syntaxmatrix 2.5.6__py3-none-any.whl → 2.5.7__py3-none-any.whl

syntaxmatrix/agentic/agents.py

@@ -42,7 +42,7 @@ def token_calculator(total_input_content, llm_profile):
         input_prompt_tokens = len(enc.encode(total_input_content))
         return input_prompt_tokens    
 
-def mlearning_agent(user_prompt, system_prompt, coding_profile, temperature=0.1, max_tokens=4096):
+def mlearning_agent(user_prompt, system_prompt, coding_profile):
     """
     Returns:
         (text, usage_dict)
@@ -95,72 +95,41 @@ def mlearning_agent(user_prompt, system_prompt, coding_profile, temperature=0.1,
     # Google
     def google_generate_code():
         nonlocal usage
-        """
-        Generates content using the Gemini API and calculates token usage
-        including Context Overhead for consistency.
-        """
-        
-        try:
-            # 1. Client Initialization
-            config = types.GenerateContentConfig(
-                system_instruction=system_prompt,
-                temperature=temperature,
-                max_output_tokens=max_tokens,
-            )
+        config = types.GenerateContentConfig(
+            system_instruction=system_prompt,
+            # Optional: Force the model to generate a Python code block as JSON
+            response_mime_type="application/json",
+            response_schema=types.Schema(
+                type=types.Type.OBJECT,
+                properties={
+                    "code": types.Schema(type=types.Type.STRING, description="The runnable Python code."),
+                    "explanation": types.Schema(type=types.Type.STRING, description="A brief explanation of the code."),
+                },
+                required=["code"]
+            ),
+        )
 
-            # 2. API Call
-            resp = _client.models.generate_content(
+        try:
+            response = _client.models.generate_content(
                 model=_model,
-                contents=[user_prompt],
+                contents=user_prompt,
                 config=config,
             )
+        except Exception as e:
+            return f"An error occurred during API call: {e}"
 
-            # 3. Token Usage Capture and Context Overhead Calculation
-            um = resp.usage_metadata
-            usage["input_tokens"] = um.prompt_token_count
-            usage["output_tokens"] = um.thoughts_token_count
-            usage["total_tokens"] = um.total_token_count
-
-            # 4. Response Extraction (same robust logic as before)
-            text = getattr(resp, "text", None)
-            if isinstance(text, str) and text.strip():
-                return text.strip()
-
-            chunks = []
-            candidates = getattr(resp, "candidates", None) or []
-            for cand in candidates:
-                content = getattr(cand, "content", None)
-                if content:
-                    parts = getattr(content, "parts", None) or []
-                    for part in parts:
-                        t = getattr(part, "text", None)
-                        if t:
-                            chunks.append(str(t))
-                            
-            text = "\n".join(chunks).strip()
-            if text:
-                return text
-
-            # 5. Handle blocked response
-            fb = getattr(resp, "prompt_feedback", None)
-            block_reason = getattr(fb, "block_reason", None) if fb else None
-            if block_reason and block_reason != types.BlockedReason.REASON_UNSPECIFIED:
-                raise RuntimeError(f"{_model} blocked the response. Reason: {block_reason.name}")
-            raise RuntimeError(f"{_model} failed to return content due to insufficient data.")
+        # 3. Token Usage Capture and Context Overhead Calculation
+        um = response.usage_metadata
+        usage["input_tokens"] = um.prompt_token_count
+        usage["output_tokens"] = um.candidates_token_count + um.thoughts_token_count
+        usage["total_tokens"] = um.total_token_count
 
-        except APIError as e:
-            error_msg = f"Gemini API Error: {e}"
-        
+        try:
+            # The response text will be a JSON string due to the config.
+            response_json = json.loads(response.text)
+            return response_json.get("code", "Error: Code field not found in response.")
         except Exception as e:
-            error_msg = f"An unexpected error occurred during API call or processing: {e}"
-
-        # --- Return the error message wrapped in the required output code structure ---
-        msg = f"I smxAI have instructed {error_msg}\n"
-        return (
-            f"# {msg}\n"
-            "from syntaxmatrix.display import show\n"
-            f"show({msg!r})\n"
-        )
+            return f"Error parsing response as JSON: {e}\nRaw Response: {response.text}"
 
     # OpenAI Responses API
     def gpt_models_latest_generate_code():
@@ -225,15 +194,14 @@ def mlearning_agent(user_prompt, system_prompt, coding_profile, temperature=0.1,
                 "from syntaxmatrix.display import show\n"
                 f"show({msg!r})\n"
             )
-        
+
     # Anthropic
     def anthropic_generate_code():  
         nonlocal usage      
         try:        
             resp = _client.messages.create(
                 model=_model,
-                max_tokens=max_tokens,
-                temperature=temperature,
+                temperature=0,
                 system=system_prompt,
                 messages=[
                     {"role": "user", "content": user_prompt}
@@ -276,40 +244,44 @@ def mlearning_agent(user_prompt, system_prompt, coding_profile, temperature=0.1,
     def openai_sdk_generate_code():
         nonlocal usage
         try:
-            resp = _client.chat.completions.create(
+            response = None
+            if _model == "deepseek-reasoner":
+                response = _client.chat.completions.create(
+                    model=_model,
+                    messages=[
+                        {"role": "system", "content": system_prompt},
+                        {"role": "user", "content": user_prompt},
+                    ],
+                    extra_body={"thinking": {"type": "enabled"}},
+                    temperature=0,
+                    stream=False
+                )
+            else:
+                response = _client.chat.completions.create(
                 model=_model,
                 messages=[
                     {"role": "system", "content": system_prompt},
                     {"role": "user", "content": user_prompt},
                 ],
-                temperature=temperature,
-                max_tokens=max_tokens,
+                extra_body={"thinking": {"type": "enabled"}},
+                temperature=0,
+                stream=False
             )
+            content = response.choices[0].message.content
 
-
-
-            um = resp.usage
+            um = response.usage
             usage["input_tokens"] = um.prompt_tokens
             usage["output_tokens"] = um.completion_tokens
             usage["total_tokens"] = um.total_tokens
 
-            text = resp.choices[0].message.content
-            if text:
-                return text
-
-            # Try to surface any block reason (safety / policy / etc.)
-            block_reason = None
-            choices = getattr(resp, "choices", None) or []
-            if choices:
-                first = choices[0]
-                fr = getattr(first, "finish_reason", None)
-                if fr and fr != "stop":
-                    block_reason = fr
-
-            if block_reason:
-                raise RuntimeError(f"{_model} stopped with reason: {block_reason}")
-             # Fallback: nothing useful came back
-            raise RuntimeError(f"{_model} returned nothing in this section due to insufficient data.")
+            code_match = re.search(r"```(?:python)?\n(.*?)```", content, re.DOTALL)
+        
+            if code_match:
+                return code_match.group(1).strip()
+            else:
+                # If no markdown blocks are found, return the raw content
+                # (assuming the model obeyed instructions to output only code)
+                return content.strip()
         
         except Exception as e:
             # IMPORTANT: return VALID PYTHON so the dashboard can show the error
@@ -318,9 +290,7 @@ def mlearning_agent(user_prompt, system_prompt, coding_profile, temperature=0.1,
                 f"# {msg}\n"
                 "from syntaxmatrix.display import show\n"
                 f"show({msg!r})\n"
-            )
-
-    # print("TTOOKKEENN: ", token_calculator(system_prompt + user_prompt, coding_profile))   
+            ) 
 
     if _provider == "google":
         code = google_generate_code()
@@ -425,11 +395,12 @@ def refine_question_agent(raw_question: str, dataset_context: str | None = None)
 
         return "Configure LLM Profiles or contact your administrator."
 
-    system_prompt = (
-        "You rewrite user questions into specification Machine Learning (ML) job description. "
-        "If a dataset summary is provided, use it to respect column and help you redefine the question. "
-        "DO NOT write andy prelude or preamble"
-    )
+    system_prompt = ("""
+        - You are a Machine Learning (ML) and Data Science (DS) expert.
+        - You rewrite user questions into clear ML job specifications to help AI assistant generate Python code that provides solution to the user question when it is run. Most user questions are vague. So, your goal is to ensure that your output guards the assistant agains making potential errors that you anticipated could arise due to the nature of the question. 
+        - If a dataset summary is provided, use it to respect column and help you rewrite the question properly.
+        - DO NOT write andy prelude or preamble"
+    """)
 
     user_prompt = f"User question:\n{raw_question}\n\n"
     if dataset_context:
@@ -446,20 +417,7 @@ def refine_question_agent(raw_question: str, dataset_context: str | None = None)
   
 
 def classify_ml_job_agent(refined_question, dataset_profile):   
-    """
-    Instructs an LLM (gemini-2.5-flash) to analyze a task description 
-    and return a list of associated machine learning job/task types. 
-    This version uses a highly extensive, generalized list of ML jobs 
-    to ensure robustness across all domains (NLP, CV, RL, etc.).
-
-    Args:
-        task_description: The detailed description of the statistical/ML task.
-
-    Returns:
-        A list of strings identifying the relevant ML jobs. Returns an empty 
-        list if the API call fails or the output cannot be parsed.
-    """
-
+    import ast
     def ml_response(user_prompt, system_prompt, profile):
         _profile = profile  # _prof.get_profile["admin"]
         
@@ -571,10 +529,14 @@ def classify_ml_job_agent(refined_question, dataset_profile):
 
         return "Configure LLM Profiles or contact your administrator."
 
-    system_prompt = (
-        "You are a strict machine learning task classifier for an ML workbench.\n"
-        "Your job is to label the user's task desc. with all relevant tags from a fixed list.\n\n"
-    )
+    system_prompt = ("""
+        You are a strict machine learning task classifier for an ML workbench.
+        Your goal is to correctly label the user's task specifications with the most relevant tags from a fixed list.
+        You Must always have 'data_preprocessing' as the 1st tag. Then add up to 4 more, as needed, to make 5 max. 
+        Your list should be 2-5 tags long. If no relevant tag, default to ["data_preprocessing"]
+        If tasks specs and `df` don't match (of different industries, return ['context mismatch']
+        You should return only your list of tags, no prelude or preamble.
+    """)
 
     # --- 1. Define the Master List of ML Tasks (Generalized) ---
     ml_task_list = [
@@ -592,7 +554,7 @@ def classify_ml_job_agent(refined_question, dataset_profile):
         "generative_modeling", "causal_inference", "risk_modeling", "graph_analysis",
         
         # Foundational/Pipeline Steps
-        "feature_engineering", "statistical_inference", "data_preprocessing", 
+        "data_preprocessing", "feature_engineering", "statistical_inference", 
         "model_validation", "hyperparameter_tuning"
     ]
     
@@ -600,7 +562,7 @@ def classify_ml_job_agent(refined_question, dataset_profile):
     task_description = refined_question
 
     user_prompt = f"""
-    Analyze the following task description:
+    Analyze and classify the following task description:
     ---
     {task_description}
     ---
@@ -625,9 +587,11 @@ def classify_ml_job_agent(refined_question, dataset_profile):
 
     llm_profile['client'] = _prof.get_client(llm_profile)
 
-    # Extract raw content
     tasks = ml_response(user_prompt, system_prompt, llm_profile)
-    return tasks
+    try:
+        return ast.literal_eval(tasks)
+    except Exception:
+        return tasks
 
 
 def text_formatter_agent(text):

syntaxmatrix/core.py

@@ -599,7 +599,7 @@ class SyntaxMUI:
         from syntaxmatrix.gpt_models_latest import extract_output_text as _out, set_args 
    
         if not self._classification_profile:
-            classification_profile = _prof.get_profile('classification') or _prof.get_profile('chat') or _prof.get_profile('admin')
+            classification_profile = _prof.get_profile('classification') or _prof.get_profile('admin')
             if not classification_profile:
                 return {"Error": "Set a profile for Classification"}
             self._classification_profile = classification_profile
@@ -1141,21 +1141,23 @@ class SyntaxMUI:
         tasks = [str(t).strip().lower() for t in tasks if str(t).strip()]
 
         ai_profile = """
-        - You are a Python expert specializing in data science and machine learning. 
+        - You are a Python expert specializing in Data Science (DS) and Machine Learning (ML).
         - Your task is to generate a single, complete, production-quality, executable Python script for a Jupyter-like Python kernel, based on the given instructions. 
         - The dataset is already loaded as a pandas DataFrame named `df` (no file I/O or file uploads).
-        - Make a copy of `df` and name it `df_copy`. Make sure `df_copy` is preprocessed and cleaned, named `df_cleaned`, if not already done so. Then use `df_cleaned` to perform the ML tasks described in the given context.
-        - Select your features and target, from `df_cleaned`, with care and name it `required_cols`
+        - Make a copy of `df` and name it `df_copy`. 
+        - Make sure `df_copy` is preprocessed and cleaned, and name it `df_cleaned`, if not already done so. 
+        - Work only with `df_cleaned` to perform the ML tasks described in the given context.
+        - Select your features and targets, from `df_cleaned`, with care and name it `required_cols`
         - Create your 'df_filtered by doing: df_filtered = df_cleaned[required_cols].
-        - Use the {TEMPLATE_CATALOGUE} below to educate yourself on which visualizations you will implement in the code.
-        - The final output MUST be the complete, executable Python code only, enclosed in a single markdown code block (```python ... ```), which is required to fulfill the user's request. See the {tasks} below.
+        - Use the {TEMPLATE_CATALOGUE} below to educate yourself on which visualizations you will implement in the code, and ensure the implementations are in the code you generate.
+        - The final output MUST BE the complete, executable Python code only, enclosed in a single markdown code block (```python ... ```), and MUST BE able to fulfill the user's request: {tasks}.
         - Do not include any explanatory text or markdown outside the code block.
         """
 
         TEMPLATE_CATALOGUE = """
         ### Available SyntaxMatrix templates (use these instead of inventing new helpers)
 
-        Visualisation templates (dataset-agnostic):
+        Visualisation templates:
         - viz_pie(df, category_col=None, top_k=8): pie/donut shares within a category.
         - viz_stacked_bar(df, x=None, hue=None, normalise=True): composition across groups.
         - viz_count_bar(df, category_col=None, top_k=12): counts/denominators by category.
@@ -1195,9 +1197,9 @@ class SyntaxMUI:
             
             """
             ### Template rules
-            - You MAY call a template if it matches the task.
+            - You MAY call 1 or more templates if they matche the task.
             - Do NOT invent template names.
-            - If no template fits, write minimal direct pandas/sklearn/seaborn code instead.
+            - If no template fits, write minimal direct pandas/sklearn/seaborn code instead, for visualization.
             - Keep the solution short: avoid writing wrappers/utilities already handled by SyntaxMatrix hardener.
 
             #### Template selection hint examples:
@@ -1220,8 +1222,7 @@ class SyntaxMUI:
                 set `random_state=42` where relevant.
             4) Be defensive, but avoid hard-failing on optional fields:
             - If the primary column, needed to answer the question, is missing, review your copy of the `df` again.
-            Make sure that you selected the proper column. 
-            Never use a column/variable which isn't available or defined.
+            - Make sure that you selected the proper column. Never use a column/variable which isn't available or defined.
             - If a secondary/extra column is missing, show a warning with `show(...)` and continue using available fields.
             - Handle missing values sensibly (drop rows for simple EDA; use `ColumnTransformer` + `SimpleImputer` for modelling).
             - For categorical features in ML, use `OneHotEncoder(handle_unknown="ignore")`
@@ -1253,6 +1254,20 @@ class SyntaxMUI:
             11) You MUST NOT reference any column outside Available columns: {AVAILABLE_COLUMNS}.
             12) If asked to predict/classify, choose the target by matching the task text to Allowed columns
                 and never invent a new name.
+            13) Treat df as the primary dataset you must work with.
+            14) The dataset is already loaded as df (no file I/O or file uploads).
+            15) All outputs must be visible to the user via the provided show(...) helper.
+            16) Never use print(...); use show(...) instead.
+            17) You MUST NOT read from or write to local files, folders, or external storage.
+            - Do not call open(...), Path(...).write_text/write_bytes, or similar file APIs.
+            - Do not use df.to_csv(...), df.to_excel(...), df.to_parquet(...),
+                df.to_pickle(...), df.to_json(...), df.to_hdf(...), or any other
+                method that writes to disk.
+            - Do not call joblib.dump(...), pickle.dump(...), torch.save(...),
+                numpy.save(...), numpy.savetxt(...), or similar saving functions.
+            - Do not call plt.savefig(..., 'somefile.png') or any variant that
+                writes an image to a filename. Plots must be rendered in-memory only.
+            18) Keep everything in memory and surface results via show(...) or plots.
             
             #### Cohort rules
             When you generate plots for cohorts or categories, you MUST obey these rules:
@@ -1302,11 +1317,11 @@ class SyntaxMUI:
         """)
 
         if not self._coding_profile:
-            coding_profile = _prof.get_profile("coding") or _prof.get_profile("admin")
+            coding_profile = _prof.get_profile("coding")  # or _prof.get_profile("admin")
             if not coding_profile:
                 return (
                     '<div class="smx-alert smx-alert-warn">'
-                        'No LLM profile configured for <code>coding</code> (or <code>admin</code>). <br>'
+                        'No LLM profile configured for <code>coding</code> <br>'
                         'Please, add the LLM profile inside the admin panel or contact your Administrator.'
                     '</div>'
                 )

syntaxmatrix/preface.py

@@ -453,23 +453,24 @@ def _safe_concat(objs, **kwargs):
     except Exception as e:
         smx_show(f'⚠ concat skipped: {e}')
         return _pd.DataFrame()
-
+    
 
 def _SMX_OHE(**k):
     # normalise arg name across sklearn versions
-    if 'sparse' in k and 'sparse_output' not in k:
-        k['sparse_output'] = k.pop('sparse')
-    k.setdefault('handle_unknown', 'ignore')
-    k.setdefault('sparse_output', False)
+    if "sparse" in k and "sparse_output" not in k:
+        k["sparse_output"] = k.pop("sparse")
+    k.setdefault("handle_unknown", "ignore")
+    k.setdefault("sparse_output", False)
     try:
-        sig = inspect.signature(OneHotEncoder)
-        if 'sparse_output' not in sig.parameters and 'sparse_output' in k:
-            k['sparse'] = k.pop('sparse_output')
-    except Exception:
-        if 'sparse_output' in k:
-            k['sparse'] = k.pop('sparse_output')
-    return OneHotEncoder(**k)
-
+        if "sparse_output" not in inspect.signature(OneHotEncoder).parameters:
+            if "sparse_output" in k:
+                k["sparse"] = k.pop("sparse_output")
+        return OneHotEncoder(**k)
+    except TypeError:
+        if "sparse_output" in k:
+            k["sparse"] = k.pop("sparse_output")
+        return OneHotEncoder(**k)
+        
 
 def _SMX_mm(a, b):
     try:

syntaxmatrix/routes.py

@@ -3047,7 +3047,7 @@ def setup_routes(smx):
                         }) + "\n\n"
 
             except GeneratorExit:
-                smx.info("Client aborted the stream.")
+                return "Client aborted the stream."
             except Exception as e:
                 smx.error(f"Stream error: {e}")
                 yield "data: " + json.dumps({"event": "error", "error": str(e)}) + "\n\n"
@@ -6513,8 +6513,7 @@ def setup_routes(smx):
                 cell["highlighted_code"] = Markup(_pygmentize(cell["code"]))
        
         highlighted_ai_code = _pygmentize(ai_code)
-        tasks = [tag.replace("_", " ").replace('"', '').capitalize() for tag in tags]
-
+        
         return render_template(
           "dashboard.html",
           section=section,
@@ -6525,7 +6524,7 @@ def setup_routes(smx):
           highlighted_ai_code=highlighted_ai_code if ai_code else None,
           askai_question=smx.sanitize_rough_to_markdown_task(askai_question),  
           refined_question=refined_question,  
-          tasks=tasks,
+          tasks=tags,
           data_cells=data_cells,
           session_id=session_id,
           llm_usage=llm_usage

syntaxmatrix/settings/model_map.py

@@ -36,12 +36,14 @@ PROVIDERS_MODELS = {
         
     ],
     #4
-    "deepseek": [                    
+    "deepseek": [  
+        "deepseek-reasoner",          
         "deepseek-chat",
     ],
     #5
     "moonshot": [
-        "kimi-k2-0905-preview",
+        "kimi-k2-thinking",
+        "kimi-k2",
     ],
     #6
     "alibaba": [   

syntaxmatrix/templates/dashboard.html

@@ -15,7 +15,7 @@
         position: fixed;
         top: 0; left: 0;
         width: 200px; height: 100vh;
-        background: #f1f4f7;
+        background: #a3a4a5ff;
         border-right: 1px solid #ccc;
         padding: 26px 10px 10px 14px;
         box-sizing: border-box;
@@ -56,7 +56,7 @@
         min-height: 100vh;
         box-sizing: border-box;
         overflow-x: auto;
-        background: #f8f9fb;
+        background: #dff0f5ff;
         font-size: clamp(0.98rem, 2vw, 1.07rem);
       }
       .dashboard-tabs {
@@ -85,7 +85,7 @@
         top: 2px;
       }
       .dashboard-content {
-        background: #fff;
+        background: #bbbbbd;
         width: 100%;
         padding: 10px;
         border-radius: 0 0 10px 10px;
@@ -95,6 +95,10 @@
         overflow-x: auto;
         margin-right: 1vw; 
       }
+
+      textarea#askai{
+        background: #e5e5e5cd;
+      }
       
       .smx-table {
         padding: clamp(3px, 1vw, 9px) clamp(4px, 2vw, 13px);
@@ -590,6 +594,31 @@
       .eda-card h3{ color: #1f2937 !important; }
       .smx-stat h4{ color: #64748b !important; }
     </style>
+    <style>
+      div.li > li {
+        margin-left: 35px;
+      }
+      /* 1. Style the arrow specifically */
+      .toggle-arrow {
+        display: inline-block;        /* CRITICAL: Allows the element to rotate */
+        transition: transform 0.2s;   /* Makes the rotation smooth */
+        margin-right: 6px;            /* Spacing between arrow and text */
+      }
+
+      /* 2. Rotate and color when the menu is OPEN */
+      details[open] summary .toggle-arrow {
+        transform: rotate(90deg);     /* Rotates the arrow downwards */
+        color: #007acc;               /* Changes color to blue */
+      }
+
+      /* 3. (Optional) Remove default browser markers to avoid double arrows */
+      details > summary {
+        list-style: none;
+      }
+      details > summary::-webkit-details-marker {
+        display: none;
+      }
+    </style>
 
     <script src="https://cdn.plot.ly/plotly-latest.min.js"></script>
 </head>
@@ -611,8 +640,7 @@
   <div class="dashboard-main">
     <button id="sidebarToggle" class="sidebar-toggle" aria-label="Open menu"></button>
     <ul class="dashboard-tabs">
-      <li class="{{ 'active' if section == 'explore' else '' }}"><a href="/dashboard?section=explore">Explore</a></li>
-      
+      <li class="{{ 'active' if section == 'explore' else '' }}"><a href="/dashboard?section=explore">Explore</a></li>  
     </ul>
     <div class="dashboard-content">
       <div class="explore-card">
@@ -679,37 +707,41 @@
               -->
               <br><br>
               <div class="refined-qblock">
-                  <span class="refined-q-label"><b>My Thought Process:</b></span><br>
-                  <span class="refined-q">{{ refined_question|safe }}</span>
-              </div><br>
-              <div class="refined-qblock">
-                {% if tasks %}
-                  <b>Tasks Performed: </b><br>
-                  {% for task in tasks %}
-                    <span class="refined-q">- {{ task }}</span><br>
-                  {% endfor %}
-                {% endif %}
-              </div><br>
-              {% if llm_usage %}
-                <div class="refined-qblock">
-                  <b>LLM: </b><span>{{ llm_usage.provider }} | {{ llm_usage.model }}</span><br>
-                  <b>Token Usage: </b>
-                    <li>- Input Tokens: {{ llm_usage.input_tokens }}</li>
-                    <li>- Output Tokens: {{ llm_usage.output_tokens }}</li>
-                    <li>- Total Tokens: {{ llm_usage.total_tokens }}</li>         
-                  </ul>
-                </div>
-              {% endif %}
+                <details>
+                  <summary class="refined-q-label" style="cursor: pointer; list-style: none;">
+                      <span class="toggle-arrow">▶</span> 
+                      <b>Thought Process</b>
+                  </summary>
+                  <div class="li" style="margin-top: 10px; padding-left: 14px; border-left: 3px solid #e0e5ee;">
+                      <span class="refined-q">{{ refined_question|safe }}</span>
+                      <br><br>
+                      {% if tasks %}
+                        <b>Tasks Performed: </b>
+                        {% for task in tasks %}
+                          <li>{{ task.replace('_', ' ').capitalize() }}</li>
+                        {% endfor %}
+                      {% endif %} 
+                      <br><br>
+                      {% if llm_usage %}
+                        <b>LLM: </b><span>{{ llm_usage.provider }} | {{ llm_usage.model }}</span><br> 
+                        <b>Token Usage: </b>
+                          <li>Input Tokens: {{ llm_usage.input_tokens }}</li>
+                          <li>Output Tokens: {{ llm_usage.output_tokens }}</li>
+                          <li>Total Tokens: {{ llm_usage.total_tokens }}</li>         
+                      </div>
+                    {% endif %} 
+                  </div>
+                </details>
             {% endif %}
             {% if ai_outputs %}
               <div class="d-flex align-items-center justify-content-between" style="margin: 12px;">
                 <h3 class="m-0">Result</h3>
+                {% for html_block in ai_outputs %}
+                    <div class="ai-output" style="margin-bottom:18px;overflow-x:auto; max-width:100%;">
+                        {{ html_block | safe }}
+                    </div>
+                {% endfor %}
               </div>
-              {% for html_block in ai_outputs %}
-                  <div class="ai-output" style="margin-bottom:18px;overflow-x:auto; max-width:100%;">
-                      {{ html_block | safe }}
-                  </div>
-              {% endfor %}
             {% endif %}
             {% if ai_code %}
               <div>

syntaxmatrix/utils.py

@@ -155,31 +155,6 @@ def harden_ai_code(code: str) -> str:
     # Remove any LLM-added try/except blocks (hardener adds its own)
     import re
 
-    def strip_placeholders(code: str) -> str:
-        code = re.sub(r"\bshow\(\s*\.\.\.\s*\)", 
-                    "show('⚠ Block skipped due to an error.')", 
-                    code)
-        code = re.sub(r"\breturn\s+\.\.\.", "return None", code)
-        return code
-
-    def _SMX_OHE(**k):
-        # normalise arg name across sklearn versions
-        if "sparse" in k and "sparse_output" not in k:
-            k["sparse_output"] = k.pop("sparse")
-        # default behaviour we want
-        k.setdefault("handle_unknown", "ignore")
-        k.setdefault("sparse_output", False)
-        try:
-            # if running on old sklearn without sparse_output, translate back
-            if "sparse_output" not in inspect.signature(OneHotEncoder).parameters:
-                if "sparse_output" in k:
-                    k["sparse"] = k.pop("sparse_output")
-            return OneHotEncoder(**k)
-        except TypeError:
-            # final fallback: try legacy name
-            if "sparse_output" in k:
-                k["sparse"] = k.pop("sparse_output")
-            return OneHotEncoder(**k)
     
     def _strip_stray_backrefs(code: str) -> str:
         code = re.sub(r'(?m)^\s*\\\d+\s*', '', code)     
@@ -497,6 +472,67 @@ def harden_ai_code(code: str) -> str:
             """
         )
 
+    def _strip_file_io_ops(code: str) -> str:
+        """
+        Remove obvious local file I/O operations in LLM code
+        so nothing writes to the container filesystem.
+        """
+        # 1) Methods like df.to_csv(...), df.to_excel(...), etc.
+        FILE_WRITE_METHODS = (
+            "to_csv", "to_excel", "to_pickle", "to_parquet",
+            "to_json", "to_hdf",
+        )
+
+        for mname in FILE_WRITE_METHODS:
+            pat = re.compile(
+                rf"(?m)^(\s*)([A-Za-z_][A-Za-z0-9_\.]*)\s*\.\s*{mname}\s*\([^)]*\)\s*$"
+            )
+
+            def _repl(match):
+                indent = match.group(1)
+                expr = match.group(2)
+                return f"{indent}# [SMX] stripped file write: {expr}.{mname}(...)"
+
+            code = pat.sub(_repl, code)
+
+        # 2) plt.savefig(...) calls
+        pat_savefig = re.compile(r"(?m)^(\s*)(plt\.savefig\s*\([^)]*\)\s*)$")
+        code = pat_savefig.sub(
+            lambda m: f"{m.group(1)}# [SMX] stripped savefig: {m.group(2).strip()}",
+            code,
+        )
+
+        # 3) with open(..., 'w'/'wb') as f:
+        pat_with_open = re.compile(
+            r"(?m)^(\s*)with\s+open\([^)]*['\"]w[b]?['\"][^)]*\)\s+as\s+([A-Za-z_][A-Za-z0-9_]*)\s*:\s*$"
+        )
+
+        def _with_open_repl(match):
+            indent = match.group(1)
+            var = match.group(2)
+            return f"{indent}if False:  # [SMX] file write stripped (was: with open(... as {var}))"
+
+        code = pat_with_open.sub(_with_open_repl, code)
+
+        # 4) joblib.dump(...), pickle.dump(...)
+        for mod in ("joblib", "pickle"):
+            pat = re.compile(rf"(?m)^(\s*){mod}\.dump\s*\([^)]*\)\s*$")
+            code = pat.sub(
+                lambda m: f"{m.group(1)}# [SMX] stripped {mod}.dump(...)",
+                code,
+            )
+
+        # 5) bare open(..., 'w'/'wb') calls
+        pat_open = re.compile(
+            r"(?m)^(\s*)open\([^)]*['\"]w[b]?['\"][^)]*\)\s*$"
+        )
+        code = pat_open.sub(
+            lambda m: f"{m.group(1)}# [SMX] stripped open(..., 'w'/'wb')",
+            code,
+        )
+
+        return code
+
     # Register and run patches once per execution
     for _patch in (
         _smx_patch_mean_squared_error_squared_kw,
@@ -598,6 +634,7 @@ def harden_ai_code(code: str) -> str:
     fixed = _wrap_metric_calls(fixed)
     fixed = _fix_unexpected_indent(fixed)
     fixed = _patch_feature_coef_dataframe(fixed)
+    fixed = _strip_file_io_ops(fixed)
 
     # Import shared preface helpers once and wrap the LLM body safely
     header = "from syntaxmatrix.preface import *\n\n"

{syntaxmatrix-2.5.6.dist-info → syntaxmatrix-2.5.7.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: syntaxmatrix
-Version: 2.5.6
+Version: 2.5.7
 Summary: SyntaxMUI: A customizable framework for Python AI Assistant Projects.
 Author: Bob Nti
 Author-email: bob.nti@syntaxmatrix.net

{syntaxmatrix-2.5.6.dist-info → syntaxmatrix-2.5.7.dist-info}/RECORD

@@ -2,7 +2,7 @@ syntaxmatrix/__init__.py,sha256=_LnTrYAW2tbYA37Y233Vv4OMOk8NUnoJi-1yzFyHxEI,2573
 syntaxmatrix/auth.py,sha256=SCD6uWojXjj9yjUTKzgV5kBYe6ZkXASEG2VopLFkEtM,18140
 syntaxmatrix/bootstrap.py,sha256=Y7ZNg-Z3ecrr1iYem5EMzPmGstXnEKmO9kqKVoOoljo,817
 syntaxmatrix/commentary.py,sha256=3uSlbaQ1zl-gYtEtEpFbv2M-IH-HSdFdMvhxa7UCNHk,12025
-syntaxmatrix/core.py,sha256=vih5LnpcLMEfEqTZeRHhHNocbjWZCSgcChFV1t45CYs,60179
+syntaxmatrix/core.py,sha256=7o5givPq7Io9DCnJQjTy-izgYvJivE2POxvG45i3dBk,61338
 syntaxmatrix/dataset_preprocessing.py,sha256=wtV4MWzkyfOsBHTsS0H1gqHho77ZQHGDI9skJryyZWA,8732
 syntaxmatrix/db.py,sha256=xkCpyhFxnAwrnZCTd13NkJsahVze0i4egjMcbB7kPfs,5000
 syntaxmatrix/display.py,sha256=TgMrE5WW80VlLcL_XvEz936mekFccJgLTfzbCIozSc8,3728
@@ -15,28 +15,28 @@ syntaxmatrix/kernel_manager.py,sha256=sE9zwuqEZq10Q4ySpGn0ilx-ui7cmZw-LEK8GxK-Hh
 syntaxmatrix/llm_store.py,sha256=c22-ahR_PmZVWB5OAKPVr01YI9rWPWDd_aSEMujhAic,7500
 syntaxmatrix/models.py,sha256=-yGj4fALYqyQqxIiB0Eh6xWSlr9GfwuoDlzAWUikye8,533
 syntaxmatrix/plottings.py,sha256=MjHQ9T1_oC5oyr4_wkM2GJDrpjp0sbvudbs2lGaMyzk,6103
-syntaxmatrix/preface.py,sha256=hAV8t0U87PXzbIOKEz-RkYxKV5a766dOoE71HWloQ_s,16960
+syntaxmatrix/preface.py,sha256=EOK3lflMJ-0B6SRJtVXhzZjhvu-bfXzw-sy1TbTYOVs,17009
 syntaxmatrix/profiles.py,sha256=0-lky7Wj-WQlP5CbvTyw1tI2M0FiqhhTkLZYLRhD5AU,2251
 syntaxmatrix/project_root.py,sha256=1ckvbFVV1szHtHsfSCoGcImHkRwbfszmPG1kGh9ZZlE,2227
-syntaxmatrix/routes.py,sha256=tGBSccUs9iNuMnjpZmtvn8jDRc1Sy2aAb0t0sKoBUmE,302995
+syntaxmatrix/routes.py,sha256=kkLn6uOfOiD8-Kl2_JKs0qcxBHkuJd6cFih9v8KuTWk,302912
 syntaxmatrix/session.py,sha256=v0qgxnVM_LEaNvZQJSa-13Q2eiwc3RDnjd2SahNnHQk,599
 syntaxmatrix/smiv.py,sha256=1lSN3UYpXvYoVNd6VrkY5iZuF_nDxD6xxvLnTn9wcbQ,1405
 syntaxmatrix/smpv.py,sha256=rrCgYqfjBaK2n5qzfQyXK3bHFMvgNcCIqPaXquOLtDM,3600
 syntaxmatrix/themes.py,sha256=qa90vPZTuNNKB37loZhChQfu5QqkaJG4JxgI_4QgCxw,3576
 syntaxmatrix/ui_modes.py,sha256=5lfKK3AKAB-JQCWfi1GRYp4sQqg4Z0fC3RJ8G3VGCMw,152
-syntaxmatrix/utils.py,sha256=0UG1e9XwU3E2m6xblI6yPavKmEQzSIYsvTzkWiN6xJ4,122458
+syntaxmatrix/utils.py,sha256=0iTu9XbUN1HsZModWmyexYrXAzjox7gpHyYV7SmW-PM,123555
 syntaxmatrix/vector_db.py,sha256=ozvOcMHt52xFAvcp-vAqT69kECPq9BwL8Rzgq3AJaMs,5824
 syntaxmatrix/vectorizer.py,sha256=5w_UQiUIirm_W-Q9TcaEI8LTcTYIuDBdKfz79T1aZ8g,1366
 syntaxmatrix/workspace_db.py,sha256=Xu9OlW8wo3iaH5Y88ZMdLOf-fiZxF1NBb5rAw3KcbfY,4715
 syntaxmatrix/agentic/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 syntaxmatrix/agentic/agent_tools.py,sha256=yQwavONP23ziMxNQf3j2Y4TVo_LxEsiAWecKuBK8WDg,866
-syntaxmatrix/agentic/agents.py,sha256=gy3XDzydAQdpp-GQyRL8TjIhAAO_ruPnmo4tvJQyARE,30588
+syntaxmatrix/agentic/agents.py,sha256=m4eqeCf4fVRodCa1Pgc3LwT4EPGip4E-7n1k1HxeKxA,29898
 syntaxmatrix/agentic/code_tools_registry.py,sha256=Wp4-KHtp0BUVciqSbmionBsQMVFOnvJPruBJeNiuwkk,1564
 syntaxmatrix/agentic/model_templates.py,sha256=A3ROE3BHkvnU9cxqSGjlCBIw9U15zRaTKgK-WxcZtUI,76033
 syntaxmatrix/settings/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 syntaxmatrix/settings/default.yaml,sha256=BznvF1D06VMPbT6UX3MQ4zUkXxTXLnAA53aUu8G4O38,569
 syntaxmatrix/settings/logging.py,sha256=U8iTDFv0H1ECdIzH9He2CtOVlK1x5KHCk126Zn5Vi7M,1362
-syntaxmatrix/settings/model_map.py,sha256=of49J294XnPRXYXUtdApg1t9VRajwZoZplVD9mBDSIc,11990
+syntaxmatrix/settings/model_map.py,sha256=M6EPTPw2m88RFLNlzxt6vQzEtdr66NcHz0t4zB7QfkU,12028
 syntaxmatrix/settings/prompts.py,sha256=dLNijnw9UHlAg5qxcSaLPhTmR7SdDDyOFcMKhlCA4eQ,21695
 syntaxmatrix/settings/string_navbar.py,sha256=NqgTzo3J9rRI4c278VG6kpoViFfmi2FKmL6sO0R-bus,83
 syntaxmatrix/static/docs.md,sha256=rWlKjNcpS2cs5DElGNYuaA-XXdGZnRGMXx62nACvDwE,11105
@@ -51,7 +51,7 @@ syntaxmatrix/static/js/sidebar.js,sha256=zHp4skKLY2Dlqx7aLPQ8_cR0iTRT17W0SC2TR38
 syntaxmatrix/static/js/widgets.js,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 syntaxmatrix/templates/change_password.html,sha256=YWEcnwJLccLyKGzQxIrc0xuP-p00BtEIwcYq4oFvJ-0,3332
 syntaxmatrix/templates/code_cell.html,sha256=LOr9VjvNQcOGKKJ1ecpcZh3C3qsUxBHueg2iQtpdxl8,638
-syntaxmatrix/templates/dashboard.html,sha256=1Qf-5IztYOkc0f7QSExV3Fo57Ks0r_BCbYVp7SqvW-4,30996
+syntaxmatrix/templates/dashboard.html,sha256=lR0wUtozTh5bDHbPSiywJiypiT_nNfzvEJJLfWckE0w,32272
 syntaxmatrix/templates/docs.html,sha256=KVi5JrZD3gwOduiZhAz7hQrKY9SrQ_bsHOODj0Nj09s,3552
 syntaxmatrix/templates/error.html,sha256=Iu5ykHnhw8jrxVBNn6B95e90W5u9I2hySCiLtaoOJMs,3290
 syntaxmatrix/templates/login.html,sha256=V_bWHozS1xCeHPsvAAfaGG-_2lAE7K8d05IarQN1PS8,2677
@@ -63,8 +63,8 @@ syntaxmatrix/vectordb/adapters/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5
 syntaxmatrix/vectordb/adapters/milvus_adapter.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 syntaxmatrix/vectordb/adapters/pgvector_adapter.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 syntaxmatrix/vectordb/adapters/sqlite_adapter.py,sha256=L8M2qHfwZRAFVxWeurUVdHaJXz6F5xTUSWh3uy6TSUs,6035
-syntaxmatrix-2.5.6.dist-info/licenses/LICENSE.txt,sha256=j1P8naTdy1JMxTC80XYQjbyAQnuOlpDusCUhncrvpy8,1083
-syntaxmatrix-2.5.6.dist-info/METADATA,sha256=vel34pxHc1Tf6YwflfaiGnw6TkUW9JnTbWIFiZijZyg,18090
-syntaxmatrix-2.5.6.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-syntaxmatrix-2.5.6.dist-info/top_level.txt,sha256=HKP_zkl4V_nt7osC15DlacoBZktHrbZYOqf_pPkF3T8,13
-syntaxmatrix-2.5.6.dist-info/RECORD,,
+syntaxmatrix-2.5.7.dist-info/licenses/LICENSE.txt,sha256=j1P8naTdy1JMxTC80XYQjbyAQnuOlpDusCUhncrvpy8,1083
+syntaxmatrix-2.5.7.dist-info/METADATA,sha256=2_xno8Sgx4iERRKmk5GfgluVQDY_94aPFdmx3NC85dw,18090
+syntaxmatrix-2.5.7.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+syntaxmatrix-2.5.7.dist-info/top_level.txt,sha256=HKP_zkl4V_nt7osC15DlacoBZktHrbZYOqf_pPkF3T8,13
+syntaxmatrix-2.5.7.dist-info/RECORD,,