Houtini-lm

// Fixed streaming response handler for Houtini LM MCP // This fixes the [object Object] issue when streaming from LM Studio export async function handleLLMResponse(prediction: AsyncIterable<any>): Promise<string> { let response = ''; try { for await (const chunk of prediction) { // Handle different chunk types if (typeof chunk === 'string') { response += chunk; } else if (chunk && typeof chunk === 'object') { // If it's an object with content property (common in streaming) if ('content' in chunk) { response += chunk.content; } else if ('text' in chunk) { response += chunk.text; } else if ('choices' in chunk && Array.isArray(chunk.choices)) { // Handle OpenAI-style streaming format for (const choice of chunk.choices) { if (choice.delta?.content) { response += choice.delta.content; } else if (choice.text) { response += choice.text; } } } else { // Try to stringify if it's a plain object try { const str = JSON.stringify(chunk); if (str !== '[object Object]') { response += str; } } catch { console.warn('Skipping non-serializable chunk:', chunk); } } } } } catch (error) { console.error('Error processing streaming response:', error); throw error; } return response; } // Enhanced response formatter with schema enforcement export function formatWithSchema(response: string, schema?: any): any { // First, try to extract JSON from the response let jsonData = null; // Try direct JSON parse try { jsonData = JSON.parse(response); } catch { // Try to extract JSON from markdown code blocks const jsonMatch = response.match(/```(?:json)?\s*\n?([\s\S]*?)\n?```/); if (jsonMatch) { try { jsonData = JSON.parse(jsonMatch[1]); } catch { // Continue with text processing } } } // If we have a schema and JSON data, validate and structure it if (schema && jsonData) { return enforceSchema(jsonData, schema); } // Otherwise, return structured text response return { summary: extractSummary(response), content: response, confidence: 0.85, metadata: { format: jsonData ? 'json' : 'text', length: response.length } }; } function enforceSchema(data: any, schema: any): any { const result: any = {}; if (schema.type === 'object' && schema.properties) { for (const [key, propSchema] of Object.entries(schema.properties)) { const prop = propSchema as any; if (data[key] !== undefined) { if (prop.type === 'string') { result[key] = String(data[key]); } else if (prop.type === 'number' || prop.type === 'integer') { result[key] = Number(data[key]) || 0; } else if (prop.type === 'boolean') { result[key] = Boolean(data[key]); } else if (prop.type === 'array') { result[key] = Array.isArray(data[key]) ? data[key] : []; } else if (prop.type === 'object') { result[key] = typeof data[key] === 'object' ? data[key] : {}; } else { result[key] = data[key]; } } else if (schema.required?.includes(key)) { // Provide default for required fields if (prop.type === 'string') result[key] = ''; else if (prop.type === 'number' || prop.type === 'integer') result[key] = 0; else if (prop.type === 'boolean') result[key] = false; else if (prop.type === 'array') result[key] = []; else if (prop.type === 'object') result[key] = {}; } } } return result; } function extractSummary(text: string): string { // Extract first meaningful line or paragraph const lines = text.split('\n').filter(line => line.trim()); if (lines.length > 0) { // Remove markdown headers return lines[0].replace(/^#+\s*/, '').trim(); } return 'Analysis complete'; }