Update app.py

app.py

@@ -1,12 +1,16 @@
 #!/usr/bin/env python3
 """
-Overthinker v20 — Gradio Server Backend
+Overthinker v24 — Gradio Server Backend
 
 Dual Prompt Architecture:
 - Input nodes -> prompts to generate OPTIONS/CHOICES/DECISIONS
 - Outcome nodes -> prompts to generate OUTCOMES/CONSEQUENCES
 - Model: nvidia/nemotron-3-nano-30b-a3b
 Port: 7860
+
+Toggle:
+- USE_HUGGINGFACE = True  -> Uses HuggingFace Inference Client (HF_TOKEN from .env)
+- USE_HUGGINGFACE = False -> Uses OpenRouter/OpenAI fallback (original behavior)
 """
 
 import os
@@ -26,10 +30,27 @@ from gradio import Server
 from fastapi import HTTPException
 from starlette.responses import HTMLResponse
 
+# ────────────────────────────────────────────────
+# HuggingFace Inference Client (optional import)
+# ────────────────────────────────────────────────
+HF_AVAILABLE = False
+try:
+    from huggingface_hub import InferenceClient
+    HF_AVAILABLE = True
+except ImportError:
+    InferenceClient = None
+    print("[Warning] huggingface_hub not installed. HuggingFace mode disabled.")
+
 load_dotenv()
 
 OPENROUTER_API_KEY = os.getenv('OPENROUTER_API_KEY', '')
 OPENAI_API_KEY = os.getenv('OPENAI_API_KEY', '')
+HF_TOKEN = os.getenv('HF_TOKEN', '')
+
+# ────────────────────────────────────────────────
+# TOGGLE: Set True to use HuggingFace Inference Client
+# ────────────────────────────────────────────────
+USE_HUGGINGFACE = True  # <-- Set to False to use OpenRouter/OpenAI instead
 
 app = Server()
 PORT = 7860
@@ -132,9 +153,39 @@ class HistoryManager:
 history_manager = HistoryManager()
 
 # ────────────────────────────────────────────────
-# LLM API Call
+# LLM API Call — with HuggingFace Toggle
 # ────────────────────────────────────────────────
 def call_api(prompt: str, system_prompt: str = "You are a helpful assistant that generates decision trees.") -> Optional[str]:
+    
+    # ─── HuggingFace Inference Client Mode ───
+    if USE_HUGGINGFACE:
+        if not HF_TOKEN:
+            print("[HF Error] No HF_TOKEN found in environment variables")
+            return None
+        if not HF_AVAILABLE:
+            print("[HF Error] huggingface_hub not installed. Run: pip install huggingface-hub>=0.23.0")
+            return None
+        
+        try:
+            client = InferenceClient(token=HF_TOKEN)
+            
+            response = client.chat.completions.create(
+                model=DEFAULT_MODEL,
+                messages=[
+                    {"role": "system", "content": system_prompt},
+                    {"role": "user", "content": prompt}
+                ],
+                temperature=0.8,
+                max_tokens=2048
+            )
+            
+            return response.choices[0].message.content
+            
+        except Exception as e:
+            print(f"[HF Exception] {e}")
+            return None
+
+    # ─── OpenRouter Mode (original fallback) ───
     if OPENROUTER_API_KEY:
         try:
             headers = {
@@ -276,12 +327,12 @@ def _fallback_outcomes(decision: str, context: str = "") -> dict:
     labels = [pos, neu, neg]
     desc_map = {
         'positive': f'If this path unfolds favorably, {pos.lower()}. This represents a best-case scenario where your decision leads to growth and improvement.',
-        'neutral': f'On this path, {neu.lower()}. The outcome is neither clearly good nor bad — it requires careful monitoring.',
+        'neutral': f'On this path, {neu.lower()}. The outcome is neither clearly good nor bad \u2014 it requires careful monitoring.',
         'negative': f'In a challenging scenario, {neg.lower()}. This represents potential risks and difficulties that may arise.'
     }
     tip_map = {
         'positive': 'Nurture this positive outcome by staying engaged and proactive.',
-        'neutral': 'Monitor this neutral path closely — small changes can shift the outcome.',
+        'neutral': 'Monitor this neutral path closely \u2014 small changes can shift the outcome.',
         'negative': 'Prepare contingency plans to mitigate this risk if it materializes.'
     }
     children = []
@@ -300,7 +351,7 @@ def _fallback_outcomes(decision: str, context: str = "") -> dict:
     return {'children': children}
 
 # ────────────────────────────────────────────────
-# Prompt Builders — Dual prompts
+# Prompt Builders \u2014 Dual prompts
 # ────────────────────────��───────────────────────
 def build_root_prompt(decision: str) -> str:
     return f'''You are an AI that helps people explore decisions by generating decision trees.
@@ -329,11 +380,11 @@ Parent node: "{decision_label}"
 Description: "{decision_desc}"
 {comment_section}
 
-Generate EXACTLY {count} child nodes that represent different OPTIONS or CHOICES the person could take. Each child should be a distinct, realistic possibility — an actionable decision they could make at this point.
+Generate EXACTLY {count} child nodes that represent different OPTIONS or CHOICES the person could take. Each child should be a distinct, realistic possibility \u2014 an actionable decision they could make at this point.
 
 IMPORTANT: Frame each child as an OPTION or CHOICE, not as an outcome. For example:
 - GOOD: "Consult a financial advisor" (describes an action/choice)
-- BAD: "Financial situation improves" (describes an outcome — DO NOT use here)
+- BAD: "Financial situation improves" (describes an outcome \u2014 DO NOT use here)
 
 Return ONLY valid JSON with exactly this structure (no markdown, no backticks):
 {{
@@ -372,7 +423,7 @@ Generate EXACTLY {count} child nodes that represent a DIVERSE RANGE of possible
 
 IMPORTANT: Frame each child as an OUTCOME or CONSEQUENCE, not as a choice someone makes. For example:
 - GOOD: "Financial stability improves" (describes a result)
-- BAD: "Consider financial planning" (describes a choice — DO NOT do this)
+- BAD: "Consider financial planning" (describes a choice \u2014 DO NOT do this)
 
 Return ONLY valid JSON with exactly this structure (no markdown, no backticks):
 {{
@@ -420,7 +471,7 @@ async def index():
     if os.path.exists(html_path):
         with open(html_path, "r", encoding="utf-8") as f:
             return HTMLResponse(content=f.read(), status_code=200)
-    return HTMLResponse(content="<h1>Overthinker v20</h1><p>index.html not found</p>", status_code=404)
+    return HTMLResponse(content="<h1>Overthinker v24</h1><p>index.html not found</p>", status_code=404)
 
 
 @app.post("/create_tree")
@@ -641,15 +692,15 @@ async def export_path_md(request: dict):
     if not tree_id or not node_id:
         raise HTTPException(status_code=400, detail="Missing tree_id or node_id")
     path = node_manager.get_path(tree_id, node_id)
-    md = '# 🧠 Overthinker — Decision Path\n\n'
+    md = '# \U0001f9e0 Overthinker \u2014 Decision Path\n\n'
     for i, node in enumerate(path):
         indent = '  ' * i
-        emoji = {'root': '🌳', 'input': '🧠', 'outcome': '📊'}.get(node.get('type', ''), '📌')
+        emoji = {'root': '\U0001f333', 'input': '\U0001f9e0', 'outcome': '\U0001f4ca'}.get(node.get('type', ''), '\U0001f4cc')
         md += f'{indent}{emoji} **{node.get("label", "")}**\n'
         if node.get('description'):
             md += f'{indent}  > {node.get("description", "")}\n'
         if node.get('tips') and len(node['tips']) > 0:
-            md += f'{indent}  > 💡 {node["tips"][0]}\n'
+            md += f'{indent}  > \U0001f4a1 {node["tips"][0]}\n'
         md += '\n'
     return md
 
@@ -658,12 +709,24 @@ async def export_path_md(request: dict):
 # Launch
 # ────────────────────────────────────────────────
 if __name__ == "__main__":
-    print(f"🧠 Overthinker v20 — Starting on port {PORT}")
-    print(f"🤖 Model: {DEFAULT_MODEL}")
-    print(f"🌐 Open http://localhost:{PORT} in your browser")
-    if not OPENROUTER_API_KEY and not OPENAI_API_KEY:
-        print("⚠️  No API key found. Using local fallback generation (limited).")
-    print(f"📚 API docs available at http://localhost:{PORT}/docs")
+    mode_str = "HuggingFace Inference" if USE_HUGGINGFACE else "OpenRouter/OpenAI"
+    print(f"\U0001f9e0 Overthinker v24 \u2014 Starting on port {PORT}")
+    print(f"\U0001f916 Model: {DEFAULT_MODEL}")
+    print(f"\U0001f500 Inference Mode: {mode_str}")
+    print(f"\U0001f310 Open http://localhost:{PORT} in your browser")
+    
+    if USE_HUGGINGFACE:
+        if not HF_TOKEN:
+            print("\u26a0\ufe0f  No HF_TOKEN found. Set it in your .env file.")
+        elif not HF_AVAILABLE:
+            print("\u26a0\ufe0f  huggingface_hub not installed. Run: pip install huggingface-hub>=0.23.0")
+        else:
+            print(f"\u2705 HuggingFace Inference Client ready (token: {HF_TOKEN[:6]}...{HF_TOKEN[-4:]})")
+    else:
+        if not OPENROUTER_API_KEY and not OPENAI_API_KEY:
+            print("\u26a0\ufe0f  No API key found. Using local fallback generation (limited).")
+    
+    print(f"\U0001f4da API docs available at http://localhost:{PORT}/docs")
     app.launch(
         server_port=PORT,
         show_error=True,