Content & website
The agent answers from what you approved.
Policies, terms, product facts and service descriptions, each with an author and a reviewer. Agents answer from these and cite them. They cannot add to them, and when two approved sources disagree they stop and ask.
What connects
Your knowledge base or approved-claims store, through its API.
What crosses the boundary
- Into the agent: approved claims with their sources and approval dates.
- Out of the agent: nothing. This connection is read-only.
- Never: draft or unapproved content, and nothing written back.
The capabilities this exposes
- Read approved knowledge
Setting it up
- 1Authorise a read-only connection.
- 2Confirm which collections are approved for agent use.
- 3Grant knowledge access to the agents that answer customers.
Permissions
Read-only, and limited to approved content. Draft and unapproved material is not exposed, because an agent quoting a draft is an agent inventing a company position.
Jobs an agent can own once this is connected
- Answer a product question and cite the source
- Refuse to promise something not in approved knowledge
- Flag a published page that contradicts an approved claim
Where it stops
Read-only, permanently: no agent can write to approved knowledge. This does not make an agent correct — if your knowledge base is wrong, the agent will be wrong and cite you. Contradictions between two approved sources halt the agent rather than being resolved by it.
Content & website
Others in this category.
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]