> ## Documentation Index
> Fetch the complete documentation index at: https://docs.alysium.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Control how your agent uses its knowledge

> Choose how closely an agent's answers follow attached knowledge and what it should say when the sources do not answer a visitor's question.

Choose how closely your agent's answers should follow its sources and what it should say when information is missing.

Attach useful knowledge first. The Behavior section appears after the agent has a file, an attached folder or project, or a website source. These settings require an agent you can edit.

## Open Behavior

1. Open [Home](https://app.alysium.ai/spaces).
2. Select the intended agent.
3. Open **Build** → **Settings**.
4. Find **Knowledge**.
5. Open **Behavior**.

If Behavior is already open, continue with [Choose the answer boundary](#choose-the-answer-boundary). The collapsed summary shows the current grounding, whether knowledge is always consulted and whether the missing-answer reply is custom.

## Choose the answer boundary

Under **Answer grounding**, choose for the agent's job:

| Choice | Intended behavior |
| - | - |
| **Strict** | Answer only from the agent's knowledge and use a not-found response for gaps. |
| **Balanced** | Prefer the knowledge and identify added general information. |
| **Open** | Use the knowledge alongside general information. |

For a customer FAQ that must not invent business policies, start with **Strict**. A brainstorming agent may suit Open or Balanced. The label describes the intended answer boundary; it does not add missing facts or guarantee correctness.

## Decide whether to search first

**Always consult knowledge before answering** tells the agent to search attached knowledge first on every factual question. Turn it on when your own notes should be the first place it looks.

This is separate from grounding. Strict defines the intended answer boundary; always-consult controls search-first behavior. Neither makes an unreadable or unattached source usable.

## Write the missing-answer reply

Use **When the answer isn't in my knowledge…** for a knowledge gap. For a fictional shop:

> I do not have that detail in the shop notes. Please contact the shop team for the current policy.

Replace the business and contact route with your own. Keep the wording within the displayed character limit. Do not claim the agent emailed a person or created a ticket.

## Apply and save a factual-FAQ setup

For an agent whose answers should stay within its supplied FAQ:

1. Under **Answer grounding**, select **Strict**.
2. Turn on **Always consult knowledge before answering**.
3. Enter your missing-answer wording.
4. Pause to let the settings save automatically.

There is no separate Design **Save Changes** step for these choices.

To check persistence, return to [Home](https://app.alysium.ai/spaces), reopen the same agent and check **Build** → **Settings** → **Knowledge** → **Behavior**.

## Try a known fact and a gap

Use the chat beside the Build settings, or **Chat Preview** within Build on a narrow screen.

1. Enter a question your actual source answers.
2. Select **Send message**, or press Enter.
3. Compare the full reply with the source.
4. Enter a question the source does not answer.
5. Select **Send message**, or press Enter.
6. Check how the reply handles the missing information.

You are checking that the instructions and knowledge fit the task, not guaranteeing future replies. [Test your agent while you build](https://docs.alysium.ai/build-your-agent/test-your-agent-while-you-build.md) explains this comparison.

## If the behavior does not match

* **A supported fact is missing:** check the exact attachment and its **Ready** or **Saved** state before changing grounding.
* **A policy is invented:** confirm the intended grounding and check **Agent Instructions** for a conflicting rule.
* **Behavior is absent:** check whether a source has been attached or whether the last source was removed.

Let any correction save, then try the question again. Continue with [Why your agent can't find something](https://docs.alysium.ai/knowledge-and-library/why-your-agent-cant-find-something.md) for source-specific checks.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.