Tools and integrations
What tools each catalog agent has access to and how integrations expand their capabilities.
Catalog agents ship with a fixed toolkit tuned for their role. Custom agents pick their tools in the builder — see Tools and integrations (builder). This page covers the concepts: what tools are, what each built-in does, how integration tools work.
Built-in tools
These tools are available to every catalog agent:
- search_kb — Search your workspace knowledge base for relevant documents
- read_file — Read the content of a file the user attached to the message
- search_org_contacts — Look up a person by name within your workspace contacts (relevant for Sales and Support primarily)
Any agent can use any of these. The Sales agent leans heavily on contact lookup; the Research agent leans on knowledge-base search. The model decides which tool to call based on the user's message.
Integration tools
When a workspace connects an integration (HubSpot, Gmail, Google Drive, etc.), agents gain access to integration-specific tools. For example:
- HubSpot connected → agents can read CRM contacts, deals, and activity
- Gmail connected → agents can search recent email threads, draft replies
- Google Drive connected → agents can search and read documents
Integrations are configured at the workspace level. Once an admin connects an integration, every agent in the catalog can use its tools — there's no per-agent gating.
See Integrations for the list of supported integrations and how to connect them.
Per-agent tool emphasis
Even with the same toolkit, the four agents emphasize different tools because of their specialties:
| Agent | Tools they reach for most |
|---|---|
| Customer Support | search_kb (find policy/product info), search_org_contacts (look up customer history) |
| Sales Assistant | search_org_contacts (qualify leads), CRM tools when integrated |
| Research Analyst | search_kb, read_file (synthesize from sources) |
| Content Creator | read_file (edit drafts), search_kb (pull product context for blog posts) |
This is emergent behavior — the agents have system prompts that orient them toward certain tools, but they're free to use any tool when relevant.
What tools the agents don't have
To keep agents safe and predictable, the catalog agents are deliberately restricted from:
- Sending email automatically (they can draft, you send)
- Creating or modifying records in connected integrations without confirmation
- Running shell commands or arbitrary code
- Interacting with the workspace's billing or admin surfaces
If you need an agent that can take destructive action automatically, build a workflow (Workflows) — the workflow can call the agent and then take the action with approval.
When the agent should have searched but didn't
The agent's system prompt tells it when to reach for tools, but it doesn't always pick the right moment. If you ask a question that should have triggered a search_kb call and the agent answered from training data instead, you can nudge it:
"Search the knowledge base for our refund policy and quote it directly."
Being explicit usually fixes the issue.
Tools you'll see in the chat
Tool calls appear inline in the chat as small pills like search_kb or read_file. The pill is the agent's way of saying "I called this tool". Hover or click the pill to see what arguments the agent passed and what was returned.
If you see no tool-call pills in a response that should have used tools, the agent answered from context alone. Often that's fine; sometimes it means you should re-ask with explicit guidance.