feat: Switch to Groq as primary LLM provider

GRADIO_APP.md

@@ -58,7 +58,7 @@ Subtle translucency (`backdrop-filter: blur(12px)`) creates a sense of depth. Th
 
 ### 🎧 The Narrator (Sonic Mode)
 - **Design**: Modeled after premium music players
-- **Tech**: Gemini 2.0 Flash summarization + ElevenLabs TTS
+- **Tech**: Groq Llama 3.1 summarization + ElevenLabs TTS
 - **Feature**: Natural sentence endings for clean audio
 
 ### 🧠 The Analyst (Intelligence)

MCP_SERVER.md

@@ -87,6 +87,7 @@ Add to your `claude_desktop_config.json`:
       "command": "python",
       "args": ["c:/Dev/Medium-Agent/medium-mcp/server.py"],
       "env": {
+        "GROQ_API_KEY": "your-key",
         "GEMINI_API_KEY": "your-key",
         "ELEVENLABS_API_KEY": "your-key",
         "OPENAI_API_KEY": "your-key"

README.md

@@ -52,7 +52,7 @@ By combining **Agentic Workflows**, **Neural Audio**, and the **Model Context Pr
 
 ### 🎨 UI Enhancements
 - **🏠 Hero Tab**: Beautiful landing page with feature overview and gradient design
-- **🎧 Improved Audio**: Gemini 2.0 Flash for summarization with natural sentence endings
+- **🎧 Improved Audio**: Groq Llama for summarization with natural sentence endings
 - **📊 Enhanced Intelligence**: Streamlined analyst reports with PDF export
 
 ---
@@ -101,7 +101,7 @@ The system operates as an autonomous research pipeline:
 1. **User** asks a question via the **Project Aether UI**
 2. **Scout Agent** searches DuckDuckGo & RSS for fresh signals
 3. **Reader Agent** extracts clean text, bypassing paywalls
-4. **Analyst Agent** (Gemini 2.0 Flash) synthesizes professional reports
+4. **Analyst Agent** (Groq Llama 3.3) synthesizes professional reports
 5. **Sonic Agent** (ElevenLabs) converts content to podcast audio
 
 ---
@@ -112,7 +112,8 @@ The system operates as an autonomous research pipeline:
 | :--- | :--- |
 | **[ElevenLabs](https://elevenlabs.io/)** | Neural TTS for Sonic Mode |
 | **[OpenAI](https://openai.com/)** | TTS fallback & embeddings |
-| **[Google Gemini 2.0](https://deepmind.google/technologies/gemini/)** | Intelligence engine for Analyst |
+| **[Groq](https://console.groq.com/)** | Primary LLM (fastest inference) |
+| **[Google Gemini](https://deepmind.google/technologies/gemini/)** | Backup LLM + Vision/Embeddings |
 | **[Gradio](https://gradio.app/)** | Project Aether web interface |
 | **[Playwright](https://playwright.dev/)** | Headless browser scraping |
 
@@ -172,7 +173,8 @@ python app.py
 
 Built with ❤️ for the **Open Source AI Community**.
 
-*   **Google DeepMind**: Gemini 2.0 Flash
+*   **Groq**: Lightning-fast LLM inference
+*   **Google DeepMind**: Gemini for Vision/Embeddings
 *   **Anthropic**: MCP Standard
 *   **Hugging Face**: Platform & Infrastructure
 *   **ElevenLabs**: Neural Voice Technology

app.py

@@ -56,8 +56,10 @@ from src.service import ScraperService
 from src.html_renderer import render_full_page, BASE_TEMPLATE as RENDERER_TEMPLATE
 from src.config import MCPConfig
 from elevenlabs_voices import ELEVENLABS_VOICES, VOICE_CATEGORIES, get_voice_id
-# Import Gemini for Analyst
+# Import Gemini for Analyst (backup)
 import google.generativeai as genai
+# Import Groq for primary LLM
+from groq import Groq
 
 # ============================================================================
 # PROJECT AETHER: VISUAL SYSTEM (ENHANCED)
@@ -976,21 +978,14 @@ async def generate_audio(url, voice, summarize, max_chars):
         if not text or len(text) < 50:
             return '<div style="background: rgba(239,68,68,0.1); border: 1px solid rgba(239,68,68,0.2); border-radius: 12px; padding: 16px; text-align: center; margin-top: 16px;"><span style="color: #fca5a5; font-weight: 600; font-size: 14px;">⚠️ Article too short</span></div>', None
         
-        # Summarize with heavily guardrailed prompt
+        # Summarize with Groq (PRIMARY) or Gemini (BACKUP)
         if summarize != "none":
+            groq_key = os.environ.get("GROQ_API_KEY")
             gemini_key = os.environ.get("GEMINI_API_KEY")
-            if gemini_key:
-                try:
-                    gr.Info("Summarizing for audio...")
-                    genai.configure(api_key=gemini_key)
-                    # Use gemini-2.0-flash as primary, fallback to 1.5 if needed
-                    try:
-                        model = genai.GenerativeModel('gemini-2.0-flash')
-                    except:
-                        model = genai.GenerativeModel('gemini-1.5-flash-latest')
-                    
-                    # HEAVILY GUARDRAILED PROMPT
-                    prompt = f"""You are a professional podcast narrator. Your ONLY task is to summarize the following article for audio narration.
+            summarize_success = False
+            
+            # HEAVILY GUARDRAILED PROMPT (shared between providers)
+            prompt = f"""You are a professional podcast narrator. Your ONLY task is to summarize the following article for audio narration.
 
 STRICT RULES:
 1. Output ONLY plain English text suitable for text-to-speech
@@ -1010,26 +1005,53 @@ Article Content:
 
 Write a clean, narration-ready summary that ends with a proper concluding sentence:"""
 
+            # Try Groq first (PRIMARY - fastest)
+            if groq_key and not summarize_success:
+                try:
+                    gr.Info("Summarizing for audio with Groq...")
+                    client = Groq(api_key=groq_key)
+                    response = client.chat.completions.create(
+                        model="llama-3.1-8b-instant",  # Fast model for summarization
+                        messages=[{"role": "user", "content": prompt}],
+                        max_tokens=500,
+                        temperature=0.7
+                    )
+                    summary = response.choices[0].message.content.strip()
+                    
+                    # Validate response
+                    if summary and len(summary) > 50:
+                        summary = clean_text_for_audio(summary)
+                        if len(summary) > 50 and not any(bad in summary.lower() for bad in ['```', 'http://', 'https://', '**', '##']):
+                            text = summary
+                            summarize_success = True
+                except Exception as e:
+                    gr.Warning(f"Groq failed: {str(e)[:50]}, trying Gemini...")
+            
+            # Fallback to Gemini (BACKUP)
+            if gemini_key and not summarize_success:
+                try:
+                    gr.Info("Summarizing for audio with Gemini...")
+                    genai.configure(api_key=gemini_key)
+                    try:
+                        model = genai.GenerativeModel('gemini-2.0-flash')
+                    except:
+                        model = genai.GenerativeModel('gemini-1.5-flash-latest')
+                    
                     response = await model.generate_content_async(prompt)
                     summary = response.text.strip()
                     
-                    # Validate response - reject if it contains gibberish markers
                     if summary and len(summary) > 50:
-                        # Additional cleaning of Gemini output
                         summary = clean_text_for_audio(summary)
-                        
-                        # Reject if too short after cleaning or contains obvious issues
                         if len(summary) > 50 and not any(bad in summary.lower() for bad in ['```', 'http://', 'https://', '**', '##']):
                             text = summary
+                            summarize_success = True
                         else:
                             gr.Warning("Summary had issues, using cleaned original")
-                            text = text[:max_chars]
-                    else:
-                        text = text[:max_chars]
                 except Exception as e:
-                    gr.Warning(f"Summarization failed: {str(e)[:50]}")
-                    text = text[:max_chars]
-            else:
+                    gr.Warning(f"Gemini also failed: {str(e)[:50]}")
+            
+            # Final fallback: truncate original
+            if not summarize_success:
                 text = text[:max_chars]
         
         # Final safety check - ensure text is clean for TTS
@@ -1092,11 +1114,12 @@ async def analyst_report(topic):
     if not topic:
         return "Please enter a topic."
     
+    groq_key = os.environ.get("GROQ_API_KEY")
     gemini_key = os.environ.get("GEMINI_API_KEY")
     openai_key = os.environ.get("OPENAI_API_KEY")
     
-    if not gemini_key and not openai_key:
-        return "⚠️ Error: No AI API keys found. Set GEMINI_API_KEY or OPENAI_API_KEY in your .env file."
+    if not groq_key and not gemini_key and not openai_key:
+        return "⚠️ Error: No AI API keys found. Set GROQ_API_KEY, GEMINI_API_KEY, or OPENAI_API_KEY in your .env file."
     
     max_articles = 5
     
@@ -1152,11 +1175,25 @@ Articles:
         gr.Info("Analyst: Synthesizing report...")
         report_content = ""
         
-        # 4. Try Gemini first
-        if gemini_key:
+        # 4. Try Groq first (PRIMARY - fastest)
+        if groq_key:
+            try:
+                client = Groq(api_key=groq_key)
+                response = client.chat.completions.create(
+                    model="llama-3.3-70b-versatile",  # Best model for synthesis
+                    messages=[{"role": "user", "content": prompt}],
+                    max_tokens=2000,
+                    temperature=0.7
+                )
+                report_content = response.choices[0].message.content
+                gr.Info("Report generated via Groq")
+            except Exception as e:
+                gr.Warning(f"Groq failed: {str(e)[:100]}, trying Gemini...")
+        
+        # 5. Fallback to Gemini
+        if not report_content and gemini_key:
             try:
                 genai.configure(api_key=gemini_key)
-                # Use gemini-2.0-flash as primary, fallback to 1.5 if needed
                 try:
                     model = genai.GenerativeModel('gemini-2.0-flash')
                 except:
@@ -1167,7 +1204,7 @@ Articles:
             except Exception as e:
                 gr.Warning(f"Gemini failed: {str(e)[:100]}, trying OpenAI...")
         
-        # 5. Fallback to OpenAI
+        # 6. Fallback to OpenAI
         if not report_content and openai_key:
             try:
                 from openai import AsyncOpenAI
@@ -1286,9 +1323,10 @@ async def export_report_pdf():
 
 def render_settings():
     keys = {
+        "Groq": "GROQ_API_KEY",           # PRIMARY LLM
         "ElevenLabs": "ELEVENLABS_API_KEY",
-        "Gemini": "GEMINI_API_KEY",
-        "OpenAI": "OPENAI_API_KEY"
+        "Gemini": "GEMINI_API_KEY",       # BACKUP LLM
+        "OpenAI": "OPENAI_API_KEY"        # BACKUP LLM
     }
     
     html = "<h3>System Status</h3>"
@@ -1382,7 +1420,7 @@ with gr.Blocks(title="Project Aether") as demo:
                             <div style="width: 44px; height: 44px; background: linear-gradient(135deg, #10b981, #34d399); border-radius: 12px; display: flex; align-items: center; justify-content: center; font-size: 22px;">🧠</div>
                             <h3 style="margin: 0; color: #fff; font-size: 1.1rem; font-weight: 600;">AI Analyst</h3>
                         </div>
-                        <p style="color: #a1a1aa; font-size: 0.9rem; margin: 0; line-height: 1.6;">Generate comprehensive intelligence reports. AI-powered synthesis using Gemini & GPT-4.</p>
+                        <p style="color: #a1a1aa; font-size: 0.9rem; margin: 0; line-height: 1.6;">Generate comprehensive intelligence reports. AI-powered synthesis using Groq & GPT-4.</p>
                     </div>
                     
                     <!-- Settings Card -->
@@ -1391,7 +1429,7 @@ with gr.Blocks(title="Project Aether") as demo:
                             <div style="width: 44px; height: 44px; background: linear-gradient(135deg, #6366f1, #818cf8); border-radius: 12px; display: flex; align-items: center; justify-content: center; font-size: 22px;">⚙️</div>
                             <h3 style="margin: 0; color: #fff; font-size: 1.1rem; font-weight: 600;">Configuration</h3>
                         </div>
-                        <p style="color: #a1a1aa; font-size: 0.9rem; margin: 0; line-height: 1.6;">Manage API keys and system status. Connect ElevenLabs, Gemini, and OpenAI.</p>
+                        <p style="color: #a1a1aa; font-size: 0.9rem; margin: 0; line-height: 1.6;">Manage API keys and system status. Connect ElevenLabs, Groq, and OpenAI.</p>
                     </div>
                     
                 </div>

requirements.txt

@@ -32,6 +32,7 @@ fastmcp>=0.2.0
 # ============================================================================
 google-generativeai>=0.3.0
 openai>=1.3.0
+groq>=0.13.0
 
 # ============================================================================
 # Text-to-Speech

server.py

@@ -28,6 +28,9 @@ from elevenlabs_voices import ELEVENLABS_VOICES, get_voice_id, get_voices_info,
 from src.service import ScraperService
 from src.html_renderer import render_article_html, render_full_page
 
+# LLM imports
+from groq import Groq
+
 
 # ============================================================================
 # LIFESPAN MANAGEMENT
@@ -234,7 +237,7 @@ async def get_server_stats() -> str:
             "35+ total domains"
         ],
         "tts_providers": ["elevenlabs", "edge-tts", "openai"],
-        "ai_providers": ["gemini", "openai"]
+        "ai_providers": ["groq", "gemini", "openai"]
     }, ensure_ascii=False)
 
 
@@ -545,14 +548,11 @@ async def medium_cast(
     )
     
     if should_summarize and summarize != "none":
-        try:
-            import google.generativeai as genai
-            gemini_key = os.environ.get("GEMINI_API_KEY")
-            if gemini_key:
-                genai.configure(api_key=gemini_key)
-                gemini_model = genai.GenerativeModel('gemini-2.5-flash')
-                
-                prompt = f"""You are creating a quick audio summary for busy professionals. In EXACTLY {max_chars} characters or less, give the ONE most valuable insight or actionable takeaway from this article.
+        groq_key = os.environ.get("GROQ_API_KEY")
+        gemini_key = os.environ.get("GEMINI_API_KEY")
+        summarize_success = False
+        
+        prompt = f"""You are creating a quick audio summary for busy professionals. In EXACTLY {max_chars} characters or less, give the ONE most valuable insight or actionable takeaway from this article.
 
 Format: Start with the key insight, then briefly explain why it matters.
 Style: Conversational, engaging, like a smart friend sharing a tip.
@@ -564,14 +564,43 @@ Article Content:
 {text[:8000]}
 
 Your {max_chars}-character summary (make every word count):"""
+
+        # Try Groq first (PRIMARY - fastest)
+        if groq_key and not summarize_success:
+            try:
+                client = Groq(api_key=groq_key)
+                response = client.chat.completions.create(
+                    model="llama-3.1-8b-instant",  # Fast model for summarization
+                    messages=[{"role": "user", "content": prompt}],
+                    max_tokens=500,
+                    temperature=0.7
+                )
+                text = response.choices[0].message.content.strip()[:max_chars]
+                summarize_success = True
+                if ctx:
+                    await ctx.info(f"Summarized with Groq: {original_length} -> {len(text)} chars")
+            except Exception as e:
+                if ctx:
+                    await ctx.warning(f"Groq failed: {e}, trying Gemini...")
+        
+        # Fallback to Gemini (BACKUP)
+        if gemini_key and not summarize_success:
+            try:
+                import google.generativeai as genai
+                genai.configure(api_key=gemini_key)
+                gemini_model = genai.GenerativeModel('gemini-2.0-flash')
                 
                 response = await gemini_model.generate_content_async(prompt)
                 text = response.text.strip()[:max_chars]
+                summarize_success = True
                 if ctx:
-                    await ctx.info(f"Summarized: {original_length} -> {len(text)} chars")
-        except Exception as e:
-            if ctx:
-                await ctx.warning(f"Summarization failed: {e}, using truncation")
+                    await ctx.info(f"Summarized with Gemini: {original_length} -> {len(text)} chars")
+            except Exception as e:
+                if ctx:
+                    await ctx.warning(f"Gemini also failed: {e}, using truncation")
+        
+        # Final fallback: truncation
+        if not summarize_success:
             text = text[:max_chars]
     else:
         # Just truncate to model limit
@@ -741,15 +770,14 @@ async def medium_synthesize(topic: str, max_articles: int = 5, ctx: Context = No
     Returns:
         Synthesized research report
     """
-    import google.generativeai as genai
-    
     app = get_app_context(ctx)
     
+    groq_key = os.environ.get("GROQ_API_KEY")
     gemini_key = os.environ.get("GEMINI_API_KEY")
     openai_key = os.environ.get("OPENAI_API_KEY")
     
-    if not gemini_key and not openai_key:
-        return "Error: Neither GEMINI_API_KEY nor OPENAI_API_KEY is set."
+    if not groq_key and not gemini_key and not openai_key:
+        return "Error: No AI API keys set (GROQ_API_KEY, GEMINI_API_KEY, or OPENAI_API_KEY)."
     
     # Scrape articles
     if ctx:
@@ -792,9 +820,25 @@ Articles:
 {context_text}
 """
     
-    # Try Gemini first
+    # Try Groq first (PRIMARY - fastest)
+    if groq_key:
+        try:
+            client = Groq(api_key=groq_key)
+            response = client.chat.completions.create(
+                model="llama-3.3-70b-versatile",  # Best model for synthesis
+                messages=[{"role": "user", "content": prompt}],
+                max_tokens=2000,
+                temperature=0.7
+            )
+            return response.choices[0].message.content
+        except Exception as e:
+            if ctx:
+                await ctx.warning(f"Groq failed: {e}")
+    
+    # Fallback to Gemini
     if gemini_key:
         try:
+            import google.generativeai as genai
             genai.configure(api_key=gemini_key)
             model = genai.GenerativeModel('gemini-2.0-flash')
             response = await model.generate_content_async(prompt)
@@ -814,7 +858,7 @@ Articles:
             )
             return response.choices[0].message.content
         except Exception as e:
-            return f"Error: Both Gemini and OpenAI failed. {e}"
+            return f"Error: All providers failed. Last error: {e}"
     
     return "Error: No AI service available."
 

src/config.py

@@ -28,6 +28,7 @@ class Config:
     DB_PATH = ":memory:" if os.getenv("SPACE_ID") else os.path.join(BASE_DIR, "articles.db")
 
     # API Keys (from shared config)
+    GROQ_API_KEY = _shared.groq_api_key or os.getenv("GROQ_API_KEY")
     GEMINI_API_KEY = _shared.gemini_api_key or os.getenv("GEMINI_API_KEY")
 
     # Scraping Settings (from shared config)  
@@ -81,6 +82,7 @@ class Config:
     @classmethod
     def reload_config(cls):
         cls._shared = SharedConfig.from_env()
+        cls.GROQ_API_KEY = cls._shared.groq_api_key or os.getenv("GROQ_API_KEY")
         cls.GEMINI_API_KEY = cls._shared.gemini_api_key or os.getenv("GEMINI_API_KEY")
         cls.TIMEOUT_MS = cls._shared.default_timeout * 1000
         cls.MAX_WORKERS = int(os.getenv("MAX_WORKERS", cls._shared.max_workers))

src/knowledge.py

@@ -1,5 +1,6 @@
 import os
 import google.generativeai as genai
+from groq import Groq
 from typing import Dict, Any, List, Optional
 import json
 
@@ -8,51 +9,79 @@ import logging
 
 logger = logging.getLogger("KnowledgeGraph")
 
-# Configure Gemini
-if Config.GEMINI_API_KEY:
-    genai.configure(api_key=Config.GEMINI_API_KEY)
-
 def extract_knowledge_graph(text: str) -> Optional[Dict[str, Any]]:
     """
-    Uses Gemini to extract a Knowledge Graph (Concepts & Relationships) from text.
+    Uses Groq (primary) or Gemini (backup) to extract a Knowledge Graph (Concepts & Relationships) from text.
     Returns a JSON object with 'concepts' and 'relationships'.
     """
-    if not Config.GEMINI_API_KEY:
-        logger.warning("GEMINI_API_KEY not set. Skipping Knowledge Graph extraction.")
+    groq_key = os.environ.get("GROQ_API_KEY") or Config.GROQ_API_KEY
+    gemini_key = os.environ.get("GEMINI_API_KEY") or Config.GEMINI_API_KEY
+    
+    if not groq_key and not gemini_key:
+        logger.warning("No LLM API key set (GROQ_API_KEY or GEMINI_API_KEY). Skipping Knowledge Graph extraction.")
         return None
 
-    try:
-        model = genai.GenerativeModel('gemini-1.5-flash')
-        
-        prompt = f"""
-        Analyze the following text and extract a Knowledge Graph.
-        Identify key "Concepts" (entities, technologies, ideas) and "Relationships" between them.
-        
-        Return ONLY a valid JSON object with this structure:
-        {{
-            "concepts": [
-                {{"id": "concept_name", "type": "technology/person/idea", "description": "short definition"}}
-            ],
-            "relationships": [
-                {{"source": "concept_name", "target": "concept_name", "relation": "uses/created/is_a"}}
-            ]
-        }}
-        
-        Text:
-        {text[:10000]}  # Limit text length to avoid token limits
-        """
-        
-        response = model.generate_content(prompt)
-        
-        # Clean up response (remove markdown code blocks if present)
-        content = response.text.strip()
-        if content.startswith("```json"):
-            content = content[7:]
-        if content.endswith("```"):
-            content = content[:-3]
+    prompt = f"""
+Analyze the following text and extract a Knowledge Graph.
+Identify key "Concepts" (entities, technologies, ideas) and "Relationships" between them.
+
+Return ONLY a valid JSON object with this structure:
+{{
+    "concepts": [
+        {{"id": "concept_name", "type": "technology/person/idea", "description": "short definition"}}
+    ],
+    "relationships": [
+        {{"source": "concept_name", "target": "concept_name", "relation": "uses/created/is_a"}}
+    ]
+}}
+
+Text:
+{text[:10000]}
+"""
+
+    # Try Groq first (PRIMARY - fastest)
+    if groq_key:
+        try:
+            client = Groq(api_key=groq_key)
+            response = client.chat.completions.create(
+                model="llama-3.3-70b-versatile",
+                messages=[{"role": "user", "content": prompt}],
+                max_tokens=2000,
+                temperature=0.3
+            )
+            content = response.choices[0].message.content.strip()
             
-        return json.loads(content)
-        
-    except Exception as e:
-        print(f"Error extracting Knowledge Graph: {e}")
-        return None
+            # Clean up response (remove markdown code blocks if present)
+            if content.startswith("```json"):
+                content = content[7:]
+            if content.startswith("```"):
+                content = content[3:]
+            if content.endswith("```"):
+                content = content[:-3]
+            
+            return json.loads(content.strip())
+        except Exception as e:
+            logger.warning(f"Groq failed for Knowledge Graph: {e}, trying Gemini...")
+
+    # Fallback to Gemini (BACKUP)
+    if gemini_key:
+        try:
+            genai.configure(api_key=gemini_key)
+            model = genai.GenerativeModel('gemini-1.5-flash')
+            
+            response = model.generate_content(prompt)
+            
+            # Clean up response (remove markdown code blocks if present)
+            content = response.text.strip()
+            if content.startswith("```json"):
+                content = content[7:]
+            if content.endswith("```"):
+                content = content[:-3]
+                
+            return json.loads(content)
+            
+        except Exception as e:
+            logger.error(f"Gemini also failed for Knowledge Graph: {e}")
+            return None
+
+    return None

src/shared_config.py

@@ -33,6 +33,7 @@ class SharedConfig:
     # ========================================================================
     # API Keys
     # ========================================================================
+    groq_api_key: Optional[str] = None
     gemini_api_key: Optional[str] = None
     openai_api_key: Optional[str] = None
     elevenlabs_api_key: Optional[str] = None
@@ -128,6 +129,7 @@ class SharedConfig:
             max_concurrency=get_env("MAX_CONCURRENCY", 5, int),
             
             # API Keys
+            groq_api_key=get_env("GROQ_API_KEY"),
             gemini_api_key=get_env("GEMINI_API_KEY"),
             openai_api_key=get_env("OPENAI_API_KEY"),
             elevenlabs_api_key=get_env("ELEVENLABS_API_KEY"),
@@ -179,7 +181,7 @@ class SharedConfig:
         safe_dict = self.to_dict()
         
         # Mask sensitive keys
-        sensitive_keys = ['gemini_api_key', 'openai_api_key', 'elevenlabs_api_key']
+        sensitive_keys = ['groq_api_key', 'gemini_api_key', 'openai_api_key', 'elevenlabs_api_key']
         for key in sensitive_keys:
             if safe_dict.get(key):
                 safe_dict[key] = safe_dict[key][:8] + "..." if safe_dict[key] else None