perf: Optimize RAM footprint to stay under 512MB for low-memory containers

Dockerfile

@@ -6,6 +6,9 @@ ENV PYTHONUNBUFFERED=1 \
     PYTHONDONTWRITEBYTECODE=1 \
     PORT=7860 \
     HOST=0.0.0.0 \
+    ENABLE_PROMPT_GUARD=false \
+    OMP_NUM_THREADS=1 \
+    MKL_NUM_THREADS=1 \
     DATA_DIR=/app/knowledge_base \
     INDEX_DIR=/app/knowledge_base/indexes \
     PROCESSED_DIR=/app/knowledge_base/processed \

config.py

@@ -170,7 +170,7 @@ MIN_CONFIDENT_MATCH_SCORE = float(os.getenv("MIN_CONFIDENT_MATCH_SCORE", "0.35")
 GROUNDING_OVERLAP_THRESHOLD = 0.30
 
 # Meta Prompt-Guard 86M Sub-10ms Neural Safety & Indirect Prompt Injection Guardrail
-ENABLE_PROMPT_GUARD = os.getenv("ENABLE_PROMPT_GUARD", "true").lower() == "true"
+ENABLE_PROMPT_GUARD = os.getenv("ENABLE_PROMPT_GUARD", "false").lower() == "true"
 PROMPT_GUARD_MODEL_NAME = os.getenv("PROMPT_GUARD_MODEL_NAME", "meta-llama/Prompt-Guard-86M")
 PROMPT_GUARD_ONNX_REPO = os.getenv("PROMPT_GUARD_ONNX_REPO", "prompt-security/Prompt-Guard-86M_onnx")
 PROMPT_GUARD_ONNX_PATH = ONNX_MODELS_DIR / "prompt_guard_86m.onnx"

scratch/test_render.py

@@ -0,0 +1,21 @@
+import urllib.request
+import urllib.parse
+import json
+
+url = "https://visionquest-1.onrender.com/query"
+
+# Test 1: Text Query
+data = urllib.parse.urlencode({"text": "Who was J. Robert Oppenheimer?"}).encode("utf-8")
+req = urllib.request.Request(url, data=data)
+
+try:
+    res = urllib.request.urlopen(req)
+    print("=== TEST 1 (Text Query) ===")
+    print("Status:", res.status)
+    body_text = res.read().decode("utf-8")
+    print("Body preview:", body_text[:300])
+except Exception as e:
+    print("=== TEST 1 ERROR ===")
+    print(e)
+    if hasattr(e, "read"):
+        print("Error content:", e.read().decode("utf-8"))