feat: use llama

.env.example

@@ -1,2 +1,3 @@
 DATABASE_URL=""
-USER_KEY=""
+USER_KEY=""
+OVHCLOUD_API_KEY=""

package.json

@@ -1,7 +1,7 @@
 {
   "name": "siri-shortcuts",
   "version": "1.0.0",
-  "description": "Siri-enabled, Anthropic-powered shortcuts",
+  "description": "Siri-enabled, Llama-powered shortcuts",
   "author": "riccardo@frompixels.com",
   "scripts": {
     "dev": "next dev",
@@ -25,12 +25,12 @@
   },
   "dependencies": {
     "@prisma/client": "^5.22.0",
-    "ai": "^5.0.68",
+    "openai": "^4.77.0",
     "axios": "^1.12.0",
     "next": "^15.5.9",
     "react": "^19.0.0",
     "react-dom": "^19.0.0",
-    "zod": "^4.1.12"
+    "zod": "^3.23.8"
   },
   "devDependencies": {
     "@commitlint/cli": "^19.6.1",

prisma/schema.prisma

@@ -7,7 +7,7 @@ datasource db {
   url      = env("DATABASE_URL")
 }
 
-model AnthropicQuery {
+model AiQuery {
   id          String   @id @default(cuid())
   question    String   @db.Text
   response    String   @db.Text
@@ -17,7 +17,7 @@ model AnthropicQuery {
   createdAt   DateTime @default(now())
   updatedAt   DateTime @updatedAt
   
-  @@map("anthropic_queries")
+  @@map("ai_queries")
   @@index([createdAt])
   @@index([success])
 }

test-request.js

@@ -4,7 +4,7 @@ const fetch = require('axios');
 async function testAPI() {
   try {
     const response = await fetch.post('http://localhost:3000/api/shortcut', {
-      command: 'anthropic',
+      command: 'llama',
       parameters: {
         question: 'What is 42?'
       },

utils/aiGatewayClient.ts

@@ -1,54 +1,90 @@
-import { generateText } from 'ai';
+import OpenAI from 'openai';
 
-interface AiGatewayResponse {
+let ovhAI: OpenAI | null = null;
+
+function getClient(): OpenAI {
+  if (!ovhAI) {
+    ovhAI = new OpenAI({
+      apiKey: process.env.OVHCLOUD_API_KEY,
+      baseURL: 'https://oai.endpoints.kepler.ai.cloud.ovh.net/v1'
+    });
+  }
+  return ovhAI;
+}
+
+interface AiResponse {
   text: string;
   tokensUsed: number;
 }
 
-export async function queryAiGateway(
+const MAX_RETRIES = 3;
-  text: string,
+
-  model: string
+export async function queryAi(text: string): Promise<AiResponse> {
-): Promise<AiGatewayResponse> {
   const requestId = Math.random().toString(36).substring(7);
   const startTime = Date.now();
 
-  console.info(`[AI-${requestId}] Starting Vercel Gateway AI request`, {
+  console.info(`[AI-${requestId}] Starting OVH AI request`, {
     promptLength: text.length,
-    model,
     timestamp: new Date().toISOString()
   });
 
-  try {
+  let lastError: Error | null = null;
-    const response = await generateText({
-      model,
-      prompt: text
-    });
 
-    const duration = Date.now() - startTime;
+  for (let attempt = 1; attempt <= MAX_RETRIES; attempt++) {
-    const tokensUsed = response.usage?.totalTokens || 0;
+    try {
+      console.info(
+        `[AI-${requestId}] OVH AI request attempt ${attempt}/${MAX_RETRIES}`
+      );
 
-    console.info(
+      const completion = await getClient().chat.completions.create({
-      `[AI-${requestId}] Vercel Gateway AI response received in ${duration}ms`,
+        model: 'Meta-Llama-3_3-70B-Instruct',
-      {
+        temperature: 0.7,
-        responseLength: response.text.length,
+        max_tokens: 4096,
-        tokensUsed,
+        messages: [
-        usage: response.usage
+          {
+            role: 'user',
+            content: text
+          }
+        ]
+      });
+
+      const responseText = completion.choices[0]?.message?.content || '';
+      const tokensUsed =
+        (completion.usage?.prompt_tokens || 0) +
+        (completion.usage?.completion_tokens || 0);
+
+      const duration = Date.now() - startTime;
+      console.info(
+        `[AI-${requestId}] OVH AI response received in ${duration}ms`,
+        {
+          responseLength: responseText.length,
+          tokensUsed,
+          usage: completion.usage
+        }
+      );
+
+      return {
+        text: responseText,
+        tokensUsed
+      };
+    } catch (error) {
+      const duration = Date.now() - startTime;
+      console.error(
+        `[AI-${requestId}] OVH AI attempt ${attempt} failed after ${duration}ms:`,
+        {
+          error: error instanceof Error ? error.message : String(error),
+          promptLength: text.length
+        }
+      );
+      lastError = error as Error;
+
+      if (attempt < MAX_RETRIES) {
+        const delay = 1000 * attempt;
+        console.info(`[AI-${requestId}] Retrying in ${delay}ms...`);
+        await new Promise(resolve => setTimeout(resolve, delay));
       }
-    );
+    }
-
-    return {
-      text: response.text,
-      tokensUsed
-    };
-  } catch (error) {
-    const duration = Date.now() - startTime;
-    console.error(
-      `[AI-${requestId}] Vercel Gateway AI error after ${duration}ms:`,
-      {
-        error: error instanceof Error ? error.message : String(error),
-        promptLength: text.length
-      }
-    );
-    throw new Error(`Vercel Gateway AI error: ${JSON.stringify(error)}.`);
   }
+
+  throw lastError || new Error('OVH AI error: all retry attempts failed.');
 }

utils/commands/llama.ts

@@ -1,8 +1,8 @@
-import { queryAiGateway } from '@utils/aiGatewayClient';
+import { queryAi } from '@utils/aiGatewayClient';
 import { ShortcutsResponse } from '../types';
 import { dbOperations } from '@utils/db';
 
-export async function anthropicCommand(
+export async function llamaCommand(
   parameters: Record<string, string> | undefined
 ): Promise<ShortcutsResponse> {
   const commandId = Math.random().toString(36).substring(7);
@@ -14,7 +14,7 @@ export async function anthropicCommand(
   let errorMessage: string | undefined;
   let tokensUsed: number | undefined;
 
-  console.info(`[CMD-${commandId}] Anthropic command started`, {
+  console.info(`[CMD-${commandId}] Llama command started`, {
     hasParameters: !!parameters,
     timestamp: new Date().toISOString()
   });
@@ -41,17 +41,14 @@ export async function anthropicCommand(
       question +
       '. Structure the response in a manner suitable for spoken communication.';
 
-    const anthropicResponse = await queryAiGateway(
+    const aiResponse = await queryAi(prompt);
-      prompt,
+    response = aiResponse.text;
-      'anthropic/claude-sonnet-4.5'
+    tokensUsed = aiResponse.tokensUsed;
-    );
-    response = anthropicResponse.text;
-    tokensUsed = anthropicResponse.tokensUsed;
     success = true;
 
     const duration = Date.now() - startTime;
     console.info(
-      `[CMD-${commandId}] Anthropic command completed in ${duration}ms`,
+      `[CMD-${commandId}] Llama command completed in ${duration}ms`,
       {
         responseLength: response.length,
         tokensUsed,
@@ -69,12 +66,12 @@ export async function anthropicCommand(
   } catch (error) {
     const duration = Date.now() - startTime;
     console.error(
-      `[CMD-${commandId}] Anthropic command failed after ${duration}ms:`,
+      `[CMD-${commandId}] Llama command failed after ${duration}ms:`,
       error
     );
     success = false;
     errorMessage = error instanceof Error ? error.message : 'Unknown error';
-    response = 'Sorry. There was a problem with Anthropic.';
+    response = 'Sorry. There was a problem with the AI service.';
 
     return {
       success: false,
@@ -84,7 +81,7 @@ export async function anthropicCommand(
     if (question) {
       try {
         console.info(`[CMD-${commandId}] Saving query to database`);
-        await dbOperations.saveAnthropicQuery({
+        await dbOperations.saveQuery({
           question,
           response,
           success,

utils/db.ts

@@ -11,7 +11,7 @@ if (process.env.NODE_ENV === 'development') {
 }
 
 export const dbOperations = {
-  async saveAnthropicQuery({
+  async saveQuery({
     question,
     response,
     success,
@@ -25,7 +25,7 @@ export const dbOperations = {
     tokensUsed?: number;
   }) {
     try {
-      return await db.anthropicQuery.create({
+      return await db.aiQuery.create({
         data: {
           question,
           response,

utils/registry.ts

@@ -1,7 +1,7 @@
 import { ShortcutsResponse } from './types';
 import { pingCommand } from './commands/ping';
 import { timeCommand } from './commands/time';
-import { anthropicCommand } from './commands/anthropic';
+import { llamaCommand } from './commands/llama';
 
 type CommandHandler = (
   parameters?: Record<string, string>
@@ -18,7 +18,7 @@ export class CommandRegistry {
   private registerDefaultCommands() {
     this.register('ping', pingCommand);
     this.register('time', timeCommand);
-    this.register('anthropic', anthropicCommand);
+    this.register('llama', llamaCommand);
   }
 
   register(command: string, handler: CommandHandler) {