Knowledge
The material an agent grounds its answers in, and how access to it is scoped.
Knowledge is the material an agent can draw on. The goal is not an agent that knows everything. It is an agent that reaches the right source at the right moment, so its answers reflect your organization rather than general reasoning.
Knowledge sits alongside skills, which define repeatable behaviour, and policies, which set the boundaries.
What goes in it
Documents and playbooks. Standard operating procedures, policies, product documentation, service manuals, quoting guidelines.
Structured records. Case histories, customer profiles, entitlement details, where access permits.
Context from conversations. Prior summaries, extracted signals, and approved memory attached to a record or to the organization.
What it changes
Grounded answers. Guidance references authoritative content instead of being reasoned from scratch.
Continuity. An agent can retrieve what is known about a record and behave as though it remembers, within whatever policy allows.
Consistency. Two teams asking the same question get the organization's answer rather than two versions of it.
How it is scoped
Access is never blanket. Policies decide which stores an agent can reach and at what level, which categories of information are restricted, and whether a given conversation may write anything back into shared memory.
Where this lives in the product
Knowledge stores are created and filled in the console under Knowledge stores, by uploading files or pasting text, and attached to agents and teams from there. The number of stores and the amount of material in them is capped by your plan.
On the phone, people can attach a store's files to a single chat message when they want an agent to look at one specific document; they cannot create or change stores. See Chats.