Connect
An agent should not invent an answer.
Agents answer from approved knowledge — your policies, terms, product facts and service descriptions — and when the knowledge does not cover a question, they say so rather than producing something plausible. This is the difference between an agent you can put in front of a customer and one you cannot.
Can we promise 24/7 support?
24/7 support is available on enterprise plans only.
Approved by the Service Lead · 2 September 2026
“Round-the-clock support is included on our enterprise plan. On your current plan, cover is 08:00 to 18:00 on weekdays.”
Anything about 24/7 cover on a plan that does not include it. The claim does not exist in approved knowledge, so the agent has no way to make it.
Approved means somebody approved it
A claim in the knowledge base has an author, a date and a reviewer. An agent quoting it is quoting your company rather than its own inference.
A question outside knowledge is escalated, not answered
Asked whether you offer 24/7 support when the approved answer is "enterprise plans only", the agent gives that answer. Asked something not covered at all, it hands over.
Contradictions stop the agent
Two approved sources that disagree produce a halt and a request for a decision, not a choice. An agent that resolves a contradiction by picking one has invented a company policy.
Customer content is data, never instruction
An email, an attachment or a web page cannot change how an agent behaves. Anybody who can contact you can introduce content, which makes this a security property rather than a design preference.
Questions this answers
The things you currently have to ask somebody.
Each of these is a question an owner asks regularly and currently answers from memory, a spreadsheet, or by interrupting whoever is nearest.
Where did the agent get that claim, and who approved it?
What happens when two of our own documents disagree?
Can a customer's email change what the agent believes?
What will the agent say when it does not know?
Where this stops
What it does not do.
Approved knowledge does not make an agent correct — it makes it attributable. If your knowledge base says something wrong, the agent will say the wrong thing and cite you. Keeping it accurate is work the platform can prompt but cannot do. Nor can agents write to it: an agent that could add to what your company considers true about itself would be a much worse product than one that cannot.
Next
Where this connects.
Give AI a job.
Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.
No code required · Full SDKs and API · Stop any agent instantly · [email protected]