Skip to main content
Marking a run as a reference sets the standard for how your agent should behave, allowing the system to learn from its most successful outputs. When an agent execution meets your quality requirements, you can validate it to ensure consistency over time and prevent performance regressions. These historical examples provide the agent with a proven foundation to build on, replicating SQL patterns and formatting styles that already work. This guide explains how to mark runs as reviewed, how the context injection works, and the benefits for your automated workflows.

​
How it works

The process for using reviewed runs as a reference follows three main stages to ensure your agents maintain high standards over time.

Mark as Reviewed

After an agent completes a successful run, navigate to the Previous Runs list. Click the Mark as reviewed button on the specific execution that produced the desired result.

Context Storage

WisdomAI stores the full context of that validated run, including the specific steps taken, the generated SQL, and the exact output formatting used.

Reference Injection

On future executions, WisdomAI automatically retrieves up to 3 of the most recently reviewed runs. These are injected as historical reference examples before the agent begins its new task, guiding it to follow the established logic and style.
Reviewed Run Agent
This feature acts as a regression guard, ensuring that once you have perfected an agent’s behavior, that quality is preserved across future updates and similar queries.

​
Key capabilities

Reviewed runs provide the following technical advantages for maintaining agent performance:
  • Manual Validation: Choose exactly which success-status runs to promote as references via the Mark as reviewed button.
  • Contextual Injection: Automatically utilize up to 3 recently reviewed runs per execution for historical guidance.
  • Logic Consistency: Improve the reliability of SQL patterns and tool call sequencing by providing a “gold standard” for the agent to follow.
  • Formatting Control: Ensure the agent maintains a consistent output structure, which is critical for teams running recurring weekly or monthly reports.
  • Reversibility: Remove the reviewed status at any time to update your reference library and ensure only the best examples are used.
Reviewed Run Agent Remove

​
Target users

The reviewed runs feature is specifically designed for:
  • Agent owners & builders who want to lock in high-quality performance once an agent run meets their exact requirements.
  • Data teams that generate recurring weekly or monthly reports and need the output structure to remain identical across every execution.

​
FAQs

Only runs with a Completed status and an associated conversation can be marked as reviewed.
Yes. You can remove the reviewed status at any time. Once unmarked, that specific run will no longer be used as a reference for future agent tasks.
WisdomAI uses up to the 3 most recent reviewed runs to provide context for a new execution.

​
Next steps

Create an Agent

Learn the basics of setting up a new AI agent for your data.

Manage Your Agents

Explore how to monitor performance and view the Previous Runs list.