Update app.py

app.py

@@ -22,6 +22,13 @@ exadata_specs = {
     "X11M": {"Quarter Rack": {"max_iops": 600000, "max_throughput": 40}, "Half Rack": {"max_iops": 1200000, "max_throughput": 80}, "Full Rack": {"max_iops": 2400000, "max_throughput": 160}},
 }
 
+# --- Supported LLM Models ---
+supported_llms = {
+    "gpt-3.5-turbo": "Fastest / lowest cost (basic analysis), Standard AWR Healthcheck",
+    "gpt-4-turbo": "Balanced (recommended default), Production Performance Analysis",
+    "gpt-4o": "Deep + technical (best), Deep Dive, Exadata, RAC Stability, Risk Audit",
+}
+
 # --- Utils ---
 def clean_awr_content(content):
     if "<html" in content.lower():
@@ -32,7 +39,7 @@ def clean_awr_content(content):
 # === AGENTS ===
 
 class CriticalAnalyzerAgent:
-    def analyze(self, content, performance_test_mode, exadata_model, rack_size):
+    def analyze(self, content, performance_test_mode, exadata_model, rack_size, llm_model):
         cleaned_content = clean_awr_content(content)
         if len(cleaned_content) > 128000:
             cleaned_content = cleaned_content[:128000] + "\n\n[TRUNCATED]..."
@@ -69,7 +76,7 @@ Compare observed vs theoretical. Recommend actions to close the performance gap.
 """
 
         response = client.chat.completions.create(
-            model="gpt-4-turbo",
+            model=llm_model,
             messages=[
                 {"role": "system", "content": "You are an expert Oracle DBA."},
                 {"role": "user", "content": prompt}
@@ -79,7 +86,7 @@ Compare observed vs theoretical. Recommend actions to close the performance gap.
         return response.choices[0].message.content.strip()
 
 class HealthAgent:
-    def check_health(self, content):
+    def check_health(self, content, llm_model):
         cleaned_content = clean_awr_content(content)
         if len(cleaned_content) > 128000:
             cleaned_content = cleaned_content[:128000] + "\n\n[TRUNCATED]..."
@@ -105,7 +112,7 @@ AWR CONTENT:
 """
 
         response = client.chat.completions.create(
-            model="gpt-4-turbo",  # or "gpt-4o" if preferred/available
+            model=llm_model,
             messages=[
                 {"role": "system", "content": "You are the strict Oracle AWR Health Analysis Agent."},
                 {"role": "user", "content": prompt}
@@ -115,26 +122,26 @@ AWR CONTENT:
         return response.choices[0].message.content.strip()
 
 class RaterAgent:
-    def rate(self, content):
+    def rate(self, content, llm_model):
         prompt = f"Rate the following analysis from 1-5 stars and explain:\n\n{content}"
         response = client.chat.completions.create(
-            model="gpt-4-turbo",
+            model=llm_model,
             messages=[{"role": "user", "content": prompt}]
         )
         return response.choices[0].message.content.strip()
 
 # === Main Process ===
-def process_awr(awr_text, threshold, performance_test_mode, exadata_model, rack_size):
+def process_awr(awr_text, threshold, performance_test_mode, exadata_model, rack_size, llm_model):
     analyzer = CriticalAnalyzerAgent()
     health = HealthAgent()
     rater = RaterAgent()
 
     if not awr_text.strip():
-        return "No AWR text provided", "", ""
+        return "No AWR text provided", "", "", ""
 
-    analysis = analyzer.analyze(awr_text, performance_test_mode, exadata_model, rack_size)
-    health_status = health.check_health(awr_text)
-    rating_text = rater.rate(analysis)
+    analysis = analyzer.analyze(awr_text, performance_test_mode, exadata_model, rack_size, llm_model)
+    health_status = health.check_health(awr_text, llm_model)
+    rating_text = rater.rate(analysis, llm_model)
 
     stars = 0
     match = re.search(r"(\d+)", rating_text)
@@ -144,8 +151,8 @@ def process_awr(awr_text, threshold, performance_test_mode, exadata_model, rack_
     retry_status = "✅ Accepted"
 
     if stars < threshold:
-        analysis_retry = analyzer.analyze(awr_text, performance_test_mode, exadata_model, rack_size)
-        rating_text_retry = rater.rate(analysis_retry)
+        analysis_retry = analyzer.analyze(awr_text, performance_test_mode, exadata_model, rack_size, llm_model)
+        rating_text_retry = rater.rate(analysis_retry, llm_model)
         retry_status = "✅ Retry Occurred"
         analysis = analysis_retry
         rating_text = rating_text_retry
@@ -161,6 +168,7 @@ with gr.Blocks() as demo:
     performance_test_mode = gr.Checkbox(label="Performance Test Mode")
     exadata_model = gr.Dropdown(choices=list(exadata_specs.keys()), label="Exadata Model", visible=False)
     rack_size = gr.Dropdown(choices=["Quarter Rack", "Half Rack", "Full Rack"], label="Rack Size", visible=False)
+    llm_selector = gr.Dropdown(choices=list(supported_llms.keys()), value="gpt-4-turbo", label="LLM Model")
 
     def toggle_visibility(mode):
         return gr.update(visible=mode), gr.update(visible=mode)
@@ -169,10 +177,12 @@ with gr.Blocks() as demo:
 
     analyze_btn = gr.Button("Analyze AWR Report")
     output = gr.Textbox(label="AWR Analysis", lines=20)
-    health = gr.Textbox(label="Health Agent Findings", lines=5)
+    health = gr.Textbox(label="Health Agent Findings", lines=10)
     rating = gr.Textbox(label="Rater", lines=3)
     retry_status = gr.Textbox(label="Retry Status")
 
-    analyze_btn.click(process_awr, inputs=[awr_text, threshold, performance_test_mode, exadata_model, rack_size], outputs=[output, health, rating, retry_status])
+    analyze_btn.click(process_awr, 
+                      inputs=[awr_text, threshold, performance_test_mode, exadata_model, rack_size, llm_selector], 
+                      outputs=[output, health, rating, retry_status])
 
-demo.launch(debug=True)
+demo.launch(debug=True)