A helpdesk for your staff that answers from your documents
Give a contained agent your handbook, policies or IT guides, and your team can ask it anything about them in the app or on a page in your Canvas theme. It reads only what you grant it.
What you get
The agent reads only what you grant: a knowledge folder, project documents and skills you approve. Everything else on your server stays out of reach.
In the CRHQ app, for the members who see the agent, or in a chat block on a page of your Canvas theme (signed-in pages for now).
Four spending limits are checked before any cost: per answer, per person per day, for the agent per day, and for all contained agents on the server per month. Contained agents run on a separate Anthropic API key you add for them.
Each conversation runs in its own sealed space, so it never reaches the rest of the server. Unless you choose one shared conversation for the agent, each colleague's conversation is their own.
When a policy changes, update the documents you granted and the agent answers from the new version.
Staff get answers about leave, expenses, equipment or security rules any time, from the documents you chose, without waiting for HR or IT.
> Agent: Up to 5 days, according to the Leave policy (section 3.2). Carried-over days must be used by 31 March.
> Colleague: And who approves it?
> Agent: Your team lead, in the HR system under Leave requests (section 3.4).
✓ Answered from the granted policy documents only
For CRHQ clients
Turn on Contained
Create an agent for this job and, in its Contained settings, switch on Contained. The first time, press “Set up on this server”. Every conversation with it now runs sealed off. Built-in and organization-managed agents keep full power.
Grant the documents
Grant the knowledge it needs: a knowledge folder with your handbook, project documents and skills you approve. Nothing else is visible to it.
Put it where your team works
Leave it in the app for the members who see it, or add a chat block to a page in your Canvas theme. Choosing who sees an agent keeps menus tidy; what keeps it safe is the containment itself.
Keep it current
Follow the setup guide.
Read conversations to spot missing answers, and update the documents you granted.
Get more out of it, safely
Grant only what it should share
The agent only knows what you give it, and anyone who can talk to it may get it to repeat what it knows. Grant documents that everyone who sees the agent may read, and never give it passwords, keys or tokens.
Who sees it is not a lock
Choosing which people see the agent in the app decides who sees it. Treat that as tidy menus, and rely on the containment and what you grant for safety.
Start with small caps
The default caps keep costs predictable. Watch real questions for a week before you raise them.
AI answers can be wrong
You can read every conversation. When an answer needs fixing, point the agent to the right document or update it.