Chat with Your Data

The Chat interface lets you ask questions about your uploaded files in plain English. No SQL or coding knowledge required.

Starting a Conversation

1. Open a workspace with uploaded files
2. Click + New in the sidebar to start a conversation
3. Type a question in the input box
4. Press Enter or click the send button

Selecting Tables

Below the chat input, you'll see a collapsible dataset selector showing how many tables are active (e.g., "2 of 3 tables selected"). Only tabular datasets appear here—documents (PDFs, Word files, etc.) are automatically included for search without needing selection.

Use the checkboxes to pick which tables the AI should query. You can Select all or Deselect all (at least one table always stays selected). If your workspace has tables but none are selected, the input placeholder changes to "Select at least one table to query..." and sending is disabled.

Attaching Files

You can upload files directly from the chat without switching to the Documents section:

• Drag and drop files onto the chat input area
• Paste files from your clipboard

Attached files appear as pills below the input showing their upload status. You can remove a file before sending by clicking the × on its pill. Once you send the message, files upload to the workspace and become available as datasets—they'll appear in the Documents section and the dataset selector.

Supported file types: CSV, Excel, Parquet, PDF, Word, PowerPoint, TXT, and Markdown.

How It Works

When you ask a question:

1. For tabular data — ParseSphere generates SQL queries and runs them
2. For documents — It searches for relevant passages using AI
3. For mixed workspaces — It automatically picks the best approach

You'll see the answer in plain language, often with data tables or charts.

Example Questions

For Tabular Data

• "What are the top 5 products by revenue?"
• "Show me sales by region for Q4"
• "How many customers made purchases last month?"
• "What's the average order value?"
• "Compare this year to last year"

For Documents

• "What are the key terms in this contract?"
• "Summarize the main findings"
• "What does it say about pricing?"
• "Find information about delivery timelines"

Follow-up Questions

The AI remembers your conversation context. Ask follow-ups naturally:

1. You: "What are the top 5 products by revenue?"
2. AI: Shows results
3. You: "Show me just Electronics"
4. AI: Filters to Electronics category

Other examples:

• "Break that down by region"
• "What about Q4?"
• "Show the trend over time"
• "Exclude returns"

Understanding Responses

Data Tables

When your question involves tabular data, you'll see:

• Answer text — Plain language explanation
• SQL query — Expandable block showing the generated query and row count. Click View Results to open the full table
• Table Results — Paginated table with search, filter, and a SQL tab with syntax highlighting
• Export options — Copy the data, download as CSV, or Save as Dataset to create a new queryable dataset from the results

Charts

For trend or comparison questions, ParseSphere may show:

• Line charts for time series
• Bar charts for comparisons
• The underlying data is always available

Document Sources

For document queries, you'll see:

• Answer — Synthesized response
• Sources — Expandable section grouped by document. Each source shows the filename, page number, a confidence badge (High, Medium, Low, or Related), and a text preview. Image sources include a thumbnail with a lightbox for full-size viewing

Data Transformations

ParseSphere can modify your datasets directly from the chat. Ask for changes in plain English:

• "Add a profit margin column calculated from revenue and cost"
• "Filter out rows where status is 'cancelled'"
• "Merge this table with the customers table by customer_id"
• "Rename the 'amt' column to 'amount_usd'"
• "Create a new dataset with only Q4 transactions"

When the AI generates a transformation, you'll see a preview card before anything changes:

• Diff summary — How many rows and columns were added, removed, or changed
• Warnings — Potential issues like data type mismatches or null values
• Before & after samples — Click Preview to compare the data side by side

Click Accept to apply the change (this creates a new version of your dataset), or Reject to discard it. Your original data is never modified — every transformation creates a new version you can roll back to later.

Reasoning

For complex questions, the AI shows its step-by-step thinking in a collapsible Reasoning block. Click it to expand and see how the AI arrived at its answer—what data it looked at, what queries it considered, and why it chose a particular approach.

Feedback

Each assistant message has thumbs up and thumbs down buttons. Use these to rate response quality—it helps improve future answers.

Credits

Every chat message consumes credits based on the tokens processed.

Where to see costs:

• Per message — Each assistant message shows its credit cost inline, next to the message actions
• Per conversation — The conversation header displays the total credits used across all messages

Click either credit display to expand a breakdown showing input credits, output credits, and cache credits. Input credits cover what you send to the AI (your question plus context from your data). Output credits cover the AI's response. Cache credits reflect savings when the AI reuses context from earlier in the conversation.

Managing Conversations

Create New Conversation

Click + New in the sidebar to start fresh.

Rename Conversation

1. Hover over a conversation in the sidebar and click the three-dot menu
2. Select Rename and enter a descriptive name

You can also click the pencil icon in the conversation header to rename the active conversation.

Delete Conversation

1. Hover over a conversation in the sidebar and click the three-dot menu
2. Select Delete

Tips for Better Results

1. Be specific — "Show Q4 revenue by product category" beats "show sales"
2. Reference column names — If you know your CSV has product_category, use that term
3. Start simple — Ask a basic question first, then drill down
4. Use follow-ups — Build on previous answers instead of starting over

What's Next?

• Document Upload — Add more files to analyze
• Parses — Parse documents without workspaces