Skip to main content

Conversations

Conversations are a standalone chat interface for querying your data. Ask questions in natural language, and the AI agent writes and runs SQL, builds charts, and joins data across databases — all within a threaded chat.

Starting a Conversation​

  1. Click New Conversation from the sidebar
  2. Give it a title (or let the agent suggest one)
  3. Choose visibility: Workspace (visible to your team) or Private
  4. Select the initial data context — which connections and tables the agent can access

Chat Interface​

The conversation interface is a message thread:

  • Your messages appear on the left with your avatar
  • Agent responses stream in real time with tool call cards showing what the agent is doing
  • The prompt input at the bottom supports natural language questions

Example questions:

  • "Show me the top 10 customers by revenue"
  • "What's the month-over-month growth rate?"
  • "Join users from production with orders from analytics and find average order value"
  • "Create a chart showing sales by region over time"

SQL Queries​

When the agent needs data, it writes and executes SQL using the RUN_SQL tool:

  • Queries run directly against your connected databases
  • Results appear inline as expandable tables with row counts and timing
  • Click to expand and sort results in a modal view
  • All queries are SELECT-only by default

Derived Queries​

Derived queries let the agent combine results from different databases — a key differentiator for teams with data spread across multiple sources.

How it works:

  1. The agent runs queries against individual databases (e.g., customers from Postgres, orders from MySQL)
  2. Results are loaded into DuckDB in-memory
  3. A derived query joins, filters, or aggregates across those result sets

Capabilities:

  • Join data from any combination of connected databases
  • Standard SQL with window functions, CTEs, and aggregations
  • Composable — derived queries can reference other derived queries
  • No ETL pipelines or data warehouses required

Charts​

The agent can create inline charts from query results:

  • Bar, line, area, and pie charts
  • Configured from the data automatically
  • Always reflects the latest query results
  • Expand to full-screen for presentation

Data Access Controls​

Control which data the agent can see per conversation:

  • Select connections — choose which databases are available
  • Filter tables — restrict to specific tables within a connection
  • Change mid-conversation — adjust the data context as your questions evolve

The agent only sees the connections and tables you allow. This prevents accidental exposure of sensitive data and keeps the agent focused on relevant sources.

Sharing​

Conversations support two visibility levels:

  • Workspace — visible to all workspace members
  • Private — only visible to you

Share Dialog​

Share a conversation with specific team members:

  • Owner — full control, can delete the conversation
  • Observer — can view and follow along, cannot send messages

Tool Approvals​

For sensitive operations, the agent asks for approval before executing:

  • Running SQL — shows the query that will execute
  • Derived queries — shows the join/aggregation query
  • Chart creation — shows the chart configuration

Click Approve to proceed or Reject to cancel. This gives you control over what the agent does with your data.

Observer Mode​

When someone shares a conversation with you as an observer, you can watch their session in real time:

  • See messages and agent responses as they stream
  • Watch tool calls execute live
  • Follow the analysis without interrupting the flow

Observer mode is useful for onboarding, pair analysis, or reviewing how a colleague approaches a data question.