Table of Contents

• The short answer

• Why "who can edit" is not enough

• How it works in Fini AI

• License and entitlement data

• Example

• A five-step rollout

• Frequently Asked Questions

The short answer

To restrict knowledge base access for an AI support agent, filter the knowledge itself, not only the people who edit it. In Fini AI, admins add Attribute Filters to article folders. When a customer's attributes match a folder's filter (for example plan = Enterprise or state = California), the agent can answer from that folder. When they don't match, the agent can't use it and answers from content that applies, or hands the conversation to a person.

Want to see answers scoped by plan on your own content? Book a demo.

Why "who can edit" is not enough

Most permission settings control your team: who can publish articles, who can change workflows. Those matter, but they don't stop an AI agent from telling a Starter customer about an Enterprise feature, or quoting a California policy to a customer in Texas.

Product tiers, add-ons, regional rules and regulated plans all need answers scoped to the person asking. An agent that reads everything will mix them, and in insurance, healthcare or finance a mixed answer is a compliance problem, not a tone problem.

For the team side (who can edit and publish), see granular permission controls for AI support chatbots and fine-grained permission controls for AI email assistants. This guide covers the customer side.

How it works in Fini AI

1. Identify the customer

Pass identity through a signed token in the chat widget, or read it from your helpdesk and systems through Actions or an MCP connection. The customer's attributes, such as plan, license tier, location or state, come with it.

2. Put filters on knowledge folders

Admins add Attribute Filters to article folders. Two kinds are available:

• User Attributes: any attribute your systems return, such as plan tier, license tier, add-on or state.

• System Attributes: such as the customer's language, so answers stay in the right language.

3. Combine filters the way your policy reads

Values of the same attribute combine as OR (plan = Pro or plan = Enterprise). Different attributes combine as AND (plan = Enterprise and state = California). If a customer has no value for an attribute a folder needs, that folder is excluded. Missing data never widens access.

4. Choose what the AI can see

You choose which attributes the AI can see, and a Confidential attributes guardrail keeps selected attribute values out of replies. An attribute can drive a folder filter without its value ever being repeated to the customer.

5. Separate audiences fully when needed

Filters are set per bot. When content must never overlap, such as customers and resellers, run separate agents per audience.

6. Audit what it used

Every conversation has an AI Steps trace that shows which knowledge the agent used and which rule ran, so you can check scoping on real conversations.

How Fini AI does this: Attribute Filters on knowledge folders, a signed widget token for identity and an AI Steps trace on every conversation, all set up on one platform. See the Fini AI platform.

License and entitlement data

Any user attribute your systems return, such as plan tier, license tier or add-on, can drive a folder filter. If your licensing system knows a customer has the analytics add-on, the analytics folder opens for that customer and stays closed for everyone else.

Example

A five-step rollout

1. List the questions whose answer changes by plan, license or region.

2. Group those articles into folders by audience.

3. Confirm which attributes your systems already return for each customer.

4. Add filters, starting with the folder where a wrong answer costs most.

5. Review AI Steps traces on real conversations for the first two weeks.

Regulated teams can read how this applies in insurance and healthcare.

See how Fini AI handles permission-aware answers. Book a demo.

Frequently Asked Questions

Can an AI support agent give different answers by customer plan?

Yes. In Fini AI, Attribute Filters on article folders limit which knowledge the agent can use for each customer, based on attributes like plan, license tier or state.

How does the AI know which plan a customer is on?

Through a signed token in the chat widget, or by reading it from your helpdesk and systems via Actions or MCP.

What happens if a customer's plan is missing?

Folders that need that attribute are excluded for that customer. Missing data never opens restricted content.

Will the AI see every customer attribute?

You choose which attributes the AI can see, and a Confidential attributes guardrail keeps selected values out of replies.

Does this work for regulated plans like Medicare Advantage?

Yes. A folder filter such as plan = Medicare Advantage scopes answers to members on that plan.