> ## Documentation Index
> Fetch the complete documentation index at: https://signalpilot.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Ask Mode

> Read-only question answering mode for exploration, explanations, and learning without modifying your notebook

# Ask Mode

**No notebook changes** — the AI answers questions about your notebook, explains code, and provides guidance without modifying anything.

<Info>
  **Best for:** Exploration, explanations, learning, and getting advice without altering your cells. Perfect when you want to understand before you act.
</Info>

## How Ask Mode Works

Ask Mode is purely conversational. The AI has read-only access to your notebook and can help you understand what's happening without making any changes:

<Steps>
  <Step title="AI reads your notebook">
    The AI can see:

    * All cells and their outputs
    * Current kernel state (variables, dataframes)
    * Database connections
    * Previous execution results

    <Note>
      The AI can **read** everything but cannot **modify** anything.
    </Note>
  </Step>

  <Step title="You ask questions">
    Ask anything about your notebook:

    * "What does this code do?"
    * "Why is this cell producing an error?"
    * "How can I optimize this query?"
    * "Explain the logic in cell 7"
    * "What's the difference between these two approaches?"
  </Step>

  <Step title="AI provides explanations">
    The AI responds with:

    * Clear explanations in natural language
    * Code breakdowns and logic walkthroughs
    * Debugging suggestions
    * Performance optimization ideas
    * Alternative approaches

    <Check>
      No code is generated or executed. It's purely advisory.
    </Check>
  </Step>

  <Step title="You decide what to do">
    Armed with the AI's explanation, you:

    * Manually edit cells based on suggestions
    * Run experiments to test ideas
    * Switch to Hands-On or Agent Mode if you want AI to implement changes
  </Step>
</Steps>

## Example Interactions

<AccordionGroup>
  <Accordion title="Understanding code" icon="code">
    **You:** "What does cell 5 do?"

    **AI:** "Cell 5 performs a left join between the `customers` and `orders` dataframes on the `customer_id` column. It then filters for orders from 2024 and groups by region to calculate total revenue. The result is sorted by revenue in descending order."

    <Tip>
      Ask Mode is perfect for onboarding to unfamiliar notebooks or understanding code written by teammates.
    </Tip>
  </Accordion>

  <Accordion title="Debugging without changes" icon="bug">
    **You:** "Why is the last cell showing a KeyError?"

    **AI:** "The KeyError on `revenue_usd` suggests that column doesn't exist in your dataframe. Looking at cell 3, the column is actually named `revenue_dollars`. You have two options:

    1. Rename the column in cell 3 to `revenue_usd`
    2. Update cell 7 to use `revenue_dollars`

    I recommend option 2 to maintain consistency with your earlier analysis."

    The AI explains the problem but doesn't automatically fix it. You decide which approach to take.
  </Accordion>

  <Accordion title="Learning and exploration" icon="graduation-cap">
    **You:** "How would I calculate a rolling average in pandas?"

    **AI:** "To calculate a rolling average in pandas, use the `.rolling()` method:

    ```python theme={null}
    df['rolling_avg'] = df['value'].rolling(window=7).mean()
    ```

    This computes a 7-period moving average. You can adjust `window` to change the period. For your `daily_revenue` dataframe, you could do:

    ```python theme={null}
    daily_revenue['7day_avg'] = daily_revenue['revenue'].rolling(window=7).mean()
    ```

    Would you like me to explain how rolling windows handle edge cases?"

    <Note>
      The AI provides educational explanations without modifying your notebook. Perfect for learning new techniques.
    </Note>
  </Accordion>

  <Accordion title="Performance optimization" icon="gauge">
    **You:** "Is there a faster way to do this groupby operation?"

    **AI:** "Yes! Your current approach loops through groups, which is slow for large datasets. Here's a vectorized alternative:

    **Current approach (slow):**

    ```python theme={null}
    for group in df.groupby('category'):
        # process each group
    ```

    **Faster vectorized approach:**

    ```python theme={null}
    df.groupby('category').agg({'revenue': 'sum', 'quantity': 'mean'})
    ```

    For your dataset with 1M rows, this could be 10-100x faster. The vectorized version lets pandas optimize operations internally instead of Python looping."

    The AI explains the optimization but lets you implement it.
  </Accordion>

  <Accordion title="Data quality checks" icon="clipboard-check">
    **You:** "Are there any issues with my data that I should be aware of?"

    **AI:** "I notice a few potential issues in your `sales_data` dataframe:

    1. **Missing values:** 342 rows have null `customer_id` (8% of total)
    2. **Negative revenue:** 17 rows have negative `revenue` values (likely refunds?)
    3. **Date range gap:** No data between March 15-22, 2024
    4. **Outliers:** 5 transactions over $100k (99th percentile is $8,500)

    Would you like suggestions on how to handle these issues?"

    <Check>
      The AI proactively identifies data quality concerns without making assumptions about how to fix them.
    </Check>
  </Accordion>
</AccordionGroup>

## When to Use Ask Mode

<CardGroup cols={2}>
  <Card title="Before making changes" icon="shield-check">
    Use Ask Mode to understand the current state before modifying anything. Get a clear picture of what exists, then switch modes to implement changes.
  </Card>

  <Card title="Code review" icon="magnifying-glass">
    Review notebooks from teammates or understand legacy code without risking accidental modifications.
  </Card>

  <Card title="Learning new techniques" icon="book">
    Ask how to implement something or understand a coding pattern. The AI teaches without writing code for you.
  </Card>

  <Card title="Sanity checks" icon="clipboard-check">
    Verify your logic, check for edge cases, or get a second opinion on your approach.
  </Card>
</CardGroup>

## Limitations

<Warning>
  **Ask Mode cannot:**

  * Generate new code cells
  * Edit existing cells
  * Execute code
  * Create visualizations
  * Modify kernel state
  * Run database queries

  For any of these actions, switch to [Hands-On Mode](/docs/reference/agent-modes/hands-on-mode) or [Agent Mode](/docs/reference/agent-modes/agent-mode).
</Warning>

## Switching from Ask Mode

When you're ready to implement changes, switching is seamless:

<Tabs>
  <Tab title="To Hands-On Mode">
    **When to switch:** You understand what needs to change and want to approve each suggestion.

    **What happens:**

    * Conversation history is preserved
    * Context from Ask Mode carries over
    * AI starts suggesting code instead of just explaining

    **Example flow:**

    1. Ask Mode: "Why is this slow?"
    2. AI explains the issue
    3. Switch to Hands-On Mode
    4. AI suggests optimized code
    5. You review and approve
  </Tab>

  <Tab title="To Agent Mode">
    **When to switch:** You want the AI to implement a complete solution autonomously.

    **What happens:**

    * AI creates a plan based on your conversation
    * Shows you the full plan for approval
    * Executes everything after you approve

    **Example flow:**

    1. Ask Mode: "How would I analyze customer churn?"
    2. AI explains the approach
    3. Switch to Agent Mode: "Implement that analysis"
    4. AI creates multi-step plan
    5. You approve and it executes
  </Tab>

  <Tab title="Manual Implementation">
    **When to switch:** You want to implement changes yourself based on AI advice.

    **What happens:**

    * You manually edit cells
    * Use AI explanations as guidance
    * Run code yourself
    * Stay in full control

    **Best for:** Learning, practicing, or when you need custom adjustments beyond AI suggestions.
  </Tab>
</Tabs>

## Conversation Continuity

One of Ask Mode's strengths is that conversations carry context:

```
You: "What does this analysis calculate?"

AI: "This calculates monthly recurring revenue (MRR) by summing active subscriptions..."

You: "What's the retention rate?"

AI: "Based on the MRR calculation we just discussed, I can see retention in cell 12..."

You: "How can I improve the accuracy?"

AI: "To improve the MRR accuracy, consider these adjustments..."
```

The AI remembers the conversation flow and builds on previous answers.

<Tip>
  Use Ask Mode for extended exploratory conversations. Once you understand the full picture, switch modes to implement changes efficiently.
</Tip>

## FAQ

<AccordionGroup>
  <Accordion title="Does Ask Mode use my data?" icon="database">
    Yes, Ask Mode reads your notebook state to answer questions accurately. It can see:

    * Cell contents and outputs
    * Kernel variables and dataframes
    * Database connections
    * Execution history

    But as always, SignalPilot follows a **zero data retention policy**. Your data never leaves your environment.
  </Accordion>

  <Accordion title="Can I ask about database schemas in Ask Mode?" icon="table">
    Yes! If you have database connections configured, Ask Mode can:

    * Describe table schemas
    * Explain relationships between tables
    * Suggest queries
    * Identify performance issues

    It can read database metadata but cannot execute queries (read-only mode).
  </Accordion>

  <Accordion title="Will Ask Mode tell me if my code has bugs?" icon="bug">
    Yes. Ask Mode can:

    * Identify logical errors
    * Spot common pitfalls (e.g., division by zero, type mismatches)
    * Warn about performance issues
    * Suggest edge cases to test

    But it won't automatically fix bugs — it explains them so you can decide how to fix them.
  </Accordion>

  <Accordion title="Can I use Ask Mode with MCP servers?" icon="plug">
    Yes! Ask Mode can read from MCP servers to provide context:

    * dbt model documentation
    * Slack discussion history
    * Jira ticket details
    * Query history

    This helps the AI give more informed answers based on your full organizational context.

    <Card title="Learn about MCP" icon="plug" href="/docs/integrations/skills-and-mcp-servers">
      See how SignalPilot connects to external tools via Model Context Protocol
    </Card>
  </Accordion>

  <Accordion title="Can I ask hypothetical questions?" icon="lightbulb">
    Absolutely! Ask Mode is perfect for:

    * "What would happen if I changed this parameter?"
    * "How would this perform on a larger dataset?"
    * "What's the trade-off between these two approaches?"
    * "If I modify this cell, what else needs to change?"

    The AI can reason about hypotheticals without executing anything.
  </Accordion>

  <Accordion title="Is Ask Mode faster than other modes?" icon="bolt">
    Generally yes, because:

    * No code generation required
    * No execution overhead
    * Simpler interactions
    * Lighter context processing

    Ask Mode responses are typically 2-3x faster than code-generating modes.
  </Accordion>
</AccordionGroup>

## Next Steps

<CardGroup cols={2}>
  <Card title="Try Hands-On Mode" icon="hand" href="/docs/reference/agent-modes/hands-on-mode">
    Get code suggestions with granular approval control
  </Card>

  <Card title="Try Agent Mode" icon="robot" href="/docs/reference/agent-modes/agent-mode">
    Let the AI plan and execute full analyses autonomously
  </Card>

  <Card title="Cell-Level Editing" icon="pen" href="/docs/reference/cell-editing/inline-edit">
    Learn about inline edits and tab autocomplete
  </Card>

  <Card title="5-Minute Quickstart" icon="rocket" href="/docs/getting-started/quickstart">
    Complete your first analysis with SignalPilot
  </Card>
</CardGroup>