Update app.py

app.py

@@ -18,8 +18,11 @@ OpenAI-only teaching rig for building AI model UIs.
 • Uses GPT-5 for text generation.
 • Uses DALL·E 3 for image generation.
 • Lets you edit the system prompt, role, tone, and output format.
-• Provides sliders and controls to experiment with sampling and behavior.
+• Provides sliders and controls to experiment with behavior.
 • Includes starter prompts to show different use cases (chat, reports, infographics, visuals).
+
+NOTE: Some newer GPT-5 variants ignore temperature/top-p/penalty parameters.
+This app keeps the controls for teaching, and encodes their values into the instructions instead.
 """
 
 DEFAULT_TEMPERATURE = 0.7
@@ -55,10 +58,15 @@ def build_system_instructions(
     assistant_role: str,
     output_mode: str,
     tone: str,
+    temperature: float,
+    top_p: float,
+    presence_penalty: float,
+    frequency_penalty: float,
 ) -> str:
     """
     Build a system prompt string combining user-provided base instructions
-    with role + format + tone metadata.
+    with role + format + tone + "virtual sampling" metadata.
+    (We encode the slider settings here since GPT-5 may not support those params directly.)
     """
     role_map = {
         "General Assistant": "Behave as a highly capable, calm general-purpose AI assistant.",
@@ -88,6 +96,16 @@ def build_system_instructions(
         "Minimalist": "Be extremely concise. Prefer fewer words and high information density.",
     }
 
+    sampling_hint = (
+        "SAMPLING HINTS FROM UI SLIDERS:\n"
+        f"- Temperature slider: {temperature:.2f} (higher = more creative and speculative).\n"
+        f"- Top-p slider: {top_p:.2f} (lower = more conservative).\n"
+        f"- Presence penalty slider: {presence_penalty:.2f} (higher = encourage new topics).\n"
+        f"- Frequency penalty slider: {frequency_penalty:.2f} (higher = reduce repetition).\n"
+        "You must interpret these values as behavioral guidance even if the underlying "
+        "model ignores sampling parameters."
+    )
+
     parts = [
         (user_system_prompt or "").strip(),
         "",
@@ -95,6 +113,8 @@ def build_system_instructions(
         f"OUTPUT MODE: {output_map.get(output_mode, '')}",
         f"TONE: {tone_map.get(tone, '')}",
         "",
+        sampling_hint,
+        "",
         "Always output clean Markdown.",
     ]
     return "\n".join(p for p in parts if p.strip())
@@ -131,23 +151,21 @@ def history_to_openai_messages(
 def call_openai_text(
     openai_key: Optional[str],
     messages: List[Dict[str, str]],
-    temperature: float,
-    top_p: float,
     max_tokens: int,
-    presence_penalty: float,
-    frequency_penalty: float,
 ) -> str:
     """
-    Call GPT-5 via Chat Completions using the new max_completion_tokens parameter.
+    Call GPT-5 via Chat Completions using only supported parameters:
+    - model
+    - messages
+    - max_completion_tokens
+
+    Newer GPT-5 variants may not support custom temperature/top_p/penalties,
+    so we rely on the system prompt for behavior control instead.
     """
     client = get_openai_client(openai_key)
     completion = client.chat.completions.create(
         model="gpt-5",  # change to exact variant you have (e.g. "gpt-5.1-mini") if needed
         messages=messages,
-        temperature=temperature,
-        top_p=top_p,
-        presence_penalty=presence_penalty,
-        frequency_penalty=frequency_penalty,
         max_completion_tokens=max_tokens,
     )
     return completion.choices[0].message.content
@@ -234,12 +252,16 @@ def agent_assembler_chat(
     if not user_message.strip():
         return chat_history, None
 
-    # Build system instructions
+    # Build system instructions (including slider hints)
     system_instructions = build_system_instructions(
         user_system_prompt=system_prompt_ui,
         assistant_role=assistant_role,
         output_mode=output_mode,
         tone=tone,
+        temperature=temperature,
+        top_p=top_p,
+        presence_penalty=presence_penalty,
+        frequency_penalty=frequency_penalty,
     )
 
     # Prepare messages for OpenAI
@@ -249,16 +271,12 @@ def agent_assembler_chat(
         system_instructions=system_instructions,
     )
 
-    # Call GPT-5
+    # Call GPT-5 (minimal parameters)
     try:
         ai_reply = call_openai_text(
             openai_key=openai_key_ui,
             messages=openai_messages,
-            temperature=temperature,
-            top_p=top_p,
             max_tokens=max_tokens,
-            presence_penalty=presence_penalty,
-            frequency_penalty=frequency_penalty,
         )
     except Exception as e:
         ai_reply = (