AutoLearn

What it is

Native tool (nicknamed AutoLearn) that automatically logs questions the agent couldn’t answer using the knowledge base. Each question becomes a learning opportunity that appears under Knowledge Bases → AutoLearn for you to review, dismiss, or approve — and when you approve, the content goes into the knowledge base and the agent answers better afterwards.

It’s the feedback loop that turns your RAG agent into something that improves on its own with use.

When to use

• Whenever you already use Query Knowledge Base and want to close the continuous improvement loop.
• To discover what’s missing in your base without having to read hundreds of conversations.
• In early phases of a new agent, while still calibrating base content.
• To spot patterns: frequent questions nobody covered in the docs.

Prerequisite

This tool only works when Query Knowledge Base is also active with at least one base selected. Without that, AutoLearn has nothing to compare against — it operates by detecting cases where the search ran but didn’t return useful results.

If you enable AutoLearn without the search configured, the form shows an explicit warning.

Configuration

1. In the agent’s Tools tab, ensure Query Knowledge Base is active with at least one base.
2. Click + Add Tool.
3. Under NATIVE TOOLS, pick AutoLearn.

No extra parameters. The tool is active immediately.

How the agent invokes

The tool is called automatically by the agent when it perceives that:

• the user asked a question within the agent’s scope;
• the base search didn’t bring enough content to answer well;
• this gap is worth logging for future review.

The user sees no difference in the response — the tool works in the background, creating the gap record while the agent keeps trying to answer with what it has.

Reviewing opportunities

Logged opportunities appear under Knowledge Bases → AutoLearn. For each one, you can:

• Dismiss: if the gap isn’t useful or is a duplicate.
• Review and include: the system suggests a target base; you review the text and add it to the knowledge base.

As you approve opportunities, the base grows — and the agent answers with more accuracy in future conversations on the same theme.

Limitations and notes

• Fully depends on search being active. Without Query Knowledge Base configured, AutoLearn does nothing.
• Not infallible. The model may fail to call the tool (misses a real gap) or call it without need (creates a disposable record). Manual review filters those cases.
• Check the queue regularly. AutoLearn’s value lies in you reviewing the opportunities — if they pile up without analysis, the improvement loop doesn’t happen.