@within-7/minto 0.1.7 → 0.2.0

package/dist/services/modelOrchestrator.js

@@ -0,0 +1,310 @@
+import { randomUUID } from "crypto";
+import { getModelManager } from "../utils/model.js";
+import { queryModel as queryModelService } from "./claude.js";
+function extractTextFromMessage(message) {
+  const content = message.message.content;
+  if (typeof content === "string") {
+    return content;
+  }
+  if (Array.isArray(content)) {
+    return content.filter(
+      (block) => block.type === "text"
+    ).map((block) => block.text).join("\n");
+  }
+  return "";
+}
+function createUserMessage(prompt) {
+  return {
+    message: { role: "user", content: prompt },
+    type: "user",
+    uuid: randomUUID()
+  };
+}
+class ModelOrchestrator {
+  modelManager = getModelManager();
+  /**
+   * Execute task with multi-model collaboration
+   */
+  async execute(task, config) {
+    const startTime = Date.now();
+    switch (config.mode) {
+      case "sequential":
+        return this.executeSequential(task, config, startTime);
+      case "parallel":
+        return this.executeParallel(task, config, startTime);
+      case "debate":
+        return this.executeDebate(task, config, startTime);
+      case "consensus":
+        return this.executeConsensus(task, config, startTime);
+      default:
+        throw new Error(`Unknown collaboration mode: ${config.mode}`);
+    }
+  }
+  /**
+   * Query a single model with a prompt
+   */
+  async querySingleModel(modelName, prompt) {
+    const startTime = Date.now();
+    try {
+      const profile = this.modelManager.resolveModel(modelName);
+      if (!profile) {
+        return {
+          model: modelName,
+          response: "",
+          success: false,
+          error: `Model not found: ${modelName}`,
+          latencyMs: Date.now() - startTime
+        };
+      }
+      const messages = [
+        createUserMessage(prompt)
+      ];
+      const response = await queryModelService("main", messages, [], void 0);
+      const responseText = extractTextFromMessage(response);
+      return {
+        model: modelName,
+        response: responseText,
+        success: true,
+        latencyMs: Date.now() - startTime
+      };
+    } catch (error) {
+      return {
+        model: modelName,
+        response: "",
+        success: false,
+        error: error instanceof Error ? error.message : String(error),
+        latencyMs: Date.now() - startTime
+      };
+    }
+  }
+  /**
+   * Sequential mode: Chain responses through models
+   * A → B → C (each model refines the previous output)
+   */
+  async executeSequential(task, config, startTime) {
+    const results = [];
+    let currentPrompt = task;
+    let finalResponse = "";
+    for (const modelName of config.models) {
+      const result = await this.querySingleModel(modelName, currentPrompt);
+      results.push(result);
+      if (!result.success) {
+        if (config.strategy?.fallback) {
+          const fallbackResult = await this.querySingleModel(
+            config.strategy.fallback,
+            currentPrompt
+          );
+          results.push(fallbackResult);
+          if (fallbackResult.success) {
+            currentPrompt = `Previous response: ${fallbackResult.response}
+
+Please review and improve.`;
+            finalResponse = fallbackResult.response;
+            continue;
+          }
+        }
+        break;
+      }
+      currentPrompt = `Previous analysis:
+${result.response}
+
+Please review and enhance this analysis.`;
+      finalResponse = result.response;
+    }
+    return {
+      mode: "sequential",
+      models: config.models,
+      finalResponse,
+      individualResults: results,
+      metadata: {
+        totalLatencyMs: Date.now() - startTime,
+        totalTokensUsed: this.sumTokens(results),
+        iterations: results.length
+      }
+    };
+  }
+  /**
+   * Parallel mode: Query all models simultaneously
+   * A + B + C (gather multiple perspectives)
+   */
+  async executeParallel(task, config, startTime) {
+    const results = await Promise.all(
+      config.models.map((model) => this.querySingleModel(model, task))
+    );
+    const successfulResults = results.filter((r) => r.success);
+    const finalResponse = this.mergeParallelResults(successfulResults);
+    return {
+      mode: "parallel",
+      models: config.models,
+      finalResponse,
+      individualResults: results,
+      metadata: {
+        totalLatencyMs: Date.now() - startTime,
+        totalTokensUsed: this.sumTokens(results)
+      }
+    };
+  }
+  /**
+   * Debate mode: Models review each other's work
+   * A proposes → B critiques → A refines
+   */
+  async executeDebate(task, config, startTime) {
+    const maxIterations = config.strategy?.maxIterations ?? 2;
+    const [modelA, modelB] = config.models;
+    if (!modelA || !modelB) {
+      throw new Error("Debate mode requires at least 2 models");
+    }
+    const results = [];
+    let currentProposal = "";
+    const proposalResult = await this.querySingleModel(modelA, task);
+    results.push(proposalResult);
+    if (!proposalResult.success) {
+      return {
+        mode: "debate",
+        models: config.models,
+        finalResponse: "",
+        individualResults: results,
+        metadata: {
+          totalLatencyMs: Date.now() - startTime,
+          iterations: 1
+        }
+      };
+    }
+    currentProposal = proposalResult.response;
+    for (let i = 0; i < maxIterations; i++) {
+      const critiquePrompt = `Review the following response and provide constructive feedback:
+
+${currentProposal}
+
+Original task: ${task}`;
+      const critiqueResult = await this.querySingleModel(modelB, critiquePrompt);
+      results.push(critiqueResult);
+      if (!critiqueResult.success) break;
+      const refinePrompt = `Original task: ${task}
+
+Your previous response:
+${currentProposal}
+
+Feedback received:
+${critiqueResult.response}
+
+Please improve your response addressing the feedback.`;
+      const refineResult = await this.querySingleModel(modelA, refinePrompt);
+      results.push(refineResult);
+      if (!refineResult.success) break;
+      currentProposal = refineResult.response;
+    }
+    return {
+      mode: "debate",
+      models: config.models,
+      finalResponse: currentProposal,
+      individualResults: results,
+      metadata: {
+        totalLatencyMs: Date.now() - startTime,
+        totalTokensUsed: this.sumTokens(results),
+        iterations: Math.ceil(results.length / 2)
+      }
+    };
+  }
+  /**
+   * Consensus mode: Models vote and merge best ideas
+   */
+  async executeConsensus(task, config, startTime) {
+    const results = await Promise.all(
+      config.models.map((model) => this.querySingleModel(model, task))
+    );
+    const successfulResults = results.filter((r) => r.success);
+    if (successfulResults.length === 0) {
+      return {
+        mode: "consensus",
+        models: config.models,
+        finalResponse: "",
+        individualResults: results,
+        metadata: {
+          totalLatencyMs: Date.now() - startTime,
+          consensusReached: false
+        }
+      };
+    }
+    const synthesisModel = config.models[0];
+    const responseSummary = successfulResults.map((r, i) => `Response ${i + 1} (${r.model}):
+${r.response}`).join("\n\n---\n\n");
+    const consensusPrompt = `You are tasked with finding the consensus among multiple AI responses.
+
+Original task: ${task}
+
+Here are the different responses:
+
+${responseSummary}
+
+Please synthesize these responses into a single, comprehensive answer that:
+1. Identifies common themes and agreements
+2. Resolves any contradictions by choosing the most well-reasoned position
+3. Combines the best elements from each response
+4. Presents a unified, coherent answer
+
+Synthesized response:`;
+    const consensusResult = await this.querySingleModel(
+      synthesisModel,
+      consensusPrompt
+    );
+    results.push(consensusResult);
+    return {
+      mode: "consensus",
+      models: config.models,
+      finalResponse: consensusResult.success ? consensusResult.response : successfulResults[0].response,
+      individualResults: results,
+      metadata: {
+        totalLatencyMs: Date.now() - startTime,
+        totalTokensUsed: this.sumTokens(results),
+        consensusReached: consensusResult.success
+      }
+    };
+  }
+  /**
+   * Merge parallel results into a combined response
+   */
+  mergeParallelResults(results) {
+    if (results.length === 0) return "";
+    if (results.length === 1) return results[0].response;
+    const formattedResponses = results.map((r, i) => `## Perspective ${i + 1} (${r.model})
+
+${r.response}`).join("\n\n---\n\n");
+    return `# Multi-Model Analysis
+
+The following perspectives were gathered from ${results.length} different models:
+
+${formattedResponses}
+
+---
+
+*Note: These are independent analyses. Consider synthesizing the best elements from each perspective.*`;
+  }
+  /**
+   * Sum token usage across results
+   */
+  sumTokens(results) {
+    const tokensUsed = results.filter((r) => r.tokenUsage).map((r) => (r.tokenUsage?.input || 0) + (r.tokenUsage?.output || 0));
+    return tokensUsed.length > 0 ? tokensUsed.reduce((a, b) => a + b, 0) : void 0;
+  }
+}
+let orchestratorInstance = null;
+function getModelOrchestrator() {
+  if (!orchestratorInstance) {
+    orchestratorInstance = new ModelOrchestrator();
+  }
+  return orchestratorInstance;
+}
+function resetModelOrchestrator() {
+  orchestratorInstance = null;
+}
+async function orchestrate(task, config) {
+  return getModelOrchestrator().execute(task, config);
+}
+export {
+  ModelOrchestrator,
+  getModelOrchestrator,
+  orchestrate,
+  resetModelOrchestrator
+};
+//# sourceMappingURL=modelOrchestrator.js.map

package/dist/services/modelOrchestrator.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/services/modelOrchestrator.ts"],
+  "sourcesContent": ["/**\n * Model Orchestrator\n *\n * Enables multi-model collaboration with different modes:\n * - Sequential: Chain of thought through multiple models\n * - Parallel: Multiple perspectives from different models\n * - Debate: Adversarial review between models\n * - Consensus: Best of both through voting\n */\n\nimport { randomUUID } from 'crypto'\nimport { getModelManager } from '@utils/model'\nimport { queryModel as queryModelService } from './claude'\nimport type { AssistantMessage, UserMessage } from '@query'\n\n/**\n * Collaboration mode for multi-model tasks\n */\nexport type CollaborationMode =\n  | 'sequential'\n  | 'parallel'\n  | 'debate'\n  | 'consensus'\n\n/**\n * Model collaboration configuration\n */\nexport interface ModelCollaborationConfig {\n  /** Collaboration mode */\n  mode: CollaborationMode\n  /** Models to use (by name or profile) */\n  models: string[]\n  /** Collaboration strategy */\n  strategy?: {\n    /** Task type to model mapping */\n    taskRouting?: Record<string, string>\n    /** Fallback model on failure */\n    fallback?: string\n    /** Consensus threshold (0-1) */\n    voteThreshold?: number\n    /** Maximum iterations for debate mode */\n    maxIterations?: number\n  }\n}\n\n/**\n * Result from a single model query\n */\nexport interface ModelQueryResult {\n  model: string\n  response: string\n  success: boolean\n  error?: string\n  latencyMs: number\n  tokenUsage?: {\n    input: number\n    output: number\n  }\n}\n\n/**\n * Combined result from orchestrated query\n */\nexport interface OrchestratedResult {\n  mode: CollaborationMode\n  models: string[]\n  finalResponse: string\n  individualResults: ModelQueryResult[]\n  metadata: {\n    totalLatencyMs: number\n    totalTokensUsed?: number\n    iterations?: number\n    consensusReached?: boolean\n  }\n}\n\n/**\n * Extract text content from an AssistantMessage\n */\nfunction extractTextFromMessage(message: AssistantMessage): string {\n  const content = message.message.content\n  if (typeof content === 'string') {\n    return content\n  }\n  if (Array.isArray(content)) {\n    return content\n      .filter(\n        (block): block is { type: 'text'; text: string } =>\n          block.type === 'text',\n      )\n      .map(block => block.text)\n      .join('\\n')\n  }\n  return ''\n}\n\n/**\n * Create a user message for the query\n */\nfunction createUserMessage(prompt: string): UserMessage {\n  return {\n    message: { role: 'user', content: prompt },\n    type: 'user',\n    uuid: randomUUID(),\n  }\n}\n\n/**\n * Model Orchestrator for multi-model collaboration\n */\nexport class ModelOrchestrator {\n  private modelManager = getModelManager()\n\n  /**\n   * Execute task with multi-model collaboration\n   */\n  async execute(\n    task: string,\n    config: ModelCollaborationConfig,\n  ): Promise<OrchestratedResult> {\n    const startTime = Date.now()\n\n    switch (config.mode) {\n      case 'sequential':\n        return this.executeSequential(task, config, startTime)\n      case 'parallel':\n        return this.executeParallel(task, config, startTime)\n      case 'debate':\n        return this.executeDebate(task, config, startTime)\n      case 'consensus':\n        return this.executeConsensus(task, config, startTime)\n      default:\n        throw new Error(`Unknown collaboration mode: ${config.mode}`)\n    }\n  }\n\n  /**\n   * Query a single model with a prompt\n   */\n  private async querySingleModel(\n    modelName: string,\n    prompt: string,\n  ): Promise<ModelQueryResult> {\n    const startTime = Date.now()\n\n    try {\n      const profile = this.modelManager.resolveModel(modelName)\n      if (!profile) {\n        return {\n          model: modelName,\n          response: '',\n          success: false,\n          error: `Model not found: ${modelName}`,\n          latencyMs: Date.now() - startTime,\n        }\n      }\n\n      // Create messages array for the query\n      const messages: (UserMessage | AssistantMessage)[] = [\n        createUserMessage(prompt),\n      ]\n\n      // Query using 'main' pointer - the actual model is determined by config\n      // For orchestration, we use the main model pointer\n      const response = await queryModelService('main', messages, [], undefined)\n      const responseText = extractTextFromMessage(response)\n\n      return {\n        model: modelName,\n        response: responseText,\n        success: true,\n        latencyMs: Date.now() - startTime,\n      }\n    } catch (error) {\n      return {\n        model: modelName,\n        response: '',\n        success: false,\n        error: error instanceof Error ? error.message : String(error),\n        latencyMs: Date.now() - startTime,\n      }\n    }\n  }\n\n  /**\n   * Sequential mode: Chain responses through models\n   * A \u2192 B \u2192 C (each model refines the previous output)\n   */\n  private async executeSequential(\n    task: string,\n    config: ModelCollaborationConfig,\n    startTime: number,\n  ): Promise<OrchestratedResult> {\n    const results: ModelQueryResult[] = []\n    let currentPrompt = task\n    let finalResponse = ''\n\n    for (const modelName of config.models) {\n      const result = await this.querySingleModel(modelName, currentPrompt)\n      results.push(result)\n\n      if (!result.success) {\n        // Try fallback if available\n        if (config.strategy?.fallback) {\n          const fallbackResult = await this.querySingleModel(\n            config.strategy.fallback,\n            currentPrompt,\n          )\n          results.push(fallbackResult)\n          if (fallbackResult.success) {\n            currentPrompt = `Previous response: ${fallbackResult.response}\\n\\nPlease review and improve.`\n            finalResponse = fallbackResult.response\n            continue\n          }\n        }\n        break\n      }\n\n      // Use this response as context for next model\n      currentPrompt = `Previous analysis:\\n${result.response}\\n\\nPlease review and enhance this analysis.`\n      finalResponse = result.response\n    }\n\n    return {\n      mode: 'sequential',\n      models: config.models,\n      finalResponse,\n      individualResults: results,\n      metadata: {\n        totalLatencyMs: Date.now() - startTime,\n        totalTokensUsed: this.sumTokens(results),\n        iterations: results.length,\n      },\n    }\n  }\n\n  /**\n   * Parallel mode: Query all models simultaneously\n   * A + B + C (gather multiple perspectives)\n   */\n  private async executeParallel(\n    task: string,\n    config: ModelCollaborationConfig,\n    startTime: number,\n  ): Promise<OrchestratedResult> {\n    const results = await Promise.all(\n      config.models.map(model => this.querySingleModel(model, task)),\n    )\n\n    const successfulResults = results.filter(r => r.success)\n    const finalResponse = this.mergeParallelResults(successfulResults)\n\n    return {\n      mode: 'parallel',\n      models: config.models,\n      finalResponse,\n      individualResults: results,\n      metadata: {\n        totalLatencyMs: Date.now() - startTime,\n        totalTokensUsed: this.sumTokens(results),\n      },\n    }\n  }\n\n  /**\n   * Debate mode: Models review each other's work\n   * A proposes \u2192 B critiques \u2192 A refines\n   */\n  private async executeDebate(\n    task: string,\n    config: ModelCollaborationConfig,\n    startTime: number,\n  ): Promise<OrchestratedResult> {\n    const maxIterations = config.strategy?.maxIterations ?? 2\n    const [modelA, modelB] = config.models\n\n    if (!modelA || !modelB) {\n      throw new Error('Debate mode requires at least 2 models')\n    }\n\n    const results: ModelQueryResult[] = []\n    let currentProposal = ''\n\n    // Initial proposal from Model A\n    const proposalResult = await this.querySingleModel(modelA, task)\n    results.push(proposalResult)\n\n    if (!proposalResult.success) {\n      return {\n        mode: 'debate',\n        models: config.models,\n        finalResponse: '',\n        individualResults: results,\n        metadata: {\n          totalLatencyMs: Date.now() - startTime,\n          iterations: 1,\n        },\n      }\n    }\n\n    currentProposal = proposalResult.response\n\n    // Debate iterations\n    for (let i = 0; i < maxIterations; i++) {\n      // Model B critiques\n      const critiquePrompt = `Review the following response and provide constructive feedback:\\n\\n${currentProposal}\\n\\nOriginal task: ${task}`\n      const critiqueResult = await this.querySingleModel(modelB, critiquePrompt)\n      results.push(critiqueResult)\n\n      if (!critiqueResult.success) break\n\n      // Model A refines based on critique\n      const refinePrompt = `Original task: ${task}\\n\\nYour previous response:\\n${currentProposal}\\n\\nFeedback received:\\n${critiqueResult.response}\\n\\nPlease improve your response addressing the feedback.`\n      const refineResult = await this.querySingleModel(modelA, refinePrompt)\n      results.push(refineResult)\n\n      if (!refineResult.success) break\n\n      currentProposal = refineResult.response\n    }\n\n    return {\n      mode: 'debate',\n      models: config.models,\n      finalResponse: currentProposal,\n      individualResults: results,\n      metadata: {\n        totalLatencyMs: Date.now() - startTime,\n        totalTokensUsed: this.sumTokens(results),\n        iterations: Math.ceil(results.length / 2),\n      },\n    }\n  }\n\n  /**\n   * Consensus mode: Models vote and merge best ideas\n   */\n  private async executeConsensus(\n    task: string,\n    config: ModelCollaborationConfig,\n    startTime: number,\n  ): Promise<OrchestratedResult> {\n    // Get responses from all models\n    const results = await Promise.all(\n      config.models.map(model => this.querySingleModel(model, task)),\n    )\n\n    const successfulResults = results.filter(r => r.success)\n\n    if (successfulResults.length === 0) {\n      return {\n        mode: 'consensus',\n        models: config.models,\n        finalResponse: '',\n        individualResults: results,\n        metadata: {\n          totalLatencyMs: Date.now() - startTime,\n          consensusReached: false,\n        },\n      }\n    }\n\n    // Use a model to synthesize and find consensus\n    const synthesisModel = config.models[0]!\n    const responseSummary = successfulResults\n      .map((r, i) => `Response ${i + 1} (${r.model}):\\n${r.response}`)\n      .join('\\n\\n---\\n\\n')\n\n    const consensusPrompt = `You are tasked with finding the consensus among multiple AI responses.\n\nOriginal task: ${task}\n\nHere are the different responses:\n\n${responseSummary}\n\nPlease synthesize these responses into a single, comprehensive answer that:\n1. Identifies common themes and agreements\n2. Resolves any contradictions by choosing the most well-reasoned position\n3. Combines the best elements from each response\n4. Presents a unified, coherent answer\n\nSynthesized response:`\n\n    const consensusResult = await this.querySingleModel(\n      synthesisModel,\n      consensusPrompt,\n    )\n    results.push(consensusResult)\n\n    return {\n      mode: 'consensus',\n      models: config.models,\n      finalResponse: consensusResult.success\n        ? consensusResult.response\n        : successfulResults[0]!.response,\n      individualResults: results,\n      metadata: {\n        totalLatencyMs: Date.now() - startTime,\n        totalTokensUsed: this.sumTokens(results),\n        consensusReached: consensusResult.success,\n      },\n    }\n  }\n\n  /**\n   * Merge parallel results into a combined response\n   */\n  private mergeParallelResults(results: ModelQueryResult[]): string {\n    if (results.length === 0) return ''\n    if (results.length === 1) return results[0]!.response\n\n    // Format all responses for presentation\n    const formattedResponses = results\n      .map((r, i) => `## Perspective ${i + 1} (${r.model})\\n\\n${r.response}`)\n      .join('\\n\\n---\\n\\n')\n\n    return `# Multi-Model Analysis\n\nThe following perspectives were gathered from ${results.length} different models:\n\n${formattedResponses}\n\n---\n\n*Note: These are independent analyses. Consider synthesizing the best elements from each perspective.*`\n  }\n\n  /**\n   * Sum token usage across results\n   */\n  private sumTokens(results: ModelQueryResult[]): number | undefined {\n    const tokensUsed = results\n      .filter(r => r.tokenUsage)\n      .map(r => (r.tokenUsage?.input || 0) + (r.tokenUsage?.output || 0))\n\n    return tokensUsed.length > 0\n      ? tokensUsed.reduce((a, b) => a + b, 0)\n      : undefined\n  }\n}\n\n// Singleton instance\nlet orchestratorInstance: ModelOrchestrator | null = null\n\n/**\n * Get the model orchestrator instance\n */\nexport function getModelOrchestrator(): ModelOrchestrator {\n  if (!orchestratorInstance) {\n    orchestratorInstance = new ModelOrchestrator()\n  }\n  return orchestratorInstance\n}\n\n/**\n * Reset the orchestrator instance (useful for testing)\n */\nexport function resetModelOrchestrator(): void {\n  orchestratorInstance = null\n}\n\n/**\n * Execute a multi-model collaboration task\n */\nexport async function orchestrate(\n  task: string,\n  config: ModelCollaborationConfig,\n): Promise<OrchestratedResult> {\n  return getModelOrchestrator().execute(task, config)\n}\n"],
+  "mappings": "AAUA,SAAS,kBAAkB;AAC3B,SAAS,uBAAuB;AAChC,SAAS,cAAc,yBAAyB;AAmEhD,SAAS,uBAAuB,SAAmC;AACjE,QAAM,UAAU,QAAQ,QAAQ;AAChC,MAAI,OAAO,YAAY,UAAU;AAC/B,WAAO;AAAA,EACT;AACA,MAAI,MAAM,QAAQ,OAAO,GAAG;AAC1B,WAAO,QACJ;AAAA,MACC,CAAC,UACC,MAAM,SAAS;AAAA,IACnB,EACC,IAAI,WAAS,MAAM,IAAI,EACvB,KAAK,IAAI;AAAA,EACd;AACA,SAAO;AACT;AAKA,SAAS,kBAAkB,QAA6B;AACtD,SAAO;AAAA,IACL,SAAS,EAAE,MAAM,QAAQ,SAAS,OAAO;AAAA,IACzC,MAAM;AAAA,IACN,MAAM,WAAW;AAAA,EACnB;AACF;AAKO,MAAM,kBAAkB;AAAA,EACrB,eAAe,gBAAgB;AAAA;AAAA;AAAA;AAAA,EAKvC,MAAM,QACJ,MACA,QAC6B;AAC7B,UAAM,YAAY,KAAK,IAAI;AAE3B,YAAQ,OAAO,MAAM;AAAA,MACnB,KAAK;AACH,eAAO,KAAK,kBAAkB,MAAM,QAAQ,SAAS;AAAA,MACvD,KAAK;AACH,eAAO,KAAK,gBAAgB,MAAM,QAAQ,SAAS;AAAA,MACrD,KAAK;AACH,eAAO,KAAK,cAAc,MAAM,QAAQ,SAAS;AAAA,MACnD,KAAK;AACH,eAAO,KAAK,iBAAiB,MAAM,QAAQ,SAAS;AAAA,MACtD;AACE,cAAM,IAAI,MAAM,+BAA+B,OAAO,IAAI,EAAE;AAAA,IAChE;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,MAAc,iBACZ,WACA,QAC2B;AAC3B,UAAM,YAAY,KAAK,IAAI;AAE3B,QAAI;AACF,YAAM,UAAU,KAAK,aAAa,aAAa,SAAS;AACxD,UAAI,CAAC,SAAS;AACZ,eAAO;AAAA,UACL,OAAO;AAAA,UACP,UAAU;AAAA,UACV,SAAS;AAAA,UACT,OAAO,oBAAoB,SAAS;AAAA,UACpC,WAAW,KAAK,IAAI,IAAI;AAAA,QAC1B;AAAA,MACF;AAGA,YAAM,WAA+C;AAAA,QACnD,kBAAkB,MAAM;AAAA,MAC1B;AAIA,YAAM,WAAW,MAAM,kBAAkB,QAAQ,UAAU,CAAC,GAAG,MAAS;AACxE,YAAM,eAAe,uBAAuB,QAAQ;AAEpD,aAAO;AAAA,QACL,OAAO;AAAA,QACP,UAAU;AAAA,QACV,SAAS;AAAA,QACT,WAAW,KAAK,IAAI,IAAI;AAAA,MAC1B;AAAA,IACF,SAAS,OAAO;AACd,aAAO;AAAA,QACL,OAAO;AAAA,QACP,UAAU;AAAA,QACV,SAAS;AAAA,QACT,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;AAAA,QAC5D,WAAW,KAAK,IAAI,IAAI;AAAA,MAC1B;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,MAAc,kBACZ,MACA,QACA,WAC6B;AAC7B,UAAM,UAA8B,CAAC;AACrC,QAAI,gBAAgB;AACpB,QAAI,gBAAgB;AAEpB,eAAW,aAAa,OAAO,QAAQ;AACrC,YAAM,SAAS,MAAM,KAAK,iBAAiB,WAAW,aAAa;AACnE,cAAQ,KAAK,MAAM;AAEnB,UAAI,CAAC,OAAO,SAAS;AAEnB,YAAI,OAAO,UAAU,UAAU;AAC7B,gBAAM,iBAAiB,MAAM,KAAK;AAAA,YAChC,OAAO,SAAS;AAAA,YAChB;AAAA,UACF;AACA,kBAAQ,KAAK,cAAc;AAC3B,cAAI,eAAe,SAAS;AAC1B,4BAAgB,sBAAsB,eAAe,QAAQ;AAAA;AAAA;AAC7D,4BAAgB,eAAe;AAC/B;AAAA,UACF;AAAA,QACF;AACA;AAAA,MACF;AAGA,sBAAgB;AAAA,EAAuB,OAAO,QAAQ;AAAA;AAAA;AACtD,sBAAgB,OAAO;AAAA,IACzB;AAEA,WAAO;AAAA,MACL,MAAM;AAAA,MACN,QAAQ,OAAO;AAAA,MACf;AAAA,MACA,mBAAmB;AAAA,MACnB,UAAU;AAAA,QACR,gBAAgB,KAAK,IAAI,IAAI;AAAA,QAC7B,iBAAiB,KAAK,UAAU,OAAO;AAAA,QACvC,YAAY,QAAQ;AAAA,MACtB;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,MAAc,gBACZ,MACA,QACA,WAC6B;AAC7B,UAAM,UAAU,MAAM,QAAQ;AAAA,MAC5B,OAAO,OAAO,IAAI,WAAS,KAAK,iBAAiB,OAAO,IAAI,CAAC;AAAA,IAC/D;AAEA,UAAM,oBAAoB,QAAQ,OAAO,OAAK,EAAE,OAAO;AACvD,UAAM,gBAAgB,KAAK,qBAAqB,iBAAiB;AAEjE,WAAO;AAAA,MACL,MAAM;AAAA,MACN,QAAQ,OAAO;AAAA,MACf;AAAA,MACA,mBAAmB;AAAA,MACnB,UAAU;AAAA,QACR,gBAAgB,KAAK,IAAI,IAAI;AAAA,QAC7B,iBAAiB,KAAK,UAAU,OAAO;AAAA,MACzC;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,MAAc,cACZ,MACA,QACA,WAC6B;AAC7B,UAAM,gBAAgB,OAAO,UAAU,iBAAiB;AACxD,UAAM,CAAC,QAAQ,MAAM,IAAI,OAAO;AAEhC,QAAI,CAAC,UAAU,CAAC,QAAQ;AACtB,YAAM,IAAI,MAAM,wCAAwC;AAAA,IAC1D;AAEA,UAAM,UAA8B,CAAC;AACrC,QAAI,kBAAkB;AAGtB,UAAM,iBAAiB,MAAM,KAAK,iBAAiB,QAAQ,IAAI;AAC/D,YAAQ,KAAK,cAAc;AAE3B,QAAI,CAAC,eAAe,SAAS;AAC3B,aAAO;AAAA,QACL,MAAM;AAAA,QACN,QAAQ,OAAO;AAAA,QACf,eAAe;AAAA,QACf,mBAAmB;AAAA,QACnB,UAAU;AAAA,UACR,gBAAgB,KAAK,IAAI,IAAI;AAAA,UAC7B,YAAY;AAAA,QACd;AAAA,MACF;AAAA,IACF;AAEA,sBAAkB,eAAe;AAGjC,aAAS,IAAI,GAAG,IAAI,eAAe,KAAK;AAEtC,YAAM,iBAAiB;AAAA;AAAA,EAAuE,eAAe;AAAA;AAAA,iBAAsB,IAAI;AACvI,YAAM,iBAAiB,MAAM,KAAK,iBAAiB,QAAQ,cAAc;AACzE,cAAQ,KAAK,cAAc;AAE3B,UAAI,CAAC,eAAe,QAAS;AAG7B,YAAM,eAAe,kBAAkB,IAAI;AAAA;AAAA;AAAA,EAAgC,eAAe;AAAA;AAAA;AAAA,EAA2B,eAAe,QAAQ;AAAA;AAAA;AAC5I,YAAM,eAAe,MAAM,KAAK,iBAAiB,QAAQ,YAAY;AACrE,cAAQ,KAAK,YAAY;AAEzB,UAAI,CAAC,aAAa,QAAS;AAE3B,wBAAkB,aAAa;AAAA,IACjC;AAEA,WAAO;AAAA,MACL,MAAM;AAAA,MACN,QAAQ,OAAO;AAAA,MACf,eAAe;AAAA,MACf,mBAAmB;AAAA,MACnB,UAAU;AAAA,QACR,gBAAgB,KAAK,IAAI,IAAI;AAAA,QAC7B,iBAAiB,KAAK,UAAU,OAAO;AAAA,QACvC,YAAY,KAAK,KAAK,QAAQ,SAAS,CAAC;AAAA,MAC1C;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,MAAc,iBACZ,MACA,QACA,WAC6B;AAE7B,UAAM,UAAU,MAAM,QAAQ;AAAA,MAC5B,OAAO,OAAO,IAAI,WAAS,KAAK,iBAAiB,OAAO,IAAI,CAAC;AAAA,IAC/D;AAEA,UAAM,oBAAoB,QAAQ,OAAO,OAAK,EAAE,OAAO;AAEvD,QAAI,kBAAkB,WAAW,GAAG;AAClC,aAAO;AAAA,QACL,MAAM;AAAA,QACN,QAAQ,OAAO;AAAA,QACf,eAAe;AAAA,QACf,mBAAmB;AAAA,QACnB,UAAU;AAAA,UACR,gBAAgB,KAAK,IAAI,IAAI;AAAA,UAC7B,kBAAkB;AAAA,QACpB;AAAA,MACF;AAAA,IACF;AAGA,UAAM,iBAAiB,OAAO,OAAO,CAAC;AACtC,UAAM,kBAAkB,kBACrB,IAAI,CAAC,GAAG,MAAM,YAAY,IAAI,CAAC,KAAK,EAAE,KAAK;AAAA,EAAO,EAAE,QAAQ,EAAE,EAC9D,KAAK,aAAa;AAErB,UAAM,kBAAkB;AAAA;AAAA,iBAEX,IAAI;AAAA;AAAA;AAAA;AAAA,EAInB,eAAe;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAUb,UAAM,kBAAkB,MAAM,KAAK;AAAA,MACjC;AAAA,MACA;AAAA,IACF;AACA,YAAQ,KAAK,eAAe;AAE5B,WAAO;AAAA,MACL,MAAM;AAAA,MACN,QAAQ,OAAO;AAAA,MACf,eAAe,gBAAgB,UAC3B,gBAAgB,WAChB,kBAAkB,CAAC,EAAG;AAAA,MAC1B,mBAAmB;AAAA,MACnB,UAAU;AAAA,QACR,gBAAgB,KAAK,IAAI,IAAI;AAAA,QAC7B,iBAAiB,KAAK,UAAU,OAAO;AAAA,QACvC,kBAAkB,gBAAgB;AAAA,MACpC;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKQ,qBAAqB,SAAqC;AAChE,QAAI,QAAQ,WAAW,EAAG,QAAO;AACjC,QAAI,QAAQ,WAAW,EAAG,QAAO,QAAQ,CAAC,EAAG;AAG7C,UAAM,qBAAqB,QACxB,IAAI,CAAC,GAAG,MAAM,kBAAkB,IAAI,CAAC,KAAK,EAAE,KAAK;AAAA;AAAA,EAAQ,EAAE,QAAQ,EAAE,EACrE,KAAK,aAAa;AAErB,WAAO;AAAA;AAAA,gDAEqC,QAAQ,MAAM;AAAA;AAAA,EAE5D,kBAAkB;AAAA;AAAA;AAAA;AAAA;AAAA,EAKlB;AAAA;AAAA;AAAA;AAAA,EAKQ,UAAU,SAAiD;AACjE,UAAM,aAAa,QAChB,OAAO,OAAK,EAAE,UAAU,EACxB,IAAI,QAAM,EAAE,YAAY,SAAS,MAAM,EAAE,YAAY,UAAU,EAAE;AAEpE,WAAO,WAAW,SAAS,IACvB,WAAW,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IACpC;AAAA,EACN;AACF;AAGA,IAAI,uBAAiD;AAK9C,SAAS,uBAA0C;AACxD,MAAI,CAAC,sBAAsB;AACzB,2BAAuB,IAAI,kBAAkB;AAAA,EAC/C;AACA,SAAO;AACT;AAKO,SAAS,yBAA+B;AAC7C,yBAAuB;AACzB;AAKA,eAAsB,YACpB,MACA,QAC6B;AAC7B,SAAO,qBAAqB,EAAE,QAAQ,MAAM,MAAM;AACpD;",
+  "names": []
+}

package/dist/services/openai.js

@@ -8,6 +8,9 @@ import {
 } from "../utils/debugLogger.js";
 import { logError } from "../utils/log.js";
 import { abortableDelay, DEFAULT_RETRY_CONFIG } from "../utils/async.js";
+import {
+  UserFriendlyError
+} from "../utils/userFriendlyError.js";
 const RETRY_CONFIG = {
   ...DEFAULT_RETRY_CONFIG
 };
@@ -638,7 +641,11 @@ async function getCompletionWithProfile(modelProfile, opts, attempt = 0, maxAtte
         // 🔧 Pass AbortSignal to recursive call
       );
     }
-    throw error;
+    throw new UserFriendlyError(error, {
+      operation: "API request",
+      model: opts.model,
+      endpoint: modelProfile.baseURL
+    });
   }
 }
 function createStreamProcessor(stream, signal) {