Engine Upgrade: Continuous Pulse Loop & Real-time Local Summarization (Qwen)

app/core/agent.py

@@ -1,5 +1,6 @@
 import asyncio
 import logging
+import random
 from datetime import datetime
 from typing import Dict, List
 from app.core.scrapers import scraper
@@ -12,7 +13,7 @@ logger = logging.getLogger("hawk.agent")
 class HawkAgent:
     def __init__(self):
         self.is_running = False
-        self.interval_hours = 6  # User mentioned 6-hour cycle
+        self.pulse_interval = settings.PULSE_INTERVAL_MINUTES
 
     def notify_jewel_discovery(self, jewel: Dict):
         """Notify the user about a high-signal discovery."""
@@ -33,6 +34,12 @@ class HawkAgent:
                 return
             logger.info(f"PHASE 1 COMPLETE: Captured {len(raw_signals)} potential signals.")
             
+            # Real-time local summarization for the telemetry stream
+            logger.info("PHASE 1.5: Generating immediate forensic gists via local Qwen...")
+            for s in raw_signals:
+                text = f"{s.get('title', '')} {s.get('description', '')}"
+                s['local_summary'] = await sieve.summarize_locally(text)
+            
             # Save raw signals for the 'All' stream in frontend
             vault.save_raw(raw_signals)
 
@@ -77,8 +84,10 @@ class HawkAgent:
         
         while self.is_running:
             await self.run_swoop_cycle()
-            logger.info(f"Hawk is resting for {self.interval_hours} hours...")
-            await asyncio.sleep(self.interval_hours * 3600)
+            # Randomized breather between pulses (jitter)
+            breather = self.pulse_interval + random.randint(-2, 5)
+            logger.info(f"Hawk is hovering for {breather} minutes before next pulse...")
+            await asyncio.sleep(max(1, breather) * 60)
 
     def stop(self):
         """Stop the autonomous loop."""

app/core/config.py

@@ -17,6 +17,7 @@ class Settings(BaseSettings):
     
     # Model Settings
     LOCAL_MODEL_NAME: str = "Qwen/Qwen2.5-0.5B-Instruct" 
+    PULSE_INTERVAL_MINUTES: int = 15 # Breather between discovery pulses
     
     # Search & Extraction Settings
     TARGET_QUERIES: list = [

app/core/filters.py

@@ -34,6 +34,25 @@ class SignalFilter:
         finally:
             self._is_loading = False
 
+    async def summarize_locally(self, text: str) -> str:
+        """Provide a 1-sentence technical gist using local LLM."""
+        await self.ensure_model_loaded()
+        if not self.model:
+            return "Local analysis offline."
+
+        prompt = f"System: You are an architectural forensic analyst. Provide a 1-sentence technical gist of the following signal.\nUser: {text[:800]}\nSummary:"
+        
+        try:
+            inputs = self.tokenizer(prompt, return_tensors="pt").to(self.model.device)
+            outputs = self.model.generate(**inputs, max_new_tokens=40)
+            response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
+            # Extract just the summary part
+            summary = response.split("Summary:")[-1].strip()
+            return summary if summary else "No distinctive pattern identified."
+        except Exception as e:
+            logger.error(f"Summarization error: {e}")
+            return "Technical synthesis failed."
+
     async def classify_with_llm(self, text: str) -> bool:
         """Use local LLM to decide if content is high-signal or junk."""
         await self.ensure_model_loaded()