Skip to main content
Lookups are read-only tools from connected integrations. They let the AI Agent fetch up-to-date information during a conversation: for example, retrieving customer details or ticket comments. For CRM knowledge base imports, see Zendesk or Salesforce.

​
Lookups vs actions

Lookups and actions behave differently when you connect an integration: Lookups run automatically once connected. Actions require explicit setup.

​
Authorized vs unauthorized lookups

When you connect an integration (MCP server, Composio, etc.), all its lookup tools become available to the AI Agent: Authorized lookups run silently. Unauthorized lookups flag the conversation for review.

​
Why unauthorized lookups exist

Many actions need supporting data to work. For example, an MCP action “Create support ticket” might require a project ID. The AI Agent can fetch that ID using a lookup, but if you haven’t explicitly added that lookup to Knowledge Hub, it’s unauthorized. Rather than block the AI or require you to manually configure every dependency, unauthorized lookups let the AI gather what it needs. The trade-off: a Human Agent must approve before anything acts on that data.

​
How approval works

Example: You configure an MCP action called “Create support ticket” that requires a project ID.
  1. Customer asks to log an issue
  2. AI Agent calls get_projects to find the project ID. This lookup isn’t in Knowledge Hub, so it’s unauthorized
  3. The lookup returns the project ID
  4. AI Agent prepares to create the ticket
  5. Human Agent sees: “Action approval required: We need your approval since we had to perform additional lookups to get information for this action.”
  6. Human Agent clicks Yes to approve or Takeover to handle manually
The lookup completes: the AI gets the data it needs. Approval is required before the AI acts on that data or sends messages to the customer. This lets the AI work while you maintain oversight.

​
Add a lookup to Knowledge Hub

Adding a lookup authorizes it: the AI Agent can use it without triggering approval workflows.
1

Open Knowledge Hub

Go to Knowledge Hub and click Add source.
2

Choose integration

Select the integration containing your lookup (for example, Linear or Salesforce).
3

Select the lookup

Choose the specific lookup tool (for example, get_customer_details). If the lookup is already used elsewhere, the picker shows a warning with the usage count.
4

Configure and save

Set the Name, description, access level, and inputs, then click Confirm.

​
Reuse the same lookup multiple times

You can authorize the same lookup tool more than once in Knowledge Hub. Each source keeps its own Name, description, and input mappings so the AI Agent can choose the right variant (for example, pending vs cancelled order searches). When you select a lookup that is already configured, the picker shows how many other sources already use it. The Sources table shows each instance under its custom Title.
Two sources with the same tool, description, and inputs are blocked. A valid variant needs a unique Name and a different description or input mapping.

​
When to authorize lookups

Authorize a lookup when:
  • It retrieves non-sensitive data (project names, status lists)
  • You want fully autonomous AI operation
  • The lookup supports frequently used actions
Keep unauthorized when:
  • It retrieves sensitive customer data
  • You want Human Agents to review each usage
  • You’re still evaluating the integration

​
Configuration

​
See also