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

# Knowledge Base Overview

> Learn how knowledge sources help your Agentsy AI assistant answer questions using your expertise, methodologies, and reference materials.

# Knowledge Base Overview

Your AI Experience becomes more useful when it can draw on the knowledge that makes your business or expertise distinctive.

In Agentsy, you can provide reference materials for your AI assistant to use when answering visitor questions. These might include your methodology, educational resources, service descriptions, frequently asked questions, or other approved business information.

A knowledge base helps the assistant provide answers grounded in those materials instead of relying only on the AI model's general knowledge.

> [!NOTE]
>
> A knowledge base improves the information available to an assistant, but it does not guarantee that every answer will be accurate. Review your sources and test the assistant regularly.

## How knowledge works in an AI Experience

Think of your AI Experience as three connected parts:

| Component | What it does | Example |
| - | - | - |
| **System instructions** | Define the assistant's role, behavior, boundaries, and conversation approach | "Explain options clearly, avoid making promises, and recommend a consultation when a visitor wants to proceed." |
| **Knowledge base** | Supplies reference information the assistant can use | A guide explaining your consulting framework, services, and common questions |
| **AI model** | Interprets questions and generates responses using available instructions and context | Explains a concept in plain language and asks an appropriate follow-up question |

These components work together. Adding a document does not automatically tell the assistant how to behave, and writing system instructions does not replace the need for accurate source material.

## What happens when a visitor asks a question?

A typical knowledge-supported interaction follows this process:

1. **The visitor asks a question.** For example: "How does your leadership assessment work?"
2. **Agentsy identifies potentially relevant knowledge.** When knowledge retrieval is configured and available, the system can retrieve material related to the question.
3. **The assistant generates an answer.** It uses the retrieved context alongside its system instructions and the AI model's capabilities.
4. **The conversation continues.** The visitor can explore the explanation, ask for clarification, or choose to speak with the expert.

This approach is often called **retrieval-augmented generation (RAG)**. The important idea is simple: the assistant can consult relevant reference material while responding.

> [!WARNING]
>
> Knowledge retrieval is not a live connection to every external system. Unless a specific real-time integration is available and configured, the assistant should not claim to know current inventory, appointment availability, account balances, or other changing information.

## What should you include?

Start with material that answers questions your visitors are likely to ask.

| Knowledge type | Examples |
| - | - |
| **Your approach** | Methodologies, frameworks, assessment processes, guiding principles |
| **Your offerings** | Services, intended audiences, engagement stages, scope and limitations |
| **Educational resources** | Guides, explainers, glossaries, training materials |
| **Frequently asked questions** | Common objections, definitions, next steps, eligibility explanations |
| **Policies and boundaries** | Approved disclaimers, escalation criteria, information the assistant must not promise |
| **Scenarios and examples** | Illustrative situations showing how your approach applies |

You do not need to upload everything you have. A smaller collection of relevant, current, well-structured documents is often easier to maintain than a large collection of duplicated or outdated material.

## Example: An advisory AI Experience

Imagine a boat-buying advisor creating an AI Experience to help visitors understand their options before speaking with a broker.

Their knowledge base could include:

* Boat types and typical use cases
* New versus used boat considerations
* Budgeting and total cost of ownership
* Inspection and maintenance basics
* Safety and regulatory considerations
* Questions to prepare before speaking with a broker

A visitor might ask, "Should I consider a center console or a cruiser for weekend family trips?"

The assistant can explain the trade-offs using the educational material and ask about intended use. If the visitor then asks, "Which boats do you have available right now?", the assistant should acknowledge that current inventory requires confirmation from the broker, unless an authorized live inventory integration is actually available.

This is the distinction Agentsy is designed to support: **help visitors make sense of their options, then create a natural path to a human conversation when needed.**

## Organize knowledge around visitor needs

A practical starting structure is:

1. **Overview and philosophy:** Who you help, your approach, and the purpose of the experience.
2. **Core concepts:** The essential ideas visitors need to understand.
3. **Options and comparisons:** Differences between approaches, services, or categories.
4. **Decision guidance:** Questions, considerations, and frameworks for evaluating options.
5. **FAQs and scenarios:** Realistic questions and illustrative examples.
6. **Limitations and handoff:** What the assistant cannot verify or decide, and when to involve a person.

You can divide these into separate documents when that makes them easier to update. Clear headings, descriptive titles, and focused sections help make the content easier to manage.

## Prepare your documents

Before adding material to Agentsy:

* **Check accuracy.** Confirm that descriptions, policies, and examples are correct.
* **Remove outdated content.** Avoid conflicting versions of the same information.
* **Use clear headings.** Make topics and sections easy to identify.
* **Explain specialized terms.** Write for your audience, not only for internal experts.
* **Separate facts from examples.** Label hypothetical scenarios clearly.
* **Protect sensitive information.** Do not upload confidential, personal, or regulated information unless you have confirmed that doing so is appropriate and permitted for your setup.
* **Identify time-sensitive claims.** Add dates where relevant and plan to review changing information.

> [!TIP]
>
> A useful first exercise is to write down the 20 questions prospects ask most often. Check whether your knowledge base provides a clear, approved answer to each one.

## Understand the limitations

A knowledge base is a reference source, not a guarantee of correctness.

Your assistant may still:

* Misinterpret a question or a source passage
* Retrieve an incomplete or less relevant section
* Encounter conflicting information across documents
* Lack the information needed for a specific answer
* Produce an answer that needs additional human review

To reduce these risks, maintain your documents, define appropriate boundaries in system instructions, and test representative visitor questions.

For high-stakes topics such as health, legal matters, or financial decisions, establish clear educational boundaries and routes to qualified human professionals. Do not treat AI-generated answers as a substitute for professional judgment.

## Test whether your knowledge is working

After adding or updating reference materials, try questions such as:

* "Explain your approach in simple terms."
* "How is option A different from option B?"
* "What should I consider before making a decision?"
* "Where does your methodology say to start?"
* "What happens if I want to speak with someone?"
* "Can you tell me something that is not covered in your materials?"

Check whether the answers are accurate, relevant, understandable, and appropriately cautious when information is missing.

## Next steps

Continue with these guides:

* [Upload documents](/knowledge-management/upload-documents) — Add source material to your AI Experience.
* [Organize your knowledge](/knowledge-management/organize-your-knowledge) — Build a maintainable knowledge structure.
* [File search and retrieval](/knowledge-management/file-search-and-retrieval) — Understand how relevant content is located.
* [Update your knowledge](/knowledge-management/update-your-knowledge) — Keep reference materials current.
* [Knowledge best practices](/knowledge-management/knowledge-best-practices) — Improve clarity and reliability.
* [Troubleshoot responses](/knowledge-management/troubleshoot-responses) — Investigate answers that are incomplete or incorrect.

> [!NOTE]
>
> Ready to define how your assistant should use this information? See [System Instructions](/create-and-configure/system-instructions) for behavior, boundaries, and human-handoff guidance.


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