Frequently Asked Questions

Platform and Architecture

What is the difference between the MCP Client and the Orchestrator?

• Think of the MCP Client as the interface — it discovers available tools, sends your requests in a structured way, and receives the final output. The look and feel of this layer is much like Claude or ChatGPT.
• The MCP Orchestrator is the brain of the operation; it's Petavue’s logic layer that intelligently decides which tools to use and in what sequence. You don’t see this layer as such you only interact with it through the MCP client.
• Together, they ensure your questions trigger the correct workflow for accurate and seamless results.

What changed with the new native Petavue MCP Client?

Previously, Petavue operated using Claude's MCP. That means you had to have a Claude account to access Petavue through Claude’s chat-based interface.

Now, we have upgraded to our own native MCP Client, which connects directly to our servers. This shift gives you:

• Better Performance: Faster and more reliable execution of analyses.
• No Context Limits: Enables longer, more complex conversations without losing context.
• Consistent Results: Reduces errors and interruptions for a smoother workflow.

Is there a limit on the number of messages I can have in one conversation thread?

While the underlying model we use has a technical context limit (around 200,000 tokens), we have implemented advanced context management to ensure you can have long, detailed conversations without worrying about this limit.

In very rare and extremely long conversations, it is possible to hit the context window, which would prevent the thread from continuing. However, we are constantly improving how we manage conversation history and are developing fallback mechanisms to prevent your analysis from ending abruptly. For all practical purposes, you should feel free to continue your analysis within a single thread.

Can I take the context from one conversation and use it in another?

No, the context from one conversation thread does not automatically carry over to a new one. Each conversation is a separate, self-contained session.

However, there are ways to reuse your work across different conversations:

• Key Definitions (KDs): Any custom metric or definition you save as a KD is available globally. You can use that KD in any new conversation, and Petavue will apply the exact same logic.
• Saved Analysis (Coming Soon): We are currently developing a feature that will allow you to save a completed analysis. You will be able to reuse this saved analysis in other conversations, which is a powerful way to transfer proven logic.

While you cannot directly move an entire conversation's context, using features like KDs allows you to maintain consistency across all your work in Petavue.

Can I still use Petavue through Claude?

Yes, you can still access Petavue through Claude's interface. However, this route uses the Claude MCP, which may have limitations like context window caps, execution errors, or downtime on Claude's side. For a seamless, end-to-end data analysis experience with flexible orchestration, we strongly recommend using the native Petavue platform.

Analyzing Data

How does an analysis work in Petavue? Every analysis follows a transparent, plan-first workflow:

1. Ask a question: You start by entering your query in plain language.
2. Review the plan: Petavue generates a clear, step-by-step analysis plan for your review.
3. Approve execution: Once you approve the plan, Petavue executes each step using the appropriate tools.
4. Get results: The findings are summarized, and you can validate the data or export it as a report.

For a detailed walkthrough, see our guide: [Performing Analysis in Petavue: A Step-by-Step Guide].

What details should I verify when reviewing a plan?

Reviewing the plan before execution is the most critical step to ensuring you get the results you want. Here is a checklist of key items to verify. (Note: This could also be a separate, more detailed guide).

• Data Preparation:
  • Source Tables: Is the plan using the correct starting tables (e.g., 'deals', 'contacts')? For later steps, is it correctly using the output from a previous step?
  • Joins: If tables are being joined, are they being connected using the correct fields?
  • Filters: Are the filters correct? Check the fields, the conditions (e.g., equals, greater than), and the values (e.g., "Close Date is after January 1, 2025").
• Calculation & Logic:
  • Grouping: Is the data being grouped correctly for aggregation (e.g., count of deals by region, sum of revenue by month)?
  • Formulas: If a metric is being calculated, is the formula accurate?
  • Key Definitions (KDs): If the plan is using a KD, does it correctly reflect the logic you intended?
  • Bucketing: If data is being segmented into buckets (e.g., deal size tiers), are the ranges and logic correct?
• Presentation:
  • Output Columns: Does the final output include all the data columns you asked for?
  • Sorting: If you requested the results to be sorted or ranked, is the plan set to order them correctly (e.g., sort by revenue descending)?

What if I want to change the plan during an analysis?

Absolutely. Petavue is designed for iterative analysis. You can edit filters, add new steps, or remove unnecessary ones from the plan at any time. If an analysis is already running, you can stop it, modify the plan in a follow-up prompt, and re-run it. Each change generates a new plan for your approval, ensuring full transparency.

Why do I see a different plan when I run the same prompt multiple times?

Petavue treats every prompt as a new request, but this doesn't mean the plans will be completely different. Instead, the system generates a directionally similar plan. You might notice minor variations, such as a different sequence of steps or subtle changes in filter conditions, especially if the prompt is high-level.

If you require consistent plans, here are a few solutions:

• Be More Specific: A high-level prompt like, "Show me deals not meeting the touch criteria," can be interpreted in slightly different ways. A more specific prompt will produce a more consistent plan. For example: "Generate a list of all deals created in 2025 that have fewer than 90 total touches."
• Use Key Definitions (KDs): If you frequently use a specific formula, metric, or sequence of steps, turn it into a KD. When you use the KD in your prompt, Petavue will always apply the exact, pre-defined logic, ensuring consistency every time.
• Review Carefully: At first glance, two plans might seem different. We recommend comparing the details side-by-side; you will likely find they are structurally and logically very similar.

We are actively working on improvements to enhance plan consistency for previously approved prompts, which will be rolled out in a future update.

Can I perform drill-downs or ask follow-up questions?

Yes. After an analysis is complete, you can ask follow-up questions like “Break this down by industry” or “Show me this data by region.” Petavue will generate a new plan for your drill-down analysis and can add the results as a new section to your existing report.

Can I see results in a table format instead of just text?

Yes, absolutely. If you want to see the underlying data in a table within the chat interface, simply ask for it. You can use prompts like, "Show me the top deals from this analysis in a table."

For performance and efficiency, the table displayed in the chat is limited to the first 25 records. To view the complete dataset with all records, click the full table link provided below the analysis step. (A screenshot showing the Petavue table link below an analysis step would be placed here).

How does Petavue handle ambiguous or conflicting questions?

Petavue uses a built-in Clarification Tool to manage ambiguity. If your instructions are unclear or conflict with previous prompts, it will automatically ask for clarification to ensure the analysis proceeds correctly. If the results still seem incorrect, you can always manually guide or correct the analysis in a follow-up prompt.

How can I trust Petavue’s analysis?

Trust is built on transparency. Every analysis in Petavue is executed based on a plan that you review and approve beforehand. To verify the process at any time, ask a question like, “How was this analysis performed?” Petavue’s Find-Out-How Tool will provide a detailed breakdown, including the plan executed, data sources used, and definitions applied.

How can I verify the results presented by Petavue?

Trust and transparency are core to Petavue. Every analysis follows the plan you approve. If a result seems unexpected or you wish to validate it, you can take the following steps:

1. Ask Follow-Up Questions: Interrogate the analysis directly in the chat to understand how a result was derived. You can ask things like:
   • "How did you calculate the 203 deals in step 2?"
   • "Can you explain in detail the steps you took for the above analysis?"
   • "How are my results different from the result in the previous analysis? I'm getting 35 decaying contacts in one and 85 in another. What changed?"
2. Inspect the Petavue Table: At the bottom of each step in an analysis, you will find a link to the corresponding Petavue table. Click this link and then select the "Find Out How" tab. This section provides detailed instructions and the exact code used for that step, allowing you to see if the execution perfectly matches the plan you approved.

If you still have questions, please don't hesitate to reach out to our support team for assistance.

What should I do if an analysis fails?

While rare, if an analysis fails mid-execution, you can easily resume it. Simply use a prompt like, “Retry the analysis from step 2.” The system will pick up where it left off. Alternatively, you can always restart the analysis from the beginning with a new plan.

What is the best practice for writing prompts if I have multiple questions?

For the most accurate and sharpest results, we strongly recommend you focus each prompt on a single, specific analysis or goal. While Petavue can maintain context across a long conversation, sending a single prompt with multiple complex and unrelated requests can lead to an overly complicated plan and reduce accuracy.

Best Practice for Complex Workflows:

Imagine you want to analyze leads, deals, and decaying contacts all at once. Instead of combining them into one large prompt like this:

"Give me a detailed analysis of all leads this month, subgrouped by week with start/end dates, and broken down by source and stage with a full contact list for each week. Also, give me a count of all converted deals this month, subgrouped by week with deal names and company details. Finally, generate a report of decaying leads between 2 weeks and 90 days old with no engagement, grouped by their creation month and week, and tell me the date range you used."

This approach can be difficult for the system to plan and for you to verify. The best practice is to break it down and iterate:

1. First Prompt (Leads): Start with your first goal. "Show me a count of all leads created this month until yesterday. Group them by week, showing the start and end date for each week, and break the count down by lead source and lifecycle stage."
2. Review and Execute: Review the plan for the lead analysis. Once you approve it and get the results, move on.
3. Second Prompt (Deals): Now ask about deals. "Using the same weekly groupings, now show me a count of all converted deals for this month until yesterday. Provide a list of the deal names and associated companies for each week."
4. Third Prompt (Decaying Leads): Finally, tackle the last part. "Generate a report of leads that are between 2 weeks and 90 days old and have had no engagement activities or associated deals. Group these decaying leads by their creation month and week."

By keeping each question focused on one core analysis, you ensure Petavue generates a concise, accurate plan, making the results easier to verify and the entire workflow more transparent and efficient.

Reports & Insights

Can I create dashboards?

Not at this time. Currently, Petavue generates document-style executive summaries that can be exported to PDF. Visually rich, HTML-based dashboards are on our product roadmap.

How do I create a report after an analysis?

Simply ask: “Create an executive summary report from this analysis.” Petavue will generate a structured report summarizing the key insights, recommendations, and supporting data. For more details, see [Creating an Artifact Report in Petavue].

What types of reports can I create for my executives?

• Currently, Petavue generates executive summary-style reports formatted in markdown. These are ideal for clear, document-based insights. Once created, these reports can be downloaded as a PDF file. Please note that other formats, such as Microsoft Word, are not supported at this time.
• We are actively developing a dashboard feature, which will allow you to create reports with interactive visuals and widgets. This functionality is a key part of our upcoming roadmap.

Can I schedule reports or set up alerts?

This feature is not yet available. Today, all reports are generated on demand and can be exported as PDFs. Automated scheduling and playbook-driven alerts are a key part of our upcoming roadmap. Soon, you’ll be able to set up triggers (e.g., “Alert me in Slack when our renewal pipeline drops by 10% month-over-month”) that deliver insights directly to your preferred tools.

Data, Definitions & Metrics

What data sources can I connect to Petavue?

Our list of native integrations is growing.

• Currently Supported: HubSpot, Salesforce.
• Coming Soon: Google Sheets, Google Analytics, Sprinto, Gainsight, Marketo, Mixpanel, and Snowflake.

If you need to connect to a source not listed here, please contact us at support@petavue.com.

How can I check if my data is ready for analysis?

You can assess your data's health in the Data Hub → Dictionary, which shows column fill rates, duplicates, and missing values. You can also ask questions directly in the chat, such as:

• “What percentage of deals have close dates?”
• “Which tables have the most null values?”

Can Petavue verify its results against a report in my CRM?

• Petavue connects directly to your CRM's underlying data objects (like deals, contacts, etc.) to perform its analysis. However, it does not have access to the pre-built reports, dashboards, or saved views that exist within your CRM platform (e.g., a Salesforce or HubSpot report). Because of this, Petavue cannot perform a direct comparison between its findings and a specific CRM report.
• Furthermore, you may notice that the results from a Petavue analysis do not perfectly match the numbers you see in a similar report within your CRM. This is often because CRMs apply their own inherent logic or built-in filters that are not always apparent in the standard user interface. Since Petavue analyzes the raw data directly, it will not include this hidden, CRM-specific logic unless you explicitly define it in your analysis plan.
• Any analysis or verification must be done based on the raw data available in the CRM's objects.

Can I create my own metrics?

Yes, you can define custom metrics, known as Key Definitions (KDs), directly from the chat interface. This allows you to create business metrics tailored to your specific needs. To learn more, read our guide on [Link: Creating Key Definitions].

Can I edit my custom metrics?

Currently, editing a metric requires you to delete the existing Key Definition (KD) and create a new one with the updated formula. Direct editing of KDs via chat is a feature we plan to release shortly.

How do I update or delete a Key Definition (KD)?

Currently, you can create new Key Definitions (KDs) directly through chat. For a step-by-step guide, please see our article on Understanding Key Definitions.

To edit or update the logic of an existing KD, you must first delete it in the Petavue UI and then recreate it. Here is the process:

1. Navigate to the UI and delete the KD you wish to update.
2. Run the analysis again with the updated formula or logic in your prompt.
3. Once the plan is executed successfully, you can save the new logic as a KD.

While the logic of a KD must be recreated, you can edit the KD's name and description directly in the UI at any time.

Deleting a KD must also be done in the UI. Full support for editing and deleting KDs directly via chat is on our roadmap and will be available in a future update.

How do I configure my company's fiscal quarter?

Standard calendar quarters (Jan-Mar, Apr-Jun, etc.) are used by default. However, we understand that many companies operate on a different fiscal calendar. If your reporting requires a custom fiscal quarter, please contact our support team on Slack or by emailing support@petavue.com. We will be happy to configure your Petavue agent to use your company's specific fiscal logic for all time-based calculations.

Security

How does Petavue ensure data privacy and security?

We adhere to strict security protocols to protect your data, including:

• Isolated data environments for each customer.
• End-to-end encryption for data in transit and at rest.
• Access-controlled connections to your data sources.

• Automated data hygiene assessments to identify potential quality issues.

  
  

Troubleshooting & Support

How do I report an issue or share feedback?

We'd love to hear from you! You can reach our support team by emailing support@petavue.com or by using the “Feedback” option directly within the platform.