(
+  messages: Array<{ role: string; content: string }>,
+): BaseMessageLike[] {
+  return messages.map((msg) => {
+    if (msg.role !== 'system') {
+      return msg as BaseMessageLike;
     }
 
-    const transformedMessages = messages.map((msg: MessageLike) => {
-      const msgType = msg._getType?.() || msg.role || 'unknown';
-      const isSystem = msgType === 'system';
-
-      if (!isSystem) return msg;
-
-      if (typeof msg.content === 'string') {
-        return {
-          ...msg,
-          content: [
-            {
-              type: 'text',
-              text: msg.content,
-              cache_control: { type: 'ephemeral' },
-            },
-          ],
-        };
-      }
-
-      if (Array.isArray(msg.content)) {
-        const blocks = msg.content as Array<{
-          type?: string;
-          text?: string;
-          cache_control?: unknown;
-        }>;
-        const hasAnyCacheControl = blocks.some((b) => b.cache_control);
-
-        if (!hasAnyCacheControl) {
-          const transformedBlocks = blocks.map((block, idx) => {
-            if (block.type === 'text' && idx === blocks.length - 1) {
-              return {
-                ...block,
-                cache_control: { type: 'ephemeral' },
-              };
-            }
-            return block;
-          });
-
-          return { ...msg, content: transformedBlocks };
-        }
-      }
-
-      return msg;
-    });
-
-    return super.invoke(transformedMessages, options);
-  }
+    return {
+      role: msg.role,
+      content: [
+        {
+          type: 'text' as const,
+          text: msg.content,
+          cache_control: { type: 'ephemeral' },
+        },
+      ],
+    } as BaseMessageLike;
+  });
 }
 
-type ChatModel = ChatOpenAI | CachingChatAnthropic;
+type ChatModel = ChatOpenAI | ChatAnthropic;
 
 // ============================================================================
 // Types
 // ============================================================================
 
-export type LLMProvider = 'openai' | 'deepseek' | 'anthropic' | 'kimi';
+export type LLMProvider = 'openai' | 'deepseek' | 'anthropic' | 'kimi' | 'openrouter';
 
 export interface ProviderConfig {
   apiKey: string;
@@ -215,10 +170,24 @@ const PROVIDER_CONFIGS: Record<LLMProvider, () => ProviderConfig> = {
     }
     return {
       apiKey,
-      baseUrl: 'https://api.moonshot.cn/v1',
+      baseUrl: 'https://api.moonshot.ai/v1',
       defaultModel: 'kimi-k2.5',
     };
   },
+  openrouter: () => {
+    const apiKey = process.env.OPEN_ROUTER_API_KEY;
+    if (!apiKey) {
+      throw new Error(
+        'OPEN_ROUTER_API_KEY environment variable is not set. ' +
+          'Please set it in your .env file or environment.',
+      );
+    }
+    return {
+      apiKey,
+      baseUrl: 'https://openrouter.ai/api/v1',
+      defaultModel: 'qwen/qwen-2.5-72b-instruct', // Default to Qwen 2.5
+    };
+  },
 };
 
 export const DEEPSEEK_MODELS = {
@@ -246,6 +215,19 @@ export const KIMI_MODELS = {
   K2_5: 'kimi-k2.5',
 } as const;
 
+export const OPENROUTER_MODELS = {
+  // Qwen models - JSON/structured data specialists
+  QWEN_2_5_72B: 'qwen/qwen-2.5-72b-instruct',
+  QWEN_2_5_CODER_32B: 'qwen/qwen-2.5-coder-32b-instruct',
+  QWEN_3_235B: 'qwen/qwen3-235b-a22b',
+  
+  // Llama models - agentic workhorses
+  LLAMA_3_3_70B: 'meta-llama/llama-3.3-70b-instruct',
+  LLAMA_3_1_405B: 'meta-llama/llama-3.1-405b-instruct',
+  LLAMA_4_MAVERICK: 'meta-llama/llama-4-maverick',
+  LLAMA_4_SCOUT: 'meta-llama/llama-4-scout',
+} as const;
+
 const DEFAULT_TEMPERATURE = 0.3;
 
 // ============================================================================
@@ -264,7 +246,9 @@ export class LLMClient {
 
   constructor(options: LLMClientOptions = {}) {
     this.provider = options.provider || 'openai';
-    this.temperature = options.temperature ?? DEFAULT_TEMPERATURE;
+    // Kimi: 0.6 when thinking disabled (our default), 1.0 when thinking enabled
+    this.temperature = options.temperature ?? 
+      (this.provider === 'kimi' ? 0.6 : DEFAULT_TEMPERATURE);
     this.streaming = options.streaming ?? false;
 
     this.providerConfig = PROVIDER_CONFIGS[this.provider]();
@@ -311,7 +295,7 @@ export class LLMClient {
     const temperature = options?.temperature ?? this.temperature;
 
     if (this.provider === 'anthropic') {
-      return new CachingChatAnthropic({
+      return new ChatAnthropic({
         apiKey: this.providerConfig.apiKey,
         model: this.modelName,
         temperature,
@@ -381,13 +365,32 @@ export class LLMClient {
 
     const timeout = this.provider === 'deepseek' ? 600000 : undefined;
 
+    // Kimi-k2.5: disable thinking to avoid reasoning_content issues with tool calls
+    // When thinking is disabled, temperature must be 0.6 (not 1.0)
+    const isKimi = this.provider === 'kimi';
+    const effectiveTemp = isKimi ? 0.6 : temperature;
+
+    // Build modelKwargs incrementally to avoid spread conflicts
+    const modelKwargs: Record<string, unknown> = {};
+    if (useCompletionTokens && maxTokens) {
+      modelKwargs.max_completion_tokens = maxTokens;
+    }
+    if (isKimi) {
+      modelKwargs.thinking = { type: 'disabled' };
+    }
+    // OpenRouter (Qwen): explicit tool_choice so the model doesn't ignore tool definitions
+    if (this.provider === 'openrouter') {
+      modelKwargs.tool_choice = 'auto';
+    }
+
     return new ChatOpenAI({
       apiKey: this.providerConfig.apiKey,
       model: this.modelName,
-      temperature: useCompletionTokens ? undefined : temperature,
+      temperature: useCompletionTokens ? undefined : effectiveTemp,
       streaming: this.streaming,
       timeout,
-      ...tokenConfig,
+      ...(Object.keys(modelKwargs).length > 0 ? { modelKwargs } : {}),
+      ...(useCompletionTokens ? {} : maxTokens ? { maxTokens } : {}),
       configuration: {
         apiKey: this.providerConfig.apiKey,
         ...(this.providerConfig.baseUrl
@@ -468,10 +471,15 @@ export class LLMClient {
               ? this.getModelWithOptions({ maxTokens, temperature })
               : this.model;
 
-          const response = await modelToUse.invoke([
+          const messages = [
             { role: 'system', content: systemPrompt },
             { role: 'user', content: currentPrompt },
-          ]);
+          ];
+          const response = await modelToUse.invoke(
+            this.provider === 'anthropic'
+              ? addCacheControlToSystemMessages(messages)
+              : messages,
+          );
 
           console.log(
             `[LLMClient:call] Model responded in ${Date.now() - invokeStartTime}ms`,
@@ -605,10 +613,15 @@ export class LLMClient {
         ? this.getModelWithOptions({ maxTokens })
         : this.model;
 
-      const response = await modelToUse.invoke([
+      const messages = [
         { role: 'system', content: systemPrompt },
         { role: 'user', content: userPrompt },
-      ]);
+      ];
+      const response = await modelToUse.invoke(
+        this.provider === 'anthropic'
+          ? addCacheControlToSystemMessages(messages)
+          : messages,
+      );
 
       let usage: LLMUsage | null = null;
       if (response.usage_metadata) {
@@ -965,3 +978,13 @@ export function createKimiClient(
     ...options,
   });
 }
+
+export function createOpenRouterClient(
+  options?: Partial<Omit<LLMClientOptions, 'provider'>>,
+): LLMClient {
+  return new LLMClient({
+    provider: 'openrouter',
+    model: OPENROUTER_MODELS.QWEN_2_5_72B,
+    ...options,
+  });
+}

package/src/index.ts

@@ -18,12 +18,14 @@ export {
   createOpenAIClient,
   createAnthropicClient,
   createKimiClient,
+  createOpenRouterClient,
   getAvailableProvider,
   isProviderAvailable,
   DEEPSEEK_MODELS,
   OPENAI_MODELS,
   ANTHROPIC_MODELS,
   KIMI_MODELS,
+  OPENROUTER_MODELS,
   type LLMProvider,
   type ProviderConfig,
   type LLMClientOptions,

package/dist/chunk-56H37PN5.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/client.ts"],"sourcesContent":["/**\n * Shared LLM Client\n *\n * Multi-provider LLM client with:\n * - OpenAI, DeepSeek, Anthropic, and Kimi support\n * - Anthropic prompt caching (CachingChatAnthropic)\n * - Rate limiting and retry logic\n * - Token tracking\n * - Structured output parsing with Zod\n *\n * @packageDocumentation\n */\n\nimport { ChatOpenAI } from '@langchain/openai';\nimport {\n  ChatAnthropic,\n  type ChatAnthropicCallOptions,\n} from '@langchain/anthropic';\nimport Anthropic from '@anthropic-ai/sdk';\nimport { z } from 'zod';\nimport {\n  RateLimiter,\n  getGlobalRateLimiter,\n  type RateLimiterOptions,\n} from './rate-limiter.js';\nimport { TokenTracker, getGlobalTokenTracker } from './token-tracker.js';\nimport { parseJsonResponse } from './json-parser.js';\n\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\ntype MessageLike = any;\n\n// ============================================================================\n// Caching Chat Anthropic Wrapper\n// ============================================================================\n\nclass CachingChatAnthropic extends ChatAnthropic {\n  async invoke(\n    input: MessageLike[] | string,\n    options?: Partial<ChatAnthropicCallOptions>,\n  ): Promise<MessageLike> {\n    let messages: MessageLike[];\n    if (typeof input === 'string') {\n      messages = [{ role: 'user', content: input }];\n    } else {\n      messages = input;\n    }\n\n    const transformedMessages = messages.map((msg: MessageLike) => {\n      const msgType = msg._getType?.() || msg.role || 'unknown';\n      const isSystem = msgType === 'system';\n\n      if (!isSystem) return msg;\n\n      if (typeof msg.content === 'string') {\n        return {\n          ...msg,\n          content: [\n            {\n              type: 'text',\n              text: msg.content,\n              cache_control: { type: 'ephemeral' },\n            },\n          ],\n        };\n      }\n\n      if (Array.isArray(msg.content)) {\n        const blocks = msg.content as Array<{\n          type?: string;\n          text?: string;\n          cache_control?: unknown;\n        }>;\n        const hasAnyCacheControl = blocks.some((b) => b.cache_control);\n\n        if (!hasAnyCacheControl) {\n          const transformedBlocks = blocks.map((block, idx) => {\n            if (block.type === 'text' && idx === blocks.length - 1) {\n              return {\n                ...block,\n                cache_control: { type: 'ephemeral' },\n              };\n            }\n            return block;\n          });\n\n          return { ...msg, content: transformedBlocks };\n        }\n      }\n\n      return msg;\n    });\n\n    return super.invoke(transformedMessages, options);\n  }\n}\n\ntype ChatModel = ChatOpenAI | CachingChatAnthropic;\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport type LLMProvider = 'openai' | 'deepseek' | 'anthropic' | 'kimi';\n\nexport interface ProviderConfig {\n  apiKey: string;\n  baseUrl?: string;\n  defaultModel: string;\n}\n\nexport interface LLMClientOptions {\n  provider?: LLMProvider;\n  model?: string;\n  temperature?: number;\n  streaming?: boolean;\n  rateLimiter?: RateLimiterOptions;\n  useGlobalRateLimiter?: boolean;\n  trackTokens?: boolean;\n}\n\nexport interface LLMCallOptions<T = unknown> {\n  systemPrompt: string;\n  userPrompt: string;\n  schema?: z.ZodSchema<T>;\n  maxRetries?: number;\n  retryWithContext?: boolean;\n  maxTokens?: number;\n  skipSchemaValidation?: boolean;\n  temperature?: number;\n}\n\nexport interface CacheableBlock {\n  type: 'text';\n  text: string;\n  cache_control?: { type: 'ephemeral' };\n}\n\nexport interface CacheAwareLLMCallOptions<T = unknown>\n  extends LLMCallOptions<T> {\n  systemBlocks?: CacheableBlock[];\n  userBlocks?: CacheableBlock[];\n  rawText?: boolean;\n}\n\nexport interface LLMUsage {\n  promptTokens: number;\n  completionTokens: number;\n  totalTokens: number;\n}\n\nexport type LLMFinishReason =\n  | 'stop'\n  | 'length'\n  | 'content_filter'\n  | 'tool_calls'\n  | null;\n\nexport interface LLMResponse<T> {\n  data: T;\n  raw: string;\n  finishReason: LLMFinishReason;\n  usage: LLMUsage | null;\n}\n\n// ============================================================================\n// Provider Configuration\n// ============================================================================\n\nconst PROVIDER_CONFIGS: Record<LLMProvider, () => ProviderConfig> = {\n  openai: () => {\n    const apiKey = process.env.OPENAI_API_KEY;\n    if (!apiKey) {\n      throw new Error(\n        'OPENAI_API_KEY environment variable is not set. ' +\n          'Please set it in your .env file or environment.',\n      );\n    }\n    return { apiKey, baseUrl: undefined, defaultModel: 'gpt-4o' };\n  },\n  deepseek: () => {\n    const apiKey = process.env.DEEPSEEK_API_KEY;\n    if (!apiKey) {\n      throw new Error(\n        'DEEPSEEK_API_KEY environment variable is not set. ' +\n          'Please set it in your .env file or environment.',\n      );\n    }\n    return {\n      apiKey,\n      baseUrl: 'https://api.deepseek.com/v1',\n      defaultModel: 'deepseek-chat',\n    };\n  },\n  anthropic: () => {\n    const apiKey = process.env.ANTHROPIC_API_KEY;\n    if (!apiKey) {\n      throw new Error(\n        'ANTHROPIC_API_KEY environment variable is not set. ' +\n          'Please set it in your .env file or environment.',\n      );\n    }\n    return {\n      apiKey,\n      baseUrl: undefined,\n      defaultModel: 'claude-sonnet-4-5-20250929',\n    };\n  },\n  kimi: () => {\n    const apiKey = process.env.KIMI_API_KEY;\n    if (!apiKey) {\n      throw new Error(\n        'KIMI_API_KEY environment variable is not set. ' +\n          'Please set it in your .env file or environment.',\n      );\n    }\n    return {\n      apiKey,\n      baseUrl: 'https://api.moonshot.cn/v1',\n      defaultModel: 'kimi-k2.5',\n    };\n  },\n};\n\nexport const DEEPSEEK_MODELS = {\n  CHAT: 'deepseek-chat',\n  CODER: 'deepseek-coder',\n  REASONER: 'deepseek-reasoner',\n} as const;\n\nexport const OPENAI_MODELS = {\n  GPT4O: 'gpt-4o',\n  GPT4O_MINI: 'gpt-4o-mini',\n  GPT4_TURBO: 'gpt-4-turbo',\n  GPT35_TURBO: 'gpt-3.5-turbo',\n  GPT_5_1: 'gpt-5.1',\n} as const;\n\nexport const ANTHROPIC_MODELS = {\n  CLAUDE_SONNET_4_5: 'claude-sonnet-4-5-20250929',\n  CLAUDE_SONNET_4: 'claude-sonnet-4-20250514',\n  CLAUDE_OPUS_4_5: 'claude-opus-4-5-20250929',\n  CLAUDE_3_5_HAIKU: 'claude-3-5-haiku-20241022',\n} as const;\n\nexport const KIMI_MODELS = {\n  K2_5: 'kimi-k2.5',\n} as const;\n\nconst DEFAULT_TEMPERATURE = 0.3;\n\n// ============================================================================\n// LLM Client\n// ============================================================================\n\nexport class LLMClient {\n  private model: ChatModel;\n  private rateLimiter: RateLimiter;\n  private tokenTracker: TokenTracker | null;\n  private modelName: string;\n  private provider: LLMProvider;\n  private providerConfig: ProviderConfig;\n  private temperature: number;\n  private streaming: boolean;\n\n  constructor(options: LLMClientOptions = {}) {\n    this.provider = options.provider || 'openai';\n    this.temperature = options.temperature ?? DEFAULT_TEMPERATURE;\n    this.streaming = options.streaming ?? false;\n\n    this.providerConfig = PROVIDER_CONFIGS[this.provider]();\n    this.modelName = options.model || this.providerConfig.defaultModel;\n\n    const keyPreview = this.providerConfig.apiKey.slice(-4);\n    console.log(\n      `[LLMClient] Provider: ${this.provider}, Model: ${this.modelName}, Key: ****${keyPreview}`,\n    );\n    if (this.providerConfig.baseUrl) {\n      console.log(\n        `[LLMClient] Using custom base URL: ${this.providerConfig.baseUrl}`,\n      );\n    }\n\n    this.model = this.createModel();\n\n    this.rateLimiter =\n      options.useGlobalRateLimiter !== false\n        ? getGlobalRateLimiter(options.rateLimiter)\n        : new RateLimiter(options.rateLimiter);\n\n    this.tokenTracker =\n      options.trackTokens !== false\n        ? getGlobalTokenTracker(this.modelName)\n        : null;\n  }\n\n  private usesMaxCompletionTokens(): boolean {\n    const model = this.modelName.toLowerCase();\n    return (\n      model.startsWith('o1') ||\n      model.startsWith('gpt-5') ||\n      model.includes('o1-') ||\n      model.includes('o3')\n    );\n  }\n\n  private createModel(options?: {\n    maxTokens?: number;\n    temperature?: number;\n  }): ChatModel {\n    const maxTokens = options?.maxTokens;\n    const temperature = options?.temperature ?? this.temperature;\n\n    if (this.provider === 'anthropic') {\n      return new CachingChatAnthropic({\n        apiKey: this.providerConfig.apiKey,\n        model: this.modelName,\n        temperature,\n        streaming: this.streaming,\n        maxTokens: maxTokens || 8192,\n        callbacks: [\n          {\n            handleLLMEnd: (output) => {\n              const generation = output.generations?.[0]?.[0];\n              const usage = (\n                generation as unknown as {\n                  message?: {\n                    usage_metadata?: {\n                      cache_creation_input_tokens?: number;\n                      cache_read_input_tokens?: number;\n                      input_tokens?: number;\n                      output_tokens?: number;\n                    };\n                  };\n                }\n              )?.message?.usage_metadata;\n\n              if (usage) {\n                const cacheCreated = usage.cache_creation_input_tokens ?? 0;\n                const cacheRead = usage.cache_read_input_tokens ?? 0;\n                const inputTokens = usage.input_tokens ?? 0;\n                const outputTokens = usage.output_tokens ?? 0;\n\n                if (cacheCreated > 0) {\n                  console.log(\n                    `[LLMClient:Anthropic] Cache WRITE: ${cacheCreated} tokens cached`,\n                  );\n                }\n                if (cacheRead > 0) {\n                  const savingsPercent = Math.round(\n                    (cacheRead / (cacheRead + inputTokens)) * 100,\n                  );\n                  console.log(\n                    `[LLMClient:Anthropic] Cache HIT: ${cacheRead} tokens (~${savingsPercent}% of prompt)`,\n                  );\n                }\n                if (cacheCreated === 0 && cacheRead === 0 && inputTokens > 0) {\n                  if (inputTokens < 500) {\n                    console.log(\n                      `[LLMClient:Anthropic] ${inputTokens} input, ${outputTokens} output tokens (likely cached)`,\n                    );\n                  } else {\n                    console.log(\n                      `[LLMClient:Anthropic] ${inputTokens} input, ${outputTokens} output tokens`,\n                    );\n                  }\n                }\n              }\n            },\n          },\n        ],\n      });\n    }\n\n    const useCompletionTokens = this.usesMaxCompletionTokens();\n\n    const tokenConfig = maxTokens\n      ? useCompletionTokens\n        ? { modelKwargs: { max_completion_tokens: maxTokens } }\n        : { maxTokens }\n      : {};\n\n    const timeout = this.provider === 'deepseek' ? 600000 : undefined;\n\n    return new ChatOpenAI({\n      apiKey: this.providerConfig.apiKey,\n      model: this.modelName,\n      temperature: useCompletionTokens ? undefined : temperature,\n      streaming: this.streaming,\n      timeout,\n      ...tokenConfig,\n      configuration: {\n        apiKey: this.providerConfig.apiKey,\n        ...(this.providerConfig.baseUrl\n          ? { baseURL: this.providerConfig.baseUrl }\n          : {}),\n      },\n    });\n  }\n\n  private getModelWithOptions(options: {\n    maxTokens?: number;\n    temperature?: number;\n  }): ChatModel {\n    return this.createModel(options);\n  }\n\n  getProvider(): LLMProvider {\n    return this.provider;\n  }\n\n  getModelName(): string {\n    return this.modelName;\n  }\n\n  getModel(): ChatModel {\n    return this.model;\n  }\n\n  getRateLimiterStatus() {\n    return this.rateLimiter.getStatus();\n  }\n\n  getTokenUsage() {\n    return this.tokenTracker?.getSummary() ?? null;\n  }\n\n  async call<T>(options: LLMCallOptions<T>): Promise<T> {\n    const response = await this.callWithMetadata(options);\n    return response.data;\n  }\n\n  async callWithMetadata<T>(options: LLMCallOptions<T>): Promise<LLMResponse<T>> {\n    const {\n      systemPrompt,\n      userPrompt,\n      schema,\n      maxRetries = 2,\n      retryWithContext = true,\n      maxTokens,\n      skipSchemaValidation = false,\n      temperature,\n    } = options;\n\n    let currentPrompt = userPrompt;\n    let lastError: Error | null = null;\n\n    console.log(\n      `[LLMClient:call] Starting call to ${this.provider}/${this.modelName}`,\n    );\n    console.log(`[LLMClient:call] Prompt length: ${userPrompt.length} chars`);\n    if (maxTokens) {\n      console.log(`[LLMClient:call] Max tokens: ${maxTokens}`);\n    }\n\n    for (let attempt = 0; attempt <= maxRetries; attempt++) {\n      try {\n        console.log(\n          `[LLMClient:call] Attempt ${attempt + 1}/${maxRetries + 1}...`,\n        );\n        const attemptStartTime = Date.now();\n\n        const result = await this.rateLimiter.execute(async () => {\n          console.log(`[LLMClient:call] Invoking model...`);\n          const invokeStartTime = Date.now();\n\n          const modelToUse =\n            maxTokens || temperature !== undefined\n              ? this.getModelWithOptions({ maxTokens, temperature })\n              : this.model;\n\n          const response = await modelToUse.invoke([\n            { role: 'system', content: systemPrompt },\n            { role: 'user', content: currentPrompt },\n          ]);\n\n          console.log(\n            `[LLMClient:call] Model responded in ${Date.now() - invokeStartTime}ms`,\n          );\n\n          let usage: LLMUsage | null = null;\n          if (response.usage_metadata) {\n            const usageMeta = response.usage_metadata as {\n              input_tokens?: number;\n              output_tokens?: number;\n            };\n            usage = {\n              promptTokens: usageMeta.input_tokens || 0,\n              completionTokens: usageMeta.output_tokens || 0,\n              totalTokens:\n                (usageMeta.input_tokens || 0) +\n                (usageMeta.output_tokens || 0),\n            };\n            console.log(\n              `[LLMClient:call] Tokens used: ${usage.promptTokens} in, ${usage.completionTokens} out`,\n            );\n\n            if (this.tokenTracker) {\n              this.tokenTracker.addUsage(\n                usage.promptTokens,\n                usage.completionTokens,\n              );\n            }\n          }\n\n          const finishReason = this.extractFinishReason(response);\n          if (finishReason === 'length') {\n            console.warn(\n              `[LLMClient:call] Response truncated (finish_reason=length)`,\n            );\n          }\n\n          const content =\n            typeof response.content === 'string'\n              ? response.content\n              : JSON.stringify(response.content);\n\n          console.log(\n            `[LLMClient:call] Response length: ${content.length} chars, finish_reason: ${finishReason}`,\n          );\n\n          return { content, finishReason, usage };\n        });\n\n        console.log(\n          `[LLMClient:call] Attempt ${attempt + 1} completed in ${Date.now() - attemptStartTime}ms, parsing response...`,\n        );\n\n        const parsed = skipSchemaValidation\n          ? (parseJsonResponse(result.content, undefined) as T)\n          : parseJsonResponse(result.content, schema);\n        console.log(\n          `[LLMClient:call] Response parsed successfully${skipSchemaValidation ? ' (schema validation skipped)' : ''}`,\n        );\n\n        return {\n          data: parsed,\n          raw: result.content,\n          finishReason: result.finishReason,\n          usage: result.usage,\n        };\n      } catch (error) {\n        lastError = error instanceof Error ? error : new Error(String(error));\n        console.error(\n          `[LLMClient:call] Attempt ${attempt + 1} failed:`,\n          lastError.message,\n        );\n\n        if (this.isRateLimitError(lastError)) {\n          console.error(`[LLMClient:call] Rate limit error, not retrying`);\n          throw lastError;\n        }\n\n        if (attempt < maxRetries && retryWithContext) {\n          console.log(`[LLMClient:call] Will retry with error context`);\n          currentPrompt =\n            `${userPrompt}\\n\\n` +\n            `[Previous attempt failed with: ${lastError.message}]\\n` +\n            `Please output valid JSON that matches the expected schema.`;\n        }\n      }\n    }\n\n    console.error(`[LLMClient:call] All attempts exhausted, throwing error`);\n    throw lastError;\n  }\n\n  private extractFinishReason(\n    response: Awaited<ReturnType<ChatOpenAI['invoke']>>,\n  ): LLMFinishReason {\n    const metadata = response.response_metadata as\n      | Record<string, unknown>\n      | undefined;\n    if (metadata?.finish_reason) {\n      const reason = metadata.finish_reason as string;\n      if (\n        reason === 'stop' ||\n        reason === 'length' ||\n        reason === 'content_filter' ||\n        reason === 'tool_calls'\n      ) {\n        return reason;\n      }\n    }\n    return null;\n  }\n\n  async callRaw(options: {\n    systemPrompt: string;\n    userPrompt: string;\n    maxTokens?: number;\n  }): Promise<string> {\n    const response = await this.callRawWithMetadata(options);\n    return response.raw;\n  }\n\n  async callRawWithMetadata(options: {\n    systemPrompt: string;\n    userPrompt: string;\n    maxTokens?: number;\n  }): Promise<Omit<LLMResponse<string>, 'data'> & { raw: string }> {\n    const { systemPrompt, userPrompt, maxTokens } = options;\n\n    return this.rateLimiter.execute(async () => {\n      const modelToUse = maxTokens\n        ? this.getModelWithOptions({ maxTokens })\n        : this.model;\n\n      const response = await modelToUse.invoke([\n        { role: 'system', content: systemPrompt },\n        { role: 'user', content: userPrompt },\n      ]);\n\n      let usage: LLMUsage | null = null;\n      if (response.usage_metadata) {\n        const usageMeta = response.usage_metadata as {\n          input_tokens?: number;\n          output_tokens?: number;\n        };\n        usage = {\n          promptTokens: usageMeta.input_tokens || 0,\n          completionTokens: usageMeta.output_tokens || 0,\n          totalTokens:\n            (usageMeta.input_tokens || 0) + (usageMeta.output_tokens || 0),\n        };\n\n        if (this.tokenTracker) {\n          this.tokenTracker.addUsage(\n            usage.promptTokens,\n            usage.completionTokens,\n          );\n        }\n      }\n\n      const finishReason = this.extractFinishReason(response);\n      const content =\n        typeof response.content === 'string'\n          ? response.content\n          : JSON.stringify(response.content);\n\n      return { raw: content, finishReason, usage };\n    });\n  }\n\n  private isRateLimitError(error: Error): boolean {\n    const message = error.message.toLowerCase();\n    return (\n      message.includes('rate limit') ||\n      message.includes('429') ||\n      message.includes('quota exceeded')\n    );\n  }\n\n  // ==========================================================================\n  // Anthropic Cache Control Support\n  // ==========================================================================\n\n  async callWithCache<T>(\n    options: CacheAwareLLMCallOptions<T>,\n  ): Promise<LLMResponse<T>> {\n    const {\n      systemPrompt,\n      userPrompt,\n      systemBlocks,\n      userBlocks,\n      schema,\n      maxRetries = 2,\n      maxTokens,\n      skipSchemaValidation = false,\n      temperature,\n      rawText = false,\n    } = options;\n\n    if (this.provider !== 'anthropic') {\n      console.log(\n        `[LLMClient:callWithCache] Provider ${this.provider} doesn't support caching, using regular call`,\n      );\n      return this.callWithMetadata(options);\n    }\n\n    const cacheableCount =\n      (systemBlocks || []).filter((b) => b.cache_control).length +\n      (userBlocks || []).filter((b) => b.cache_control).length;\n    console.log(\n      `[LLMClient:callWithCache] ${cacheableCount} cacheable block(s)`,\n    );\n\n    let lastError: Error | null = null;\n\n    for (let attempt = 0; attempt <= maxRetries; attempt++) {\n      try {\n        console.log(\n          `[LLMClient:callWithCache] Attempt ${attempt + 1}/${maxRetries + 1}...`,\n        );\n\n        const result = await this.rateLimiter.execute(async () => {\n          const anthropic = new Anthropic();\n\n          const systemContent =\n            systemBlocks && systemBlocks.length > 0\n              ? systemBlocks.map((b) => ({\n                  type: 'text' as const,\n                  text: b.text,\n                  ...(b.cache_control\n                    ? { cache_control: b.cache_control }\n                    : {}),\n                }))\n              : systemPrompt\n                ? [{ type: 'text' as const, text: systemPrompt }]\n                : [];\n\n          const userContent =\n            userBlocks && userBlocks.length > 0\n              ? userBlocks.map((b) => ({\n                  type: 'text' as const,\n                  text: b.text,\n                  ...(b.cache_control\n                    ? { cache_control: b.cache_control }\n                    : {}),\n                }))\n              : userPrompt\n                ? [{ type: 'text' as const, text: userPrompt }]\n                : [];\n\n          const response = await anthropic.messages.create({\n            model: this.modelName,\n            max_tokens: maxTokens || 8192,\n            temperature: temperature ?? 0,\n            system: systemContent,\n            messages: [{ role: 'user', content: userContent }],\n          });\n\n          const textContent = response.content.find((c) => c.type === 'text');\n          const content =\n            textContent && 'text' in textContent ? textContent.text : '';\n\n          const apiUsage = response.usage as {\n            input_tokens: number;\n            output_tokens: number;\n            cache_creation_input_tokens?: number;\n            cache_read_input_tokens?: number;\n          };\n\n          const cacheRead = apiUsage.cache_read_input_tokens || 0;\n          const cacheCreation = apiUsage.cache_creation_input_tokens || 0;\n\n          if (cacheCreation > 0) {\n            console.log(\n              `[LLMClient:callWithCache] Cache WRITE: ${cacheCreation} tokens`,\n            );\n          }\n          if (cacheRead > 0) {\n            const savingsPercent = Math.round(\n              (cacheRead / (cacheRead + apiUsage.input_tokens)) * 100,\n            );\n            console.log(\n              `[LLMClient:callWithCache] Cache HIT: ${cacheRead} tokens (~${savingsPercent}% of prompt)`,\n            );\n          }\n          if (cacheCreation === 0 && cacheRead === 0) {\n            console.log(\n              `[LLMClient:callWithCache] No caching: ${apiUsage.input_tokens} input tokens`,\n            );\n          }\n\n          const usage: LLMUsage = {\n            promptTokens: apiUsage.input_tokens,\n            completionTokens: apiUsage.output_tokens,\n            totalTokens: apiUsage.input_tokens + apiUsage.output_tokens,\n          };\n\n          if (this.tokenTracker) {\n            this.tokenTracker.addUsage(\n              usage.promptTokens,\n              usage.completionTokens,\n            );\n          }\n\n          const finishReason =\n            response.stop_reason === 'end_turn'\n              ? 'stop'\n              : response.stop_reason;\n\n          return {\n            content,\n            finishReason: finishReason as LLMFinishReason,\n            usage,\n          };\n        });\n\n        let parsed: T;\n        if (rawText) {\n          parsed = result.content as unknown as T;\n        } else if (skipSchemaValidation) {\n          parsed = parseJsonResponse(result.content, undefined) as T;\n        } else {\n          parsed = parseJsonResponse(result.content, schema);\n        }\n\n        return {\n          data: parsed,\n          raw: result.content,\n          finishReason: result.finishReason,\n          usage: result.usage,\n        };\n      } catch (error) {\n        lastError = error instanceof Error ? error : new Error(String(error));\n        console.error(\n          `[LLMClient:callWithCache] Attempt ${attempt + 1} failed:`,\n          lastError.message,\n        );\n\n        if (this.isRateLimitError(lastError)) {\n          throw lastError;\n        }\n      }\n    }\n\n    throw lastError;\n  }\n\n  static cacheableBlock(text: string, cache = true): CacheableBlock {\n    return cache\n      ? { type: 'text', text, cache_control: { type: 'ephemeral' } }\n      : { type: 'text', text };\n  }\n}\n\n// ============================================================================\n// Singleton Instances\n// ============================================================================\n\nconst sharedClients: Partial<Record<LLMProvider, LLMClient>> = {};\n\nexport function getSharedLLMClient(options?: LLMClientOptions): LLMClient {\n  const provider = options?.provider || 'openai';\n  if (!sharedClients[provider]) {\n    sharedClients[provider] = new LLMClient(options);\n  }\n  return sharedClients[provider]!;\n}\n\nexport function resetSharedLLMClient(provider?: LLMProvider): void {\n  if (provider) {\n    delete sharedClients[provider];\n  } else {\n    for (const key of Object.keys(sharedClients) as LLMProvider[]) {\n      delete sharedClients[key];\n    }\n  }\n}\n\n// ============================================================================\n// Provider Detection\n// ============================================================================\n\nexport function getAvailableProvider(): LLMProvider {\n  if (process.env.ANTHROPIC_API_KEY) return 'anthropic';\n  if (process.env.DEEPSEEK_API_KEY) return 'deepseek';\n  if (process.env.KIMI_API_KEY) return 'kimi';\n  if (process.env.OPENAI_API_KEY) return 'openai';\n  throw new Error(\n    'No LLM API key found. Please set ANTHROPIC_API_KEY, OPENAI_API_KEY, DEEPSEEK_API_KEY, or KIMI_API_KEY.',\n  );\n}\n\nexport function isProviderAvailable(provider: LLMProvider): boolean {\n  switch (provider) {\n    case 'openai':\n      return !!process.env.OPENAI_API_KEY;\n    case 'deepseek':\n      return !!process.env.DEEPSEEK_API_KEY;\n    case 'anthropic':\n      return !!process.env.ANTHROPIC_API_KEY;\n    case 'kimi':\n      return !!process.env.KIMI_API_KEY;\n    default:\n      return false;\n  }\n}\n\n// ============================================================================\n// Convenience Functions\n// ============================================================================\n\nexport function createRequirementsClient(\n  options?: Partial<LLMClientOptions>,\n): LLMClient {\n  const provider = options?.provider || getAvailableProvider();\n  const defaultModel =\n    provider === 'deepseek' ? DEEPSEEK_MODELS.CHAT : OPENAI_MODELS.GPT_5_1;\n  return new LLMClient({\n    provider,\n    model: defaultModel,\n    temperature: 0.3,\n    ...options,\n  });\n}\n\nexport function createCreativeClient(\n  options?: Partial<LLMClientOptions>,\n): LLMClient {\n  const provider = options?.provider || getAvailableProvider();\n  const defaultModel =\n    provider === 'deepseek' ? DEEPSEEK_MODELS.REASONER : OPENAI_MODELS.GPT4O;\n  return new LLMClient({\n    provider,\n    model: defaultModel,\n    temperature: 0.7,\n    ...options,\n  });\n}\n\nexport function createFixClient(\n  options?: Partial<LLMClientOptions>,\n): LLMClient {\n  const provider = options?.provider || getAvailableProvider();\n  const defaultModel =\n    provider === 'deepseek'\n      ? DEEPSEEK_MODELS.CHAT\n      : OPENAI_MODELS.GPT4O_MINI;\n  return new LLMClient({\n    provider,\n    model: defaultModel,\n    temperature: 0.2,\n    ...options,\n  });\n}\n\nexport function createDeepSeekClient(\n  options?: Partial<Omit<LLMClientOptions, 'provider'>>,\n): LLMClient {\n  return new LLMClient({\n    provider: 'deepseek',\n    model: DEEPSEEK_MODELS.CHAT,\n    ...options,\n  });\n}\n\nexport function createOpenAIClient(\n  options?: Partial<Omit<LLMClientOptions, 'provider'>>,\n): LLMClient {\n  return new LLMClient({\n    provider: 'openai',\n    model: OPENAI_MODELS.GPT4O,\n    ...options,\n  });\n}\n\nexport function createAnthropicClient(\n  options?: Partial<Omit<LLMClientOptions, 'provider'>>,\n): LLMClient {\n  return new LLMClient({\n    provider: 'anthropic',\n    model: ANTHROPIC_MODELS.CLAUDE_SONNET_4_5,\n    ...options,\n  });\n}\n\nexport function createKimiClient(\n  options?: Partial<Omit<LLMClientOptions, 'provider'>>,\n): LLMClient {\n  return new LLMClient({\n    provider: 'kimi',\n    model: KIMI_MODELS.K2_5,\n    ...options,\n  });\n}\n"],"mappings":";;;;;;;;;;AAaA,SAAS,kBAAkB;AAC3B;AAAA,EACE;AAAA,OAEK;AACP,OAAO,eAAe;AAiBtB,IAAM,uBAAN,cAAmC,cAAc;AAAA,EAC/C,MAAM,OACJ,OACA,SACsB;AACtB,QAAI;AACJ,QAAI,OAAO,UAAU,UAAU;AAC7B,iBAAW,CAAC,EAAE,MAAM,QAAQ,SAAS,MAAM,CAAC;AAAA,IAC9C,OAAO;AACL,iBAAW;AAAA,IACb;AAEA,UAAM,sBAAsB,SAAS,IAAI,CAAC,QAAqB;AAC7D,YAAM,UAAU,IAAI,WAAW,KAAK,IAAI,QAAQ;AAChD,YAAM,WAAW,YAAY;AAE7B,UAAI,CAAC,SAAU,QAAO;AAEtB,UAAI,OAAO,IAAI,YAAY,UAAU;AACnC,eAAO;AAAA,UACL,GAAG;AAAA,UACH,SAAS;AAAA,YACP;AAAA,cACE,MAAM;AAAA,cACN,MAAM,IAAI;AAAA,cACV,eAAe,EAAE,MAAM,YAAY;AAAA,YACrC;AAAA,UACF;AAAA,QACF;AAAA,MACF;AAEA,UAAI,MAAM,QAAQ,IAAI,OAAO,GAAG;AAC9B,cAAM,SAAS,IAAI;AAKnB,cAAM,qBAAqB,OAAO,KAAK,CAAC,MAAM,EAAE,aAAa;AAE7D,YAAI,CAAC,oBAAoB;AACvB,gBAAM,oBAAoB,OAAO,IAAI,CAAC,OAAO,QAAQ;AACnD,gBAAI,MAAM,SAAS,UAAU,QAAQ,OAAO,SAAS,GAAG;AACtD,qBAAO;AAAA,gBACL,GAAG;AAAA,gBACH,eAAe,EAAE,MAAM,YAAY;AAAA,cACrC;AAAA,YACF;AACA,mBAAO;AAAA,UACT,CAAC;AAED,iBAAO,EAAE,GAAG,KAAK,SAAS,kBAAkB;AAAA,QAC9C;AAAA,MACF;AAEA,aAAO;AAAA,IACT,CAAC;AAED,WAAO,MAAM,OAAO,qBAAqB,OAAO;AAAA,EAClD;AACF;AA0EA,IAAM,mBAA8D;AAAA,EAClE,QAAQ,MAAM;AACZ,UAAM,SAAS,QAAQ,IAAI;AAC3B,QAAI,CAAC,QAAQ;AACX,YAAM,IAAI;AAAA,QACR;AAAA,MAEF;AAAA,IACF;AACA,WAAO,EAAE,QAAQ,SAAS,QAAW,cAAc,SAAS;AAAA,EAC9D;AAAA,EACA,UAAU,MAAM;AACd,UAAM,SAAS,QAAQ,IAAI;AAC3B,QAAI,CAAC,QAAQ;AACX,YAAM,IAAI;AAAA,QACR;AAAA,MAEF;AAAA,IACF;AACA,WAAO;AAAA,MACL;AAAA,MACA,SAAS;AAAA,MACT,cAAc;AAAA,IAChB;AAAA,EACF;AAAA,EACA,WAAW,MAAM;AACf,UAAM,SAAS,QAAQ,IAAI;AAC3B,QAAI,CAAC,QAAQ;AACX,YAAM,IAAI;AAAA,QACR;AAAA,MAEF;AAAA,IACF;AACA,WAAO;AAAA,MACL;AAAA,MACA,SAAS;AAAA,MACT,cAAc;AAAA,IAChB;AAAA,EACF;AAAA,EACA,MAAM,MAAM;AACV,UAAM,SAAS,QAAQ,IAAI;AAC3B,QAAI,CAAC,QAAQ;AACX,YAAM,IAAI;AAAA,QACR;AAAA,MAEF;AAAA,IACF;AACA,WAAO;AAAA,MACL;AAAA,MACA,SAAS;AAAA,MACT,cAAc;AAAA,IAChB;AAAA,EACF;AACF;AAEO,IAAM,kBAAkB;AAAA,EAC7B,MAAM;AAAA,EACN,OAAO;AAAA,EACP,UAAU;AACZ;AAEO,IAAM,gBAAgB;AAAA,EAC3B,OAAO;AAAA,EACP,YAAY;AAAA,EACZ,YAAY;AAAA,EACZ,aAAa;AAAA,EACb,SAAS;AACX;AAEO,IAAM,mBAAmB;AAAA,EAC9B,mBAAmB;AAAA,EACnB,iBAAiB;AAAA,EACjB,iBAAiB;AAAA,EACjB,kBAAkB;AACpB;AAEO,IAAM,cAAc;AAAA,EACzB,MAAM;AACR;AAEA,IAAM,sBAAsB;AAMrB,IAAM,YAAN,MAAgB;AAAA,EAUrB,YAAY,UAA4B,CAAC,GAAG;AAC1C,SAAK,WAAW,QAAQ,YAAY;AACpC,SAAK,cAAc,QAAQ,eAAe;AAC1C,SAAK,YAAY,QAAQ,aAAa;AAEtC,SAAK,iBAAiB,iBAAiB,KAAK,QAAQ,EAAE;AACtD,SAAK,YAAY,QAAQ,SAAS,KAAK,eAAe;AAEtD,UAAM,aAAa,KAAK,eAAe,OAAO,MAAM,EAAE;AACtD,YAAQ;AAAA,MACN,yBAAyB,KAAK,QAAQ,YAAY,KAAK,SAAS,cAAc,UAAU;AAAA,IAC1F;AACA,QAAI,KAAK,eAAe,SAAS;AAC/B,cAAQ;AAAA,QACN,sCAAsC,KAAK,eAAe,OAAO;AAAA,MACnE;AAAA,IACF;AAEA,SAAK,QAAQ,KAAK,YAAY;AAE9B,SAAK,cACH,QAAQ,yBAAyB,QAC7B,qBAAqB,QAAQ,WAAW,IACxC,IAAI,YAAY,QAAQ,WAAW;AAEzC,SAAK,eACH,QAAQ,gBAAgB,QACpB,sBAAsB,KAAK,SAAS,IACpC;AAAA,EACR;AAAA,EAEQ,0BAAmC;AACzC,UAAM,QAAQ,KAAK,UAAU,YAAY;AACzC,WACE,MAAM,WAAW,IAAI,KACrB,MAAM,WAAW,OAAO,KACxB,MAAM,SAAS,KAAK,KACpB,MAAM,SAAS,IAAI;AAAA,EAEvB;AAAA,EAEQ,YAAY,SAGN;AACZ,UAAM,YAAY,SAAS;AAC3B,UAAM,cAAc,SAAS,eAAe,KAAK;AAEjD,QAAI,KAAK,aAAa,aAAa;AACjC,aAAO,IAAI,qBAAqB;AAAA,QAC9B,QAAQ,KAAK,eAAe;AAAA,QAC5B,OAAO,KAAK;AAAA,QACZ;AAAA,QACA,WAAW,KAAK;AAAA,QAChB,WAAW,aAAa;AAAA,QACxB,WAAW;AAAA,UACT;AAAA,YACE,cAAc,CAAC,WAAW;AACxB,oBAAM,aAAa,OAAO,cAAc,CAAC,IAAI,CAAC;AAC9C,oBAAM,QACJ,YAUC,SAAS;AAEZ,kBAAI,OAAO;AACT,sBAAM,eAAe,MAAM,+BAA+B;AAC1D,sBAAM,YAAY,MAAM,2BAA2B;AACnD,sBAAM,cAAc,MAAM,gBAAgB;AAC1C,sBAAM,eAAe,MAAM,iBAAiB;AAE5C,oBAAI,eAAe,GAAG;AACpB,0BAAQ;AAAA,oBACN,sCAAsC,YAAY;AAAA,kBACpD;AAAA,gBACF;AACA,oBAAI,YAAY,GAAG;AACjB,wBAAM,iBAAiB,KAAK;AAAA,oBACzB,aAAa,YAAY,eAAgB;AAAA,kBAC5C;AACA,0BAAQ;AAAA,oBACN,oCAAoC,SAAS,aAAa,cAAc;AAAA,kBAC1E;AAAA,gBACF;AACA,oBAAI,iBAAiB,KAAK,cAAc,KAAK,cAAc,GAAG;AAC5D,sBAAI,cAAc,KAAK;AACrB,4BAAQ;AAAA,sBACN,yBAAyB,WAAW,WAAW,YAAY;AAAA,oBAC7D;AAAA,kBACF,OAAO;AACL,4BAAQ;AAAA,sBACN,yBAAyB,WAAW,WAAW,YAAY;AAAA,oBAC7D;AAAA,kBACF;AAAA,gBACF;AAAA,cACF;AAAA,YACF;AAAA,UACF;AAAA,QACF;AAAA,MACF,CAAC;AAAA,IACH;AAEA,UAAM,sBAAsB,KAAK,wBAAwB;AAEzD,UAAM,cAAc,YAChB,sBACE,EAAE,aAAa,EAAE,uBAAuB,UAAU,EAAE,IACpD,EAAE,UAAU,IACd,CAAC;AAEL,UAAM,UAAU,KAAK,aAAa,aAAa,MAAS;AAExD,WAAO,IAAI,WAAW;AAAA,MACpB,QAAQ,KAAK,eAAe;AAAA,MAC5B,OAAO,KAAK;AAAA,MACZ,aAAa,sBAAsB,SAAY;AAAA,MAC/C,WAAW,KAAK;AAAA,MAChB;AAAA,MACA,GAAG;AAAA,MACH,eAAe;AAAA,QACb,QAAQ,KAAK,eAAe;AAAA,QAC5B,GAAI,KAAK,eAAe,UACpB,EAAE,SAAS,KAAK,eAAe,QAAQ,IACvC,CAAC;AAAA,MACP;AAAA,IACF,CAAC;AAAA,EACH;AAAA,EAEQ,oBAAoB,SAGd;AACZ,WAAO,KAAK,YAAY,OAAO;AAAA,EACjC;AAAA,EAEA,cAA2B;AACzB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,eAAuB;AACrB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,WAAsB;AACpB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,uBAAuB;AACrB,WAAO,KAAK,YAAY,UAAU;AAAA,EACpC;AAAA,EAEA,gBAAgB;AACd,WAAO,KAAK,cAAc,WAAW,KAAK;AAAA,EAC5C;AAAA,EAEA,MAAM,KAAQ,SAAwC;AACpD,UAAM,WAAW,MAAM,KAAK,iBAAiB,OAAO;AACpD,WAAO,SAAS;AAAA,EAClB;AAAA,EAEA,MAAM,iBAAoB,SAAqD;AAC7E,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,MACA;AAAA,MACA,aAAa;AAAA,MACb,mBAAmB;AAAA,MACnB;AAAA,MACA,uBAAuB;AAAA,MACvB;AAAA,IACF,IAAI;AAEJ,QAAI,gBAAgB;AACpB,QAAI,YAA0B;AAE9B,YAAQ;AAAA,MACN,qCAAqC,KAAK,QAAQ,IAAI,KAAK,SAAS;AAAA,IACtE;AACA,YAAQ,IAAI,mCAAmC,WAAW,MAAM,QAAQ;AACxE,QAAI,WAAW;AACb,cAAQ,IAAI,gCAAgC,SAAS,EAAE;AAAA,IACzD;AAEA,aAAS,UAAU,GAAG,WAAW,YAAY,WAAW;AACtD,UAAI;AACF,gBAAQ;AAAA,UACN,4BAA4B,UAAU,CAAC,IAAI,aAAa,CAAC;AAAA,QAC3D;AACA,cAAM,mBAAmB,KAAK,IAAI;AAElC,cAAM,SAAS,MAAM,KAAK,YAAY,QAAQ,YAAY;AACxD,kBAAQ,IAAI,oCAAoC;AAChD,gBAAM,kBAAkB,KAAK,IAAI;AAEjC,gBAAM,aACJ,aAAa,gBAAgB,SACzB,KAAK,oBAAoB,EAAE,WAAW,YAAY,CAAC,IACnD,KAAK;AAEX,gBAAM,WAAW,MAAM,WAAW,OAAO;AAAA,YACvC,EAAE,MAAM,UAAU,SAAS,aAAa;AAAA,YACxC,EAAE,MAAM,QAAQ,SAAS,cAAc;AAAA,UACzC,CAAC;AAED,kBAAQ;AAAA,YACN,uCAAuC,KAAK,IAAI,IAAI,eAAe;AAAA,UACrE;AAEA,cAAI,QAAyB;AAC7B,cAAI,SAAS,gBAAgB;AAC3B,kBAAM,YAAY,SAAS;AAI3B,oBAAQ;AAAA,cACN,cAAc,UAAU,gBAAgB;AAAA,cACxC,kBAAkB,UAAU,iBAAiB;AAAA,cAC7C,cACG,UAAU,gBAAgB,MAC1B,UAAU,iBAAiB;AAAA,YAChC;AACA,oBAAQ;AAAA,cACN,iCAAiC,MAAM,YAAY,QAAQ,MAAM,gBAAgB;AAAA,YACnF;AAEA,gBAAI,KAAK,cAAc;AACrB,mBAAK,aAAa;AAAA,gBAChB,MAAM;AAAA,gBACN,MAAM;AAAA,cACR;AAAA,YACF;AAAA,UACF;AAEA,gBAAM,eAAe,KAAK,oBAAoB,QAAQ;AACtD,cAAI,iBAAiB,UAAU;AAC7B,oBAAQ;AAAA,cACN;AAAA,YACF;AAAA,UACF;AAEA,gBAAM,UACJ,OAAO,SAAS,YAAY,WACxB,SAAS,UACT,KAAK,UAAU,SAAS,OAAO;AAErC,kBAAQ;AAAA,YACN,qCAAqC,QAAQ,MAAM,0BAA0B,YAAY;AAAA,UAC3F;AAEA,iBAAO,EAAE,SAAS,cAAc,MAAM;AAAA,QACxC,CAAC;AAED,gBAAQ;AAAA,UACN,4BAA4B,UAAU,CAAC,iBAAiB,KAAK,IAAI,IAAI,gBAAgB;AAAA,QACvF;AAEA,cAAM,SAAS,uBACV,kBAAkB,OAAO,SAAS,MAAS,IAC5C,kBAAkB,OAAO,SAAS,MAAM;AAC5C,gBAAQ;AAAA,UACN,gDAAgD,uBAAuB,iCAAiC,EAAE;AAAA,QAC5G;AAEA,eAAO;AAAA,UACL,MAAM;AAAA,UACN,KAAK,OAAO;AAAA,UACZ,cAAc,OAAO;AAAA,UACrB,OAAO,OAAO;AAAA,QAChB;AAAA,MACF,SAAS,OAAO;AACd,oBAAY,iBAAiB,QAAQ,QAAQ,IAAI,MAAM,OAAO,KAAK,CAAC;AACpE,gBAAQ;AAAA,UACN,4BAA4B,UAAU,CAAC;AAAA,UACvC,UAAU;AAAA,QACZ;AAEA,YAAI,KAAK,iBAAiB,SAAS,GAAG;AACpC,kBAAQ,MAAM,iDAAiD;AAC/D,gBAAM;AAAA,QACR;AAEA,YAAI,UAAU,cAAc,kBAAkB;AAC5C,kBAAQ,IAAI,gDAAgD;AAC5D,0BACE,GAAG,UAAU;AAAA;AAAA,iCACqB,UAAU,OAAO;AAAA;AAAA,QAEvD;AAAA,MACF;AAAA,IACF;AAEA,YAAQ,MAAM,yDAAyD;AACvE,UAAM;AAAA,EACR;AAAA,EAEQ,oBACN,UACiB;AACjB,UAAM,WAAW,SAAS;AAG1B,QAAI,UAAU,eAAe;AAC3B,YAAM,SAAS,SAAS;AACxB,UACE,WAAW,UACX,WAAW,YACX,WAAW,oBACX,WAAW,cACX;AACA,eAAO;AAAA,MACT;AAAA,IACF;AACA,WAAO;AAAA,EACT;AAAA,EAEA,MAAM,QAAQ,SAIM;AAClB,UAAM,WAAW,MAAM,KAAK,oBAAoB,OAAO;AACvD,WAAO,SAAS;AAAA,EAClB;AAAA,EAEA,MAAM,oBAAoB,SAIuC;AAC/D,UAAM,EAAE,cAAc,YAAY,UAAU,IAAI;AAEhD,WAAO,KAAK,YAAY,QAAQ,YAAY;AAC1C,YAAM,aAAa,YACf,KAAK,oBAAoB,EAAE,UAAU,CAAC,IACtC,KAAK;AAET,YAAM,WAAW,MAAM,WAAW,OAAO;AAAA,QACvC,EAAE,MAAM,UAAU,SAAS,aAAa;AAAA,QACxC,EAAE,MAAM,QAAQ,SAAS,WAAW;AAAA,MACtC,CAAC;AAED,UAAI,QAAyB;AAC7B,UAAI,SAAS,gBAAgB;AAC3B,cAAM,YAAY,SAAS;AAI3B,gBAAQ;AAAA,UACN,cAAc,UAAU,gBAAgB;AAAA,UACxC,kBAAkB,UAAU,iBAAiB;AAAA,UAC7C,cACG,UAAU,gBAAgB,MAAM,UAAU,iBAAiB;AAAA,QAChE;AAEA,YAAI,KAAK,cAAc;AACrB,eAAK,aAAa;AAAA,YAChB,MAAM;AAAA,YACN,MAAM;AAAA,UACR;AAAA,QACF;AAAA,MACF;AAEA,YAAM,eAAe,KAAK,oBAAoB,QAAQ;AACtD,YAAM,UACJ,OAAO,SAAS,YAAY,WACxB,SAAS,UACT,KAAK,UAAU,SAAS,OAAO;AAErC,aAAO,EAAE,KAAK,SAAS,cAAc,MAAM;AAAA,IAC7C,CAAC;AAAA,EACH;AAAA,EAEQ,iBAAiB,OAAuB;AAC9C,UAAM,UAAU,MAAM,QAAQ,YAAY;AAC1C,WACE,QAAQ,SAAS,YAAY,KAC7B,QAAQ,SAAS,KAAK,KACtB,QAAQ,SAAS,gBAAgB;AAAA,EAErC;AAAA;AAAA;AAAA;AAAA,EAMA,MAAM,cACJ,SACyB;AACzB,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,aAAa;AAAA,MACb;AAAA,MACA,uBAAuB;AAAA,MACvB;AAAA,MACA,UAAU;AAAA,IACZ,IAAI;AAEJ,QAAI,KAAK,aAAa,aAAa;AACjC,cAAQ;AAAA,QACN,sCAAsC,KAAK,QAAQ;AAAA,MACrD;AACA,aAAO,KAAK,iBAAiB,OAAO;AAAA,IACtC;AAEA,UAAM,kBACH,gBAAgB,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,aAAa,EAAE,UACnD,cAAc,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,aAAa,EAAE;AACpD,YAAQ;AAAA,MACN,6BAA6B,cAAc;AAAA,IAC7C;AAEA,QAAI,YAA0B;AAE9B,aAAS,UAAU,GAAG,WAAW,YAAY,WAAW;AACtD,UAAI;AACF,gBAAQ;AAAA,UACN,qCAAqC,UAAU,CAAC,IAAI,aAAa,CAAC;AAAA,QACpE;AAEA,cAAM,SAAS,MAAM,KAAK,YAAY,QAAQ,YAAY;AACxD,gBAAM,YAAY,IAAI,UAAU;AAEhC,gBAAM,gBACJ,gBAAgB,aAAa,SAAS,IAClC,aAAa,IAAI,CAAC,OAAO;AAAA,YACvB,MAAM;AAAA,YACN,MAAM,EAAE;AAAA,YACR,GAAI,EAAE,gBACF,EAAE,eAAe,EAAE,cAAc,IACjC,CAAC;AAAA,UACP,EAAE,IACF,eACE,CAAC,EAAE,MAAM,QAAiB,MAAM,aAAa,CAAC,IAC9C,CAAC;AAET,gBAAM,cACJ,cAAc,WAAW,SAAS,IAC9B,WAAW,IAAI,CAAC,OAAO;AAAA,YACrB,MAAM;AAAA,YACN,MAAM,EAAE;AAAA,YACR,GAAI,EAAE,gBACF,EAAE,eAAe,EAAE,cAAc,IACjC,CAAC;AAAA,UACP,EAAE,IACF,aACE,CAAC,EAAE,MAAM,QAAiB,MAAM,WAAW,CAAC,IAC5C,CAAC;AAET,gBAAM,WAAW,MAAM,UAAU,SAAS,OAAO;AAAA,YAC/C,OAAO,KAAK;AAAA,YACZ,YAAY,aAAa;AAAA,YACzB,aAAa,eAAe;AAAA,YAC5B,QAAQ;AAAA,YACR,UAAU,CAAC,EAAE,MAAM,QAAQ,SAAS,YAAY,CAAC;AAAA,UACnD,CAAC;AAED,gBAAM,cAAc,SAAS,QAAQ,KAAK,CAAC,MAAM,EAAE,SAAS,MAAM;AAClE,gBAAM,UACJ,eAAe,UAAU,cAAc,YAAY,OAAO;AAE5D,gBAAM,WAAW,SAAS;AAO1B,gBAAM,YAAY,SAAS,2BAA2B;AACtD,gBAAM,gBAAgB,SAAS,+BAA+B;AAE9D,cAAI,gBAAgB,GAAG;AACrB,oBAAQ;AAAA,cACN,0CAA0C,aAAa;AAAA,YACzD;AAAA,UACF;AACA,cAAI,YAAY,GAAG;AACjB,kBAAM,iBAAiB,KAAK;AAAA,cACzB,aAAa,YAAY,SAAS,gBAAiB;AAAA,YACtD;AACA,oBAAQ;AAAA,cACN,wCAAwC,SAAS,aAAa,cAAc;AAAA,YAC9E;AAAA,UACF;AACA,cAAI,kBAAkB,KAAK,cAAc,GAAG;AAC1C,oBAAQ;AAAA,cACN,yCAAyC,SAAS,YAAY;AAAA,YAChE;AAAA,UACF;AAEA,gBAAM,QAAkB;AAAA,YACtB,cAAc,SAAS;AAAA,YACvB,kBAAkB,SAAS;AAAA,YAC3B,aAAa,SAAS,eAAe,SAAS;AAAA,UAChD;AAEA,cAAI,KAAK,cAAc;AACrB,iBAAK,aAAa;AAAA,cAChB,MAAM;AAAA,cACN,MAAM;AAAA,YACR;AAAA,UACF;AAEA,gBAAM,eACJ,SAAS,gBAAgB,aACrB,SACA,SAAS;AAEf,iBAAO;AAAA,YACL;AAAA,YACA;AAAA,YACA;AAAA,UACF;AAAA,QACF,CAAC;AAED,YAAI;AACJ,YAAI,SAAS;AACX,mBAAS,OAAO;AAAA,QAClB,WAAW,sBAAsB;AAC/B,mBAAS,kBAAkB,OAAO,SAAS,MAAS;AAAA,QACtD,OAAO;AACL,mBAAS,kBAAkB,OAAO,SAAS,MAAM;AAAA,QACnD;AAEA,eAAO;AAAA,UACL,MAAM;AAAA,UACN,KAAK,OAAO;AAAA,UACZ,cAAc,OAAO;AAAA,UACrB,OAAO,OAAO;AAAA,QAChB;AAAA,MACF,SAAS,OAAO;AACd,oBAAY,iBAAiB,QAAQ,QAAQ,IAAI,MAAM,OAAO,KAAK,CAAC;AACpE,gBAAQ;AAAA,UACN,qCAAqC,UAAU,CAAC;AAAA,UAChD,UAAU;AAAA,QACZ;AAEA,YAAI,KAAK,iBAAiB,SAAS,GAAG;AACpC,gBAAM;AAAA,QACR;AAAA,MACF;AAAA,IACF;AAEA,UAAM;AAAA,EACR;AAAA,EAEA,OAAO,eAAe,MAAc,QAAQ,MAAsB;AAChE,WAAO,QACH,EAAE,MAAM,QAAQ,MAAM,eAAe,EAAE,MAAM,YAAY,EAAE,IAC3D,EAAE,MAAM,QAAQ,KAAK;AAAA,EAC3B;AACF;AAMA,IAAM,gBAAyD,CAAC;AAEzD,SAAS,mBAAmB,SAAuC;AACxE,QAAM,WAAW,SAAS,YAAY;AACtC,MAAI,CAAC,cAAc,QAAQ,GAAG;AAC5B,kBAAc,QAAQ,IAAI,IAAI,UAAU,OAAO;AAAA,EACjD;AACA,SAAO,cAAc,QAAQ;AAC/B;AAEO,SAAS,qBAAqB,UAA8B;AACjE,MAAI,UAAU;AACZ,WAAO,cAAc,QAAQ;AAAA,EAC/B,OAAO;AACL,eAAW,OAAO,OAAO,KAAK,aAAa,GAAoB;AAC7D,aAAO,cAAc,GAAG;AAAA,IAC1B;AAAA,EACF;AACF;AAMO,SAAS,uBAAoC;AAClD,MAAI,QAAQ,IAAI,kBAAmB,QAAO;AAC1C,MAAI,QAAQ,IAAI,iBAAkB,QAAO;AACzC,MAAI,QAAQ,IAAI,aAAc,QAAO;AACrC,MAAI,QAAQ,IAAI,eAAgB,QAAO;AACvC,QAAM,IAAI;AAAA,IACR;AAAA,EACF;AACF;AAEO,SAAS,oBAAoB,UAAgC;AAClE,UAAQ,UAAU;AAAA,IAChB,KAAK;AACH,aAAO,CAAC,CAAC,QAAQ,IAAI;AAAA,IACvB,KAAK;AACH,aAAO,CAAC,CAAC,QAAQ,IAAI;AAAA,IACvB,KAAK;AACH,aAAO,CAAC,CAAC,QAAQ,IAAI;AAAA,IACvB,KAAK;AACH,aAAO,CAAC,CAAC,QAAQ,IAAI;AAAA,IACvB;AACE,aAAO;AAAA,EACX;AACF;AAMO,SAAS,yBACd,SACW;AACX,QAAM,WAAW,SAAS,YAAY,qBAAqB;AAC3D,QAAM,eACJ,aAAa,aAAa,gBAAgB,OAAO,cAAc;AACjE,SAAO,IAAI,UAAU;AAAA,IACnB;AAAA,IACA,OAAO;AAAA,IACP,aAAa;AAAA,IACb,GAAG;AAAA,EACL,CAAC;AACH;AAEO,SAAS,qBACd,SACW;AACX,QAAM,WAAW,SAAS,YAAY,qBAAqB;AAC3D,QAAM,eACJ,aAAa,aAAa,gBAAgB,WAAW,cAAc;AACrE,SAAO,IAAI,UAAU;AAAA,IACnB;AAAA,IACA,OAAO;AAAA,IACP,aAAa;AAAA,IACb,GAAG;AAAA,EACL,CAAC;AACH;AAEO,SAAS,gBACd,SACW;AACX,QAAM,WAAW,SAAS,YAAY,qBAAqB;AAC3D,QAAM,eACJ,aAAa,aACT,gBAAgB,OAChB,cAAc;AACpB,SAAO,IAAI,UAAU;AAAA,IACnB;AAAA,IACA,OAAO;AAAA,IACP,aAAa;AAAA,IACb,GAAG;AAAA,EACL,CAAC;AACH;AAEO,SAAS,qBACd,SACW;AACX,SAAO,IAAI,UAAU;AAAA,IACnB,UAAU;AAAA,IACV,OAAO,gBAAgB;AAAA,IACvB,GAAG;AAAA,EACL,CAAC;AACH;AAEO,SAAS,mBACd,SACW;AACX,SAAO,IAAI,UAAU;AAAA,IACnB,UAAU;AAAA,IACV,OAAO,cAAc;AAAA,IACrB,GAAG;AAAA,EACL,CAAC;AACH;AAEO,SAAS,sBACd,SACW;AACX,SAAO,IAAI,UAAU;AAAA,IACnB,UAAU;AAAA,IACV,OAAO,iBAAiB;AAAA,IACxB,GAAG;AAAA,EACL,CAAC;AACH;AAEO,SAAS,iBACd,SACW;AACX,SAAO,IAAI,UAAU;AAAA,IACnB,UAAU;AAAA,IACV,OAAO,YAAY;AAAA,IACnB,GAAG;AAAA,EACL,CAAC;AACH;","names":[]}

package/dist/client.d.ts

@@ -1,136 +0,0 @@
-import { a as RateLimiterOptions, b as TokenUsage } from './rate-limiter-9XAWfHwe.js';
-import { ChatOpenAI } from '@langchain/openai';
-import { ChatAnthropic, ChatAnthropicCallOptions } from '@langchain/anthropic';
-import { z } from 'zod';
-
-type MessageLike = any;
-declare class CachingChatAnthropic extends ChatAnthropic {
-    invoke(input: MessageLike[] | string, options?: Partial<ChatAnthropicCallOptions>): Promise<MessageLike>;
-}
-type ChatModel = ChatOpenAI | CachingChatAnthropic;
-type LLMProvider = 'openai' | 'deepseek' | 'anthropic' | 'kimi';
-interface ProviderConfig {
-    apiKey: string;
-    baseUrl?: string;
-    defaultModel: string;
-}
-interface LLMClientOptions {
-    provider?: LLMProvider;
-    model?: string;
-    temperature?: number;
-    streaming?: boolean;
-    rateLimiter?: RateLimiterOptions;
-    useGlobalRateLimiter?: boolean;
-    trackTokens?: boolean;
-}
-interface LLMCallOptions<T = unknown> {
-    systemPrompt: string;
-    userPrompt: string;
-    schema?: z.ZodSchema<T>;
-    maxRetries?: number;
-    retryWithContext?: boolean;
-    maxTokens?: number;
-    skipSchemaValidation?: boolean;
-    temperature?: number;
-}
-interface CacheableBlock {
-    type: 'text';
-    text: string;
-    cache_control?: {
-        type: 'ephemeral';
-    };
-}
-interface CacheAwareLLMCallOptions<T = unknown> extends LLMCallOptions<T> {
-    systemBlocks?: CacheableBlock[];
-    userBlocks?: CacheableBlock[];
-    rawText?: boolean;
-}
-interface LLMUsage {
-    promptTokens: number;
-    completionTokens: number;
-    totalTokens: number;
-}
-type LLMFinishReason = 'stop' | 'length' | 'content_filter' | 'tool_calls' | null;
-interface LLMResponse<T> {
-    data: T;
-    raw: string;
-    finishReason: LLMFinishReason;
-    usage: LLMUsage | null;
-}
-declare const DEEPSEEK_MODELS: {
-    readonly CHAT: "deepseek-chat";
-    readonly CODER: "deepseek-coder";
-    readonly REASONER: "deepseek-reasoner";
-};
-declare const OPENAI_MODELS: {
-    readonly GPT4O: "gpt-4o";
-    readonly GPT4O_MINI: "gpt-4o-mini";
-    readonly GPT4_TURBO: "gpt-4-turbo";
-    readonly GPT35_TURBO: "gpt-3.5-turbo";
-    readonly GPT_5_1: "gpt-5.1";
-};
-declare const ANTHROPIC_MODELS: {
-    readonly CLAUDE_SONNET_4_5: "claude-sonnet-4-5-20250929";
-    readonly CLAUDE_SONNET_4: "claude-sonnet-4-20250514";
-    readonly CLAUDE_OPUS_4_5: "claude-opus-4-5-20250929";
-    readonly CLAUDE_3_5_HAIKU: "claude-3-5-haiku-20241022";
-};
-declare const KIMI_MODELS: {
-    readonly K2_5: "kimi-k2.5";
-};
-declare class LLMClient {
-    private model;
-    private rateLimiter;
-    private tokenTracker;
-    private modelName;
-    private provider;
-    private providerConfig;
-    private temperature;
-    private streaming;
-    constructor(options?: LLMClientOptions);
-    private usesMaxCompletionTokens;
-    private createModel;
-    private getModelWithOptions;
-    getProvider(): LLMProvider;
-    getModelName(): string;
-    getModel(): ChatModel;
-    getRateLimiterStatus(): {
-        queueLength: number;
-        activeRequests: number;
-        minuteTokens: number;
-        secondTokens: number;
-        backoffMs: number;
-    };
-    getTokenUsage(): TokenUsage | null;
-    call<T>(options: LLMCallOptions<T>): Promise<T>;
-    callWithMetadata<T>(options: LLMCallOptions<T>): Promise<LLMResponse<T>>;
-    private extractFinishReason;
-    callRaw(options: {
-        systemPrompt: string;
-        userPrompt: string;
-        maxTokens?: number;
-    }): Promise<string>;
-    callRawWithMetadata(options: {
-        systemPrompt: string;
-        userPrompt: string;
-        maxTokens?: number;
-    }): Promise<Omit<LLMResponse<string>, 'data'> & {
-        raw: string;
-    }>;
-    private isRateLimitError;
-    callWithCache<T>(options: CacheAwareLLMCallOptions<T>): Promise<LLMResponse<T>>;
-    static cacheableBlock(text: string, cache?: boolean): CacheableBlock;
-}
-declare function getSharedLLMClient(options?: LLMClientOptions): LLMClient;
-declare function resetSharedLLMClient(provider?: LLMProvider): void;
-declare function getAvailableProvider(): LLMProvider;
-declare function isProviderAvailable(provider: LLMProvider): boolean;
-declare function createRequirementsClient(options?: Partial<LLMClientOptions>): LLMClient;
-declare function createCreativeClient(options?: Partial<LLMClientOptions>): LLMClient;
-declare function createFixClient(options?: Partial<LLMClientOptions>): LLMClient;
-declare function createDeepSeekClient(options?: Partial<Omit<LLMClientOptions, 'provider'>>): LLMClient;
-declare function createOpenAIClient(options?: Partial<Omit<LLMClientOptions, 'provider'>>): LLMClient;
-declare function createAnthropicClient(options?: Partial<Omit<LLMClientOptions, 'provider'>>): LLMClient;
-declare function createKimiClient(options?: Partial<Omit<LLMClientOptions, 'provider'>>): LLMClient;
-
-export { ANTHROPIC_MODELS, type CacheAwareLLMCallOptions, type CacheableBlock, DEEPSEEK_MODELS, KIMI_MODELS, type LLMCallOptions, LLMClient, type LLMClientOptions, type LLMFinishReason, type LLMProvider, type LLMResponse, type LLMUsage, OPENAI_MODELS, type ProviderConfig, createAnthropicClient, createCreativeClient, createDeepSeekClient, createFixClient, createKimiClient, createOpenAIClient, createRequirementsClient, getAvailableProvider, getSharedLLMClient, isProviderAvailable, resetSharedLLMClient };

package/dist/index.d.ts

@@ -1,67 +0,0 @@
-import { LLMFinishReason, LLMClient } from './client.js';
-export { ANTHROPIC_MODELS, CacheAwareLLMCallOptions, CacheableBlock, DEEPSEEK_MODELS, KIMI_MODELS, LLMCallOptions, LLMClientOptions, LLMProvider, LLMResponse, LLMUsage, OPENAI_MODELS, ProviderConfig, createAnthropicClient, createCreativeClient, createDeepSeekClient, createFixClient, createKimiClient, createOpenAIClient, createRequirementsClient, getAvailableProvider, getSharedLLMClient, isProviderAvailable, resetSharedLLMClient } from './client.js';
-export { R as RateLimiter, a as RateLimiterOptions, T as TokenTracker, b as TokenUsage, g as getGlobalRateLimiter, c as getGlobalTokenTracker, r as resetGlobalRateLimiter, d as resetGlobalTokenTracker } from './rate-limiter-9XAWfHwe.js';
-export { autoCloseJson, extractJsonFromText, isValidJson, parseJsonResponse, safeParseJson } from './json-parser.js';
-import { z } from 'zod';
-export { JsonSchema, STRUCTURED_OUTPUT_MODELS, StructuredGenerationOptions, StructuredGenerationResult, StructuredOutputClient, StructuredOutputOptions, getStructuredOutputClient, isStructuredOutputAvailable, resetStructuredOutputClient } from './structured-output.js';
-import '@langchain/openai';
-import '@langchain/anthropic';
-
-/**
- * Truncation Detector
- *
- * Utilities for detecting when LLM output has been truncated and
- * extracting usable content from partial responses.
- *
- * @packageDocumentation
- */
-
-type TruncationReason = 'finish_reason' | 'json_incomplete' | 'bracket_mismatch' | 'none';
-interface TruncationResult {
-    isTruncated: boolean;
-    reason: TruncationReason;
-    partialContent?: string;
-    lastCompleteElement?: unknown;
-    missingCloseBrackets?: number;
-    missingCloseBraces?: number;
-}
-declare function detectTruncation(response: string, finishReason: LLMFinishReason): TruncationResult;
-declare function findLastCompleteElement(json: string): unknown | null;
-declare function isLikelyTruncated(content: string): boolean;
-
-/**
- * LLM Continuation Utility
- *
- * Handles truncated LLM responses with automatic continuation.
- * - Detects truncation via finish_reason and JSON structure
- * - Automatically continues with full context
- * - Merges partial and continuation responses
- * - Salvages partial data if max continuations reached
- *
- * @packageDocumentation
- */
-
-interface ContinuationOptions<T> {
-    client: LLMClient;
-    systemPrompt: string;
-    userPrompt: string;
-    schema?: z.ZodSchema<T>;
-    maxTokens?: number;
-    maxContinuations?: number;
-    maxRetries?: number;
-    buildContinuationPrompt: (partialResponse: string, attempt: number) => string;
-    continuationSystemPrompt?: string;
-}
-interface ContinuationResult<T> {
-    data: T;
-    raw: string;
-    continuationCount: number;
-    warnings: string[];
-    wasSalvaged: boolean;
-}
-declare function mergeResponses(previous: string, continuation: string): string;
-declare function salvagePartialResponse<T>(rawResponse: string): T | null;
-declare function callWithContinuation<T>(options: ContinuationOptions<T>): Promise<ContinuationResult<T>>;
-declare function buildGenericContinuationPrompt(context: string, partialResponse: string, attempt: number, maxAttempts?: number): string;
-
-export { type ContinuationOptions, type ContinuationResult, LLMClient, LLMFinishReason, type TruncationReason, type TruncationResult, buildGenericContinuationPrompt, callWithContinuation, detectTruncation, findLastCompleteElement, isLikelyTruncated, mergeResponses, salvagePartialResponse };

package/dist/json-parser.d.ts

@@ -1,43 +0,0 @@
-import { z } from 'zod';
-
-/**
- * JSON Parser Utilities
- *
- * Robust JSON parsing for LLM responses that may contain:
- * - Markdown code blocks
- * - Extra text before/after JSON
- * - Minor formatting issues
- *
- * @packageDocumentation
- */
-
-/**
- * Extract JSON from LLM response text.
- *
- * Handles markdown code blocks, raw JSON objects/arrays, and primitive values.
- */
-declare function extractJsonFromText(text: string): string | null;
-/**
- * Parse JSON from LLM response with optional Zod schema validation.
- */
-declare function parseJsonResponse<T>(response: string, schema?: z.ZodSchema<T>): T;
-/**
- * Safely parse JSON without throwing.
- */
-declare function safeParseJson<T>(response: string, schema?: z.ZodSchema<T>): {
-    success: true;
-    data: T;
-} | {
-    success: false;
-    error: Error;
-};
-/**
- * Check if a string is valid JSON.
- */
-declare function isValidJson(str: string): boolean;
-/**
- * Attempt to auto-close unclosed JSON brackets.
- */
-declare function autoCloseJson(json: string): string;
-
-export { autoCloseJson, extractJsonFromText, isValidJson, parseJsonResponse, safeParseJson };