From d1210773e74d7fdefc64b50b542345e1d2932933 Mon Sep 17 00:00:00 2001
From: drjones <drjones@thetempleofdoom.com>
Date: Thu, 13 Aug 2026 23:45:02 +0000
Subject: [PATCH] Switch AI tool generator from Gemini to local Ollama
 (gemma4:26b @ shadow-death .128)

---
 src/server/aiGenerator.ts | 51 +++++++++++++++++++++------------------
 1 file changed, 28 insertions(+), 23 deletions(-)

diff --git a/src/server/aiGenerator.ts b/src/server/aiGenerator.ts
index f45b0b3..6ea6750 100644
--- a/src/server/aiGenerator.ts
+++ b/src/server/aiGenerator.ts
@@ -1,16 +1,31 @@
-import { GoogleGenAI } from '@google/genai';
 import { MCPServerDefinition, ToolDefinition } from '../types.js';
 
-let aiClient: GoogleGenAI | null = null;
+const OLLAMA_HOST = process.env.OLLAMA_HOST || 'http://10.30.20.128:11434';
+const OLLAMA_MODEL = process.env.OLLAMA_MODEL || 'gemma4:26b';
 
-function getAIClient(): GoogleGenAI | null {
-  if (!aiClient) {
-    const key = process.env.GEMINI_API_KEY;
-    if (key && key !== 'MY_GEMINI_API_KEY') {
-      aiClient = new GoogleGenAI({ apiKey: key });
-    }
+async function callLocalModel(prompt: string): Promise<string> {
+  const res = await fetch(`${OLLAMA_HOST}/api/chat`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({
+      model: OLLAMA_MODEL,
+      messages: [{ role: 'user', content: prompt }],
+      stream: false,
+      format: 'json',
+      keep_alive: '30m',
+      options: { num_predict: 4096, temperature: 0.2 },
+    }),
+    signal: AbortSignal.timeout(180000),
+  });
+  if (!res.ok) {
+    throw new Error(`Ollama HTTP ${res.status} (${res.statusText})`);
   }
-  return aiClient;
+  const data = await res.json();
+  let content = data?.message?.content || '';
+  if (!content) content = data?.message?.thinking || '';
+  content = content.trim();
+  if (!content) throw new Error('Empty response from local model');
+  return content;
 }
 
 export interface GeneratedMCPResponse {
@@ -36,17 +51,8 @@ export interface GeneratedMCPResponse {
 }
 
 export async function generateMCPToolFromPrompt(prompt: string): Promise<GeneratedMCPResponse> {
-  const ai = getAIClient();
-  
-  if (!ai) {
-    // Fallback template generator when GEMINI_API_KEY is not configured
-    return generateFallbackMCP(prompt);
-  }
-
   try {
-    const response = await ai.models.generateContent({
-      model: 'gemini-2.5-flash',
-      contents: `You are an expert Model Context Protocol (MCP) tool builder.
+    const systemPrompt = `You are an expert Model Context Protocol (MCP) tool builder.
 Given the following user request or API prompt, construct a complete MCP Tool configuration in strict JSON.
 
 User Request: "${prompt}"
@@ -75,15 +81,14 @@ Respond ONLY with valid JSON (no markdown ticks or commentary) matching this Typ
   "sampleArgs": { "param1": "sample_value" }
 }
 
-Provide executable JavaScript code for customScript or accurate URL template for webhookConfig. Use fetch() if HTTP calls are needed inside customScript.`,
-    });
+Provide executable JavaScript code for customScript or accurate URL template for webhookConfig. Use fetch() if HTTP calls are needed inside customScript.`;
 
-    const rawText = response.text || '';
+    const rawText = await callLocalModel(systemPrompt);
     const cleanJson = rawText.replace(/```json/g, '').replace(/```/g, '').trim();
     const parsed = JSON.parse(cleanJson);
     return parsed;
   } catch (err) {
-    console.warn('Gemini API call failed, using heuristic fallback:', err);
+    console.warn('Local model generation failed, using heuristic fallback:', err);
     return generateFallbackMCP(prompt);
   }
 }