Skip to main content

Knowledge bases

4 min read - Arm your AI agent with knowledge to succeed.

Updated this week

Many people will find teaching their AI agent to answer questions using content from their institutional website enough. This approach has the benefit of your website being the single source of truth, and any updates made or new pages published will continue to improve your agent without any action required from you.

However, the occasion may arise where you want more granular control over what your bot can see and how it replies. This can be done in one of two ways:

  • Knowledge bases - "If a student asks x, always reply with y."

  • External knowledge sources - "Sharing internal guides/resources safely with the bot without putting them on our website."

Getting started with knowledge bases

Think of Knowledge bases as your AI agent's specialized memory. Your bot will draw upon what it finds on your website for most answers. However, you may want the bot to deviate from this or provide a pre-written, stock answer.

That's where knowledge bases come in. They allow you to group items / owned answers containing pre-written replies, as well as the questions or phrases students might use that should to trigger them.

Knowledge bases and their items can be used by any number of unique AI agents/bots. They can be assigned to (or unassigned from) a bot by heading to Conversations > Chatbots > {your bot} > Knowledge bases.

Creating Knowledge bases

Head to Conversations > Knowledge bases and click + Create a knowledge base in the top-right corner.

Give your knowledge base a name, choose whether you'd like to assign your knowledge base to individual agents or teams, and what, if any, labels you'd like to use to describe this knowledge base. Assignments and labels can be used to filter knowledge bases and are a great way of better delineating ownership swim lanes.

Creating items

To add a new item to your knowledge base, head to Conversations > Knowledge bases > {your knowledge base} and click + Create item in the top-right corner.

Items are divided into four main sections:

Basic information

This section is all about good housekeeping. How will you internally name, assign, and tag your owned answer/item?

  • Name - The internal name of your item. This is never visible to students.

  • Assignments - Optional. Assign this answer to agents or teams for easy auditing.

  • Labels - Optional. Assign labels to describe the types of information your item provides.

Responses

This section is all about the type of response you'd like your bot to give. You can use multiple response types per item. Responses will be displayed in order, from top-to-bottom:

  • Message - A rich-text enabled text block. Type out the message you'd like your bot to use. Highlight any words/sections to apply rich text editing.

  • Display message buttons - Begin by picking a button style (list or pill) and click + Add item to add a new button to your list. Click your button to edit its label and choose what action should trigger if the button is clicked:

    • Send to a webpage - The URL users should be redirected to when clicked.

    • Invoke another item - Trigger another knowledge base item from any knowledge base.

    • Prompt chatbot response - Provide a detailed, specific prompt/question for your generative AI bot to reply to. This prompt will not be printed in the conversation.

      Example: Imagine our item is to help triage users to find Scholarships. We use a 'Message' and 'Display Message Buttons' to invite students to tell us their level of study. We then discreetly prompt our AI bot to generate a cogent reply:

  • Embedded form - Select a pre-made Gecko form and invite your AI bot to capture responses. Any conditional logic, workflows, or integrations will fire as usual. You'll only be able to select from the forms your user has permission to view.

Trigger questions

This section is all about defining which questions or phrases should trigger your item. Begin by adding an initial question and then click Generate trigger questions to use AI to quickly generate additional variants of your question, or Manually add a trigger question. We recommend a minimum of 10 trigger questions and a maximum of 15.

Actions

When your item is triggered, in addition to your Responses being invoked, you can also immediately run any number of the following actions:

  • Assign conversation - To any selected agents or teams.

  • Add a label - Apply any pre-made labels to the convo.

  • Add a note to the contact - Add a note to the conversation for agents to view.

  • End conversation with bot - The convo will remain open, but the bot will be unassigned and will stop responding to future student messages.

After your item sections are configured, click Save and publish item to make your item immediately available to any bot using your knowledge base, or click Save as draft to save an unpublished copy of the item in your knowledge base.

Once created, your knowledge base items can be published or unpublished at any time from the Conversations > Knowledge bases > {your knowledge base} table. Click the breadcrumb [...] button to alternate between the two states. Published items can be used by any bot with access to the knowledge base whereas unpublished items will never be used.

External knowledge sources

Sharing documents with your agent

Sharing your treasure trove of internal documents, FAQs and guides with your AI agent can help make it better at answering questions. Happily, this process is quick and easy.

This quick guide covers the process for sharing .pdf, .docx, .xlsx, and .csv files.

Organize Your Documents

  • Gather all .pdf, .txt, .docx, .xlsx, and .csv files you'd like your AI agent to learn.

  • Make sure the files are up-to-date and contain relevant information.

Upload to Google Drive / Microsoft OneDrive

  • Log into your Google Drive / Microsoft OneDrive account.

  • Create a new folder for your AI bot documents (optional but recommended).

  • Upload your files to Google Drive / MS OneDrive.

Publish to the web

  • Open each document

  • Click File > Share > Publish to web

  • Choose "Entire Document" in the first dropdown.

  • Click "Publish" and copy the provided link.

  • For more information, check out this Google Support guide.

Doing so will provide a unique static URL per document, meaning any changes you make to the bot will be pulled through into your AI Agent when it next scrapes the document.

Finally, share your link with your Gecko account, who can point your bot at the URL as requested!

AI agent x Sharepoint integration

Should you want us to include SharePoint content within your chatbot, you will need to provide us with a way to authenticate. This is done by registering an application in Microsoft Entra ID (formerly Azure Active Directory) and granting it the appropriate permissions to access your SharePoint site via the Microsoft Graph API.

We use the Microsoft Graph API to access your site's content rather than traditional scraping techniques.

Step 1: Register an app

  1. Sign in to the Azure Portal with an account that has admin privileges for your tenant.

  2. Navigate to Microsoft Entra ID → App registrations.

  3. Click + New registration.

  4. Fill in the following:

    • Name: Gecko

    • Supported account types: Accounts in this organizational directory only (Single tenant)

  5. Click Register.

Once registered, you will be taken to the app's Overview page. Make a note of:

  • Application (client) ID

  • Directory (tenant) ID

Next, create a client secret:

  1. In the left menu, click Certificates & secrets.

  2. Under Client secrets, click + New client secret.

  3. Give it a description (e.g. Gecko Integration) and choose an expiry period.

  4. Click Add.

  5. Immediately copy the secret Value - it will not be shown again.

Step 2: Grant Microsoft Graph Permissions

  1. In your app registration, click API permissions in the left menu.

  2. Click + Add a permission → Microsoft Graph → Application permissions.

  3. Search for and select: Sites.Read.All

  4. Click Add permissions.

  5. Back on the API permissions page, click Grant admin consent for [Your Tenant] and confirm.

You should see a green tick confirming that admin consent has been granted.

For institutions requiring granular access control: If your IT policy requires restricting app access to specific SharePoint sites (rather than all sites in the tenant), you can use the Sites.Selected permission instead of Sites.Read.All:

  1. In your app registration, click API permissions in the left menu.

  2. Click + Add a permission → Microsoft Graph → Application permissions.

  3. Search for and select: Sites.Selected

  4. Click Add permissions.

  5. Click Grant admin consent for [Your Tenant] and confirm.

Important note: At this point, the app has the Sites.Selected permission, but cannot yet access any site. You now need to grant it access to your specific SharePoint site. If you're sharing multiple sites, this step will need to be be repeated for each SharePoint site you'd like to share with your Gecko AI agent.

Step 3: Send Credentials to Gecko

Once the above steps are completed, please provide us with:

  • Client ID (Application ID)

  • Client Secret

  • Tenant ID (Directory ID)

  • SharePoint Site URL

We will then conduct a test to ensure the connection is working as expected.

Any questions? Feel free to start a live chat with a member of our support team or explore the rest of our academy at your leisure.

Related Articles
Users and permissions
Workflows
Slate
Chatbots
Unibuddy
Did this answer your question?