fix: Remove temperature parameter for GPT-5 compatibility (only supports default)

analyzer/chat/demo_app.py

@@ -438,8 +438,8 @@ class GrantAnalystDemo:
                 messages=self.messages,
                 tools=self.available_tools,
                 tool_choice="auto",
-                temperature=0.2,  # Lower temperature for more consistent, thorough responses
                 max_completion_tokens=4096,  # GPT-5 uses max_completion_tokens
+                # Note: GPT-5 only supports temperature=1.0 (default), so we don't set it
             )
             timing_info["llm_call"] = time.time() - llm_start
 
@@ -481,8 +481,8 @@ class GrantAnalystDemo:
                 final_response = self.llm_client.client.chat.completions.create(
                     model=self.llm_client.model,
                     messages=self.messages,
-                    temperature=0.2,  # Lower temperature for more complete, consistent responses
                     max_completion_tokens=4096,  # GPT-5 uses max_completion_tokens
+                    # Note: GPT-5 only supports temperature=1.0 (default), so we don't set it
                 )
                 timing_info["final_llm_call"] = time.time() - final_start
 

analyzer/llm_client.py

@@ -228,12 +228,13 @@ class LLMClient:
             api_params = {
                 "model": model,
                 "messages": messages,
-                "temperature": temperature,
-                "top_p": top_p,
                 "max_completion_tokens": max_tokens,  # GPT-5 uses max_completion_tokens
                 "stream": stream,
             }
 
+            # GPT-5 only supports temperature=1.0 (default), so don't set it
+            # For other models, we could add temperature back if needed
+
             # Add GPT-5 specific parameters if provided
             if verbosity is not None:
                 api_params["verbosity"] = verbosity

src/analyzer/chat/demo_app.py

@@ -438,8 +438,8 @@ class GrantAnalystDemo:
                 messages=self.messages,
                 tools=self.available_tools,
                 tool_choice="auto",
-                temperature=0.2,  # Lower temperature for more consistent, thorough responses
                 max_completion_tokens=4096,  # GPT-5 uses max_completion_tokens
+                # Note: GPT-5 only supports temperature=1.0 (default), so we don't set it
             )
             timing_info["llm_call"] = time.time() - llm_start
 
@@ -481,8 +481,8 @@ class GrantAnalystDemo:
                 final_response = self.llm_client.client.chat.completions.create(
                     model=self.llm_client.model,
                     messages=self.messages,
-                    temperature=0.2,  # Lower temperature for more complete, consistent responses
                     max_completion_tokens=4096,  # GPT-5 uses max_completion_tokens
+                    # Note: GPT-5 only supports temperature=1.0 (default), so we don't set it
                 )
                 timing_info["final_llm_call"] = time.time() - final_start
 

src/analyzer/llm_client.py

@@ -228,12 +228,13 @@ class LLMClient:
             api_params = {
                 "model": model,
                 "messages": messages,
-                "temperature": temperature,
-                "top_p": top_p,
                 "max_completion_tokens": max_tokens,  # GPT-5 uses max_completion_tokens
                 "stream": stream,
             }
 
+            # GPT-5 only supports temperature=1.0 (default), so don't set it
+            # For other models, we could add temperature back if needed
+
             # Add GPT-5 specific parameters if provided
             if verbosity is not None:
                 api_params["verbosity"] = verbosity