Ambiance MCP Server

/** * @fileOverview: Token Parameter Utilities for AI Tools * @module: TokenUtils * @keyFunctions: * - determineTokenParameter: Choose between max_tokens (legacy), max_completion_tokens (modern), and max_output_tokens (reasoning) * - buildApiRequest: Build API request with appropriate token parameter * @context: Handles compatibility between deprecated max_tokens, new max_completion_tokens, and reasoning-aware max_output_tokens */ import { logger } from '../../../utils/logger'; /** * Determine which token parameter to use based on the model */ export function determineTokenParameter( model: string ): 'max_tokens' | 'max_completion_tokens' | 'max_output_tokens' { const normalizedModel = model.toLowerCase(); if (normalizedModel.startsWith('gpt-5')) { return 'max_output_tokens'; } // Legacy models that still use max_tokens const isLegacyModel = normalizedModel.includes('gpt-4.1') || normalizedModel.includes('gpt-4o') || normalizedModel.includes('gpt-4o-mini') || normalizedModel.includes('gpt-3.5') || normalizedModel.includes('text-') || normalizedModel.includes('davinci') || normalizedModel.includes('curie') || normalizedModel.includes('babbage') || normalizedModel.includes('ada') || normalizedModel.includes('claude-') || // Anthropic models normalizedModel.includes('gemini-') || // Google models normalizedModel.includes('command-') || // Cohere models normalizedModel.includes('qwen-'); // Qwen models return isLegacyModel ? 'max_tokens' : 'max_completion_tokens'; } /** * Build API request with appropriate token parameter */ export function buildApiRequest( model: string, messages: any[], maxTokens: number, temperature: number = 0.3, additionalParams: Record<string, any> = {} ): any { const tokenParam = determineTokenParameter(model); logger.info('🔍 Token parameter selection', { model, isLegacyModel: tokenParam === 'max_tokens', usingParameter: tokenParam, tokenValue: maxTokens, }); const apiRequest: any = { model, messages, temperature, ...additionalParams, }; // Add the appropriate token parameter apiRequest[tokenParam] = maxTokens; if (tokenParam === 'max_output_tokens' && 'temperature' in apiRequest) { logger.info('ℹ️ Reasoning models ignore temperature; relying on reasoning.effort instead', { model, }); delete apiRequest.temperature; } return apiRequest; }