Multi-Turn Context

Grepp is built for back-and-forth conversations. You don't need to start fresh for every question — the agent remembers what you've discussed, what queries have been run, and what's on your dashboard, so each message can build on the last.

Refining queries

After the agent runs a query or builds a chart, you can refine it naturally:

• "Filter that to just the last 30 days"
• "Group by product category instead"
• "Add a where clause for active users only"

The agent understands these as modifications to the previous result, not new standalone requests.

Building on earlier results

As your conversation develops, the agent keeps track of everything created or changed:

• Charts and their underlying queries
• Dashboards and their components
• Filters, date ranges, and groupings you've applied

This means you can say "save that to my dashboard" after a long back-and-forth, and the agent knows exactly which chart you mean.

Dashboard-aware responses

The agent always knows the current state of your dashboard — which components exist, what queries power them, and how they're arranged. This lets you make changes like:

• "Update the revenue chart to use a line chart"
• "Move the user growth chart to the top"
• "Delete the old funnel chart"

The agent finds the right component without you having to specify IDs or exact names.

Confirming destructive actions

Before irreversible actions — like deleting a dashboard — the agent will ask you to confirm:

The agent waits for your explicit confirmation before proceeding.

Looking back in the conversation

The agent can also search back through your conversation history when you reference something from earlier:

• "What was the SQL we used for the revenue chart?"
• "Can you clone the chart we built at the start of this session?"
• "Remind me what tables we've queried"

See Recall for more on how the agent retrieves past work.