Skip to main content
Agent memory can be modified by admins and developers only
AI agents can now learn from your corrections and feedback. When you correct an agent’s response or guide it to better understand your data, the agent can automatically update its own instructions to remember these preferences for future conversations. The memory will only be saved if you approve the agent’s suggested learning.
Memory is for things the agent picks up over time from your corrections. If you already know what context the agent needs — a glossary, fiscal calendar conventions, definitions of internal terms — it’s faster to give it upfront via knowledge documents than to wait for it to learn through feedback.
Memory vs. Reviews: Memory captures corrections per-agent as they happen in chat — it’s reactive and lives in the agent’s instructions. Reviews does the boring work for you: it scans agent turns in the background, groups likely-wrong answers by root cause, and proposes one-click fixes to your semantic layer or to a shared lightdash.project_context.yml file that every agent reads. Use memory for in-the-moment course corrections; use Reviews when you want admins and developers to improve agents at the source.

​
How it works

  • When an agent makes a mistake or you provide clarification, it can capture this feedback
  • The agent updates its instructions field with the new learning
  • All future conversations with that agent will benefit from this accumulated knowledge
  • Memories are stored directly in the agent’s instructions, which you can view and edit in agent settings

​
What agents can learn

Agents can learn various types of corrections and preferences:
  • Which tables or explores to use for specific types of queries
  • Field selection preferences like “always use net_revenue instead of gross_revenue when generating revenue charts”
  • Filter logic such as “exclude test accounts when counting customers”
  • Chart type preferences like “use bar charts for categorical comparisons” or “prefer tables for detailed breakdowns over scatter plots”
  • Complex chart patterns that require specific guidance, such as period-over-period analysis where you teach the agent what the expected output should look like
  • General preferences about formatting, ordering, or analysis approaches

​
Providing feedback on responses

AI agents can improve over time with your feedback. After receiving an answer from your agent, you can provide feedback to help improve future responses: How to provide feedback:
  1. Look for the thumbs up/down icons on any agent response
  2. Click thumbs down if the answer wasn’t helpful or accurate
  3. Optionally provide detailed feedback explaining what went wrong
  4. The agent will use your feedback when you follow up with additional questions in the same conversation
Feedback interface in Lightdash web application
Your feedback is valuable for:
  • Immediate improvement - When you follow up after providing feedback, the agent uses your feedback to adjust its next response
  • Identifying common issues - Helps spot patterns in misunderstandings or incorrect interpretations
  • Long-term learning - Improves the agent’s understanding of your data and business context over time
The more specific your feedback, the more helpful it is. Instead of just clicking thumbs down, try to explain what was wrong (e.g., “Used the wrong metric”, “Chart type doesn’t match the question”, “Missing important filter”).