Skip to main content
Golden Assets are the primary mechanism for teaching the agent how your team actually answers questions. While the system prompt encodes general rules and the data model encodes structure, Golden Assets encode patterns. The concept is a simple chain: a recurring question your team gets asked → the confirmed, trusted SQL that answers it correctly → and optionally the chart that presents it best. At chat time, the agent retrieves Golden Assets via RAG (Retrieval-Augmented Generation): when a user asks something, it searches your library for semantically similar questions and uses the matches as reference while writing its own SQL and choosing a visualization. The more Golden Assets you add, the more consistent the agent becomes — each one is a demonstration of correct behavior the agent can generalize from.
Golden Assets panel showing the Golden Queries and Golden Charts tabs

​
Golden queries

A Golden Query is a natural language question paired with the correct SQL that answers it. When a user asks a question, the agent searches your Golden Query library for semantically similar examples and uses them as reference when writing its own SQL — reducing errors and improving consistency on the question types your team cares about most.
Golden Queries tab showing an example query with its SQL
Adding a Golden Query
  1. Open your agent and go to the Assets tab in the configuration panel
  2. Select Golden Queries
  3. Click + Add Query
  4. Enter the question as a user would phrase it (e.g. “Show sales by owner segmented by product”)
  5. Enter the correct SQL that answers it
  6. Save
The question field is used for semantic similarity matching — write it the way a real user would ask, not as a technical specification. The SQL field should be the exact query your team would run, including any business logic, filters, or aggregations that make it correct for your data model. What makes a good Golden Query Good Golden Queries cover your highest-frequency question patterns and the cases where getting the SQL wrong would be most costly. Prioritize:
  • Questions with non-obvious join logic or filter requirements specific to your schema
  • KPI calculations that have a precise definition (e.g. “revenue” as a specific formula, not just a column name)
  • Questions where the correct answer depends on business rules the agent can’t infer from column names alone
  • Common questions that appeared in your evaluation suite and the agent previously got wrong
A library of 15–25 well-chosen Golden Queries covering your core use cases will have a dramatically larger impact on accuracy than 100 generic ones.

​
Golden charts

A Golden Chart pairs a question and SQL with a specific chart configuration — the visualization type, axis mappings, and formatting settings that produce the right output for that question. When the agent answers a similar question, it uses the Golden Chart as a reference for how to render the result.
Golden Charts tab showing an example chart configuration
Adding a Golden Chart
  1. Open your agent and go to the Assets tab
  2. Select Golden Charts
  3. Click + Add Chart
  4. Enter the question, the SQL, and configure the chart type and axis settings
  5. Save
Golden Charts are most valuable when the correct visualization for a question isn’t obvious from the data shape alone — for example, when a question about revenue trends should always use a line chart with a specific date grouping, or when a comparison question should use a grouped bar chart rather than a stacked one.
Golden Queries and Golden Charts are retrieved based on semantic similarity to the user’s question, not keyword matching. Write the question field in natural language — the same way a business user would ask it — to maximize the quality of retrieval.

​
Next steps

System Prompts

Encode business rules and behavioral guardrails that apply across all queries.

Skills

Turn repeated procedures into reusable know-how the agent loads on demand.

Knowledge Base

Store the reference documents Golden Asset SQL assumptions are based on.

User Memory

Let the agent adapt outputs to each end user’s preferences.

Agents

See the full agent configuration including evaluations and versioning.

MCP as Context Source

Connect external tools and live systems as additional agent context.