Skip to main content

​
What is the AI Assistant?

The AI Assistant is an interactive chat interface that lets you test your agent in real-time, debug issues, and iterate on your workflow before automating it.

Key Difference

Agent = Automated execution (runs on schedule or trigger)AI Assistant = Interactive testing (you chat with the agent)

​
Where to Find It

The AI Assistant panel appears on the right side of your agent page.
1

Open Your Agent

Click any agent from your home page
2

Look Right

The AI Assistant chat panel is on the right side of the screen
3

Start Chatting

Type a message and hit Enter to interact with your agent
You can collapse/expand the AI Assistant panel with the toggle button in the top-right.

​
How It Works

The AI Assistant uses the same AI engine as your automated agent, with access to:
  • Your agent’s instructions
  • All uploaded sources
  • Your worksheet data
  • Connected integrations
The difference: Instead of running automatically, you control it through conversation.

​
Testing Your Agent

Use the AI Assistant to test your agent before going live.

​
Basic Testing Workflow

1

Prepare Test Data

Add a few sample rows to your worksheet (e.g., 3-5 test cases)
2

Give a Command

In the AI Assistant, type:
Process the first 3 rows in the worksheet
3

Watch It Work

The AI Assistant will:
  • Read the instructions
  • Search relevant sources
  • Process the data
  • Show its reasoning step-by-step
  • Update the worksheet
4

Review Results

Switch to the Worksheet tab to see the output
5

Iterate

If results aren’t perfect, refine instructions and test again

​
Example Test Commands

Simple test:
Generate a quote for Acme Corp.
They need 10x Product A and 5x Product B.
Apply a 10% discount.
Batch test:
Process all rows in the "Quote Requests" sheet
where Status = "Pending"
Edge case test:
What happens if a customer requests a product
that's not in the price list?

​
Debugging with AI Assistant

When your agent doesn’t behave as expected, the AI Assistant helps you debug.

​
Common Debugging Scenarios

Ask:
What sources do you have access to?
Can you show me what's in the "Price List" source?
The assistant will:
  • List all available sources
  • Confirm they’re indexed
  • Show sample content from the source
If source is missing:
  • Go to Sources tab
  • Upload the missing document
  • Wait for indexing to complete
  • Test again
Ask:
Show me exactly what you would write to row 10
of the worksheet, but don't write it yet.
The assistant will:
  • Show the exact values for each column
  • Explain its reasoning
  • Let you verify before actually writing
Fix format issues:
  • Add examples to instructions
  • Specify exact column names
  • Show desired output format
Ask:
Can you explain your understanding of your job?
What are you supposed to do?
The assistant will:
  • Summarize its understanding
  • Show what it thinks it should do
  • Reveal any misinterpretations
Refine instructions:
  • Clarify ambiguous parts
  • Add specific examples
  • Break complex steps into simpler ones
Ask:
Can you see the data in the "Customers" sheet?
Show me the first 3 rows.
The assistant will:
  • Confirm if it can access the sheet
  • Show sample data
  • Reveal any access issues
Common fixes:
  • Check sheet name spelling in instructions
  • Verify worksheet is saved
  • Refresh the page if data was just added

​
Interactive Iteration

The AI Assistant excels at iterative refinement.

​
Iteration Pattern

1

Initial Attempt

Generate a quote for this customer
Reviews output - not quite right
2

Refine Request

Actually, can you format the quote like the example
in the "Quote Template.pdf" source?
Better, but missing something
3

Add Details

Also include a breakdown showing:
- Subtotal
- Discount amount
- Tax
- Grand total
Perfect!
4

Update Instructions

Copy the working approach into your agent instructions

​
Example Iteration Session

You: 
Process the first customer request
AI Assistant: 
I've generated a quote for Acme Corp:
- 10x Widget A at $50 = $500
- Total: $500

Added to row 10 of the Quotes sheet.
You: 
Wait, you forgot to apply the 10% discount they requested
AI Assistant: 
You're right! Let me recalculate:
- Subtotal: $500
- 10% discount: -$50
- Total: $450

Should I update row 10 with the corrected total?
You: 
Yes, and also show the discount as a separate line item
AI Assistant: 
Updated row 10:
- Subtotal: $500
- Discount: -$50 (10%)
- Total: $450

I'll remember to always show discounts separately.

​
Advanced Features

​
Multi-Turn Conversations

The AI Assistant maintains context across messages: 
You: "Find all customers with pending quotes"
AI: "Found 5 customers: Acme Corp, TechCo..."

You: "Generate quotes for the first 3"
AI: "Processing Acme Corp, TechCo, StartupXYZ..."

You: "Send the quotes via email"
AI: "Sending emails via Gmail integration..."

​
Ask Questions

Get clarification on your workflow: 
What should I do if a product is out of stock?

How do I handle customers who request custom pricing?

What information do you need to generate a quote?

​
Dry Runs

Test without making changes: 
Show me what would happen if I processed all pending rows,
but don't actually write anything to the worksheet yet.
The AI Assistant will simulate the execution and show you the results before committing.

​
Best Practices

Start small: Test with 1-2 rows before processing hundreds
Use specific commands: “Process row 5” is clearer than “do the thing”
Ask for explanations: “Why did you choose that value?” helps debug
Iterate incrementally: Make one change at a time, test, repeat
Save working approaches: When something works well, copy it into your instructions
Use dry runs: Preview results before actually writing to worksheets

​
AI Assistant vs. Automated Runs

Understanding when to use each:
FeatureAI AssistantAutomated Run
ControlInteractive, step-by-stepFully automated
SpeedManual, real-timeScheduled or triggered
Testing✅ Perfect for testing❌ Not for testing
Debugging✅ See reasoning liveView logs after
Iteration✅ Instant feedbackRun, review, modify, repeat
Scale❌ Manual for each task✅ Processes batches
Production❌ Not for production✅ For production
Workflow:
  1. Build and test with AI Assistant
  2. Once working, automate with Runs
  3. Debug issues with AI Assistant
  4. Optimize and return to Runs

​
Common Use Cases

Before your agent runs automatically, test it thoroughly:
  1. Test with sample data: 3-5 representative rows
  2. Test edge cases: Missing data, unusual values, errors
  3. Test integrations: Verify emails send, CRM updates work
  4. Verify output: Check worksheet updates are correct
  5. Refine instructions: Based on test results
When an automated run fails:
  1. Review run logs to see what went wrong
  2. Recreate the scenario in AI Assistant
  3. Ask questions: “Why did you fail on row 10?”
  4. Test fixes: Try different approaches interactively
  5. Update instructions once working
  6. Re-run the automated agent
For tasks you only need to do once:
Generate a summary report of all quotes from Q4 2024

Find all invoices from vendor "Acme Corp" and calculate total spent

Enrich the top 10 leads with industry and employee count
No need to create a fully automated agent - just chat with the AI Assistant!
Explore your data interactively:
What's the average order value in the worksheet?

Show me the top 5 customers by revenue

Are there any patterns in which products are frequently ordered together?
The AI Assistant can analyze and summarize your data on the fly.
Use the AI Assistant to help write better instructions:
I want to create an agent that processes invoices.
What information should I include in the instructions?

Can you suggest a good structure for my invoice processing workflow?

What edge cases should I account for?

​
Tips & Tricks

​
Get Better Responses

Be Specific

Instead of: “Do the thing”Try: “Generate a quote for the customer in row 5”

Ask Why

Instead of: Accepting wrong outputTry: “Why did you calculate the total as $500?”

Show Examples

Instead of: Vague descriptionsTry: “Format it like this: [example]”

Test Edge Cases

Instead of: Only happy pathsTry: “What if the customer field is empty?”

​
Conversation Starters

Not sure what to ask? Try these: Understanding:
  • “Explain your job in your own words”
  • “What are the main steps in your workflow?”
  • “What data do you have access to?”
Testing:
  • “Process the first row and show your work”
  • “What would you do in this scenario: [describe edge case]”
  • “Show me sample output before writing to the worksheet”
Debugging:
  • “Why did the last run fail?”
  • “What sources did you use for that decision?”
  • “Can you walk me through your reasoning?”
Optimization:
  • “How can I improve these instructions?”
  • “What’s the fastest way to process 1,000 rows?”
  • “Are there any bottlenecks in the workflow?”

​
Limitations

The AI Assistant has the same limitations as your automated agent:
  • Can only access uploaded sources (can’t browse the web unless you add links)
  • Respects your integration permissions
  • Limited by rate limits of external APIs
  • Can’t execute code or install packages

​
What AI Assistant CAN’T Do

❌ Run on a schedule (use automated runs for that) ❌ Process data while you’re away (use triggers) ❌ Handle thousands of items efficiently (use batch runs) ❌ Make changes without your permission (you control each step)

​
What AI Assistant CAN Do

✅ Test workflows before automation ✅ Debug failed runs interactively ✅ Explore and analyze data ✅ Answer questions about your workflow ✅ Generate one-off outputs ✅ Help refine instructions

​
Next Steps

Working with Agents

Learn how to create and configure agents

Monitoring Runs

Understand automated execution and logs

Quickstart

Build your first agent with AI Assistant

FAQ

Common questions and troubleshooting