Cortex rework in progress

cortex/reasoning/reflection.py

@@ -1,42 +1,57 @@
 # reflection.py
-from llm.llm_router import call_llm
 import json
+import os
+import re
+from llm.llm_router import call_llm
 
 
 async def reflect_notes(intake_summary: str, identity_block: dict | None) -> dict:
     """
-    Generate reflection notes (internal guidance) for the reasoning engine.
-    These notes help simulate continuity and identity without being shown to the user.
+    Produce short internal reflection notes for Cortex.
+    These are NOT shown to the user.
     """
 
+    # -----------------------------
+    # Build the prompt
+    # -----------------------------
     identity_text = ""
     if identity_block:
         identity_text = f"Identity:\n{identity_block}\n\n"
 
     prompt = (
-    f"{identity_text}"
-    f"Recent summary:\n{intake_summary}\n\n"
-    "You are Lyra's meta-awareness layer. Your job is to produce short, directive "
-    "internal notes that guide Lyra’s reasoning engine. These notes are NEVER "
-    "shown to the user.\n\n"
-    "Rules for output:\n"
-    "1. Return ONLY valid JSON.\n"
-    "2. JSON must have exactly one key: \"notes\".\n"
-    "3. \"notes\" must be a list of 3 to 6 short strings.\n"
-    "4. Notes must be actionable (e.g., \"keep it concise\", \"maintain context\").\n"
-    "5. No markdown, no apologies, no explanations.\n\n"
-    "Return JSON:\n"
-    "{ \"notes\": [\"...\"] }\n"
+        f"{identity_text}"
+        f"Recent summary:\n{intake_summary}\n\n"
+        "You are Lyra's meta-awareness layer. Your job is to produce short, directive "
+        "internal notes that guide Lyra’s reasoning engine. These notes are NEVER "
+        "shown to the user.\n\n"
+        "Rules for output:\n"
+        "1. Return ONLY valid JSON.\n"
+        "2. JSON must have exactly one key: \"notes\".\n"
+        "3. \"notes\" must be a list of 3 to 6 short strings.\n"
+        "4. Notes must be actionable (e.g., \"keep it concise\", \"maintain context\").\n"
+        "5. No markdown, no apologies, no explanations.\n\n"
+        "Return JSON:\n"
+        "{ \"notes\": [\"...\"] }\n"
     )
 
-    import os
-    backend = os.getenv("LLM_FORCE_BACKEND", "primary")
+    # -----------------------------
+    # Module-specific backend choice
+    # -----------------------------
+    reflection_backend = os.getenv("REFLECTION_LLM")
+    cortex_backend = os.getenv("CORTEX_LLM", "PRIMARY").upper()
 
+    # Reflection uses its own backend if set, otherwise cortex backend
+    backend = (reflection_backend or cortex_backend).upper()
+
+    # -----------------------------
+    # Call the selected LLM backend
+    # -----------------------------
     raw = await call_llm(prompt, backend=backend)
-  
     print("[Reflection-Raw]:", raw)
 
-
+    # -----------------------------
+    # Try direct JSON
+    # -----------------------------
     try:
         parsed = json.loads(raw.strip())
         if isinstance(parsed, dict) and "notes" in parsed:
@@ -44,10 +59,11 @@ async def reflect_notes(intake_summary: str, identity_block: dict | None) -> dic
     except:
         pass
 
-    # Try to extract JSON inside text
+    # -----------------------------
+    # Try JSON extraction
+    # -----------------------------
     try:
-        import re
-        match = re.search(r'\{.*?\}', raw, re.S)   # <-- non-greedy !
+        match = re.search(r"\{.*?\}", raw, re.S)
         if match:
             parsed = json.loads(match.group(0))
             if isinstance(parsed, dict) and "notes" in parsed:
@@ -55,5 +71,7 @@ async def reflect_notes(intake_summary: str, identity_block: dict | None) -> dic
     except:
         pass
 
-    # Final fallback
-    return {"notes": [raw.strip()]}
+    # -----------------------------
+    # Fallback — treat raw text as a single note
+    # -----------------------------
+    return {"notes": [raw.strip()]}