Skip to main content

​
What are tools?

When your AI phone agent talks to a caller, it can do more than have a conversation. It can take real actions during the call:
  • Schedule a meeting on Cal.com when a caller wants to book time
  • Create a contact in Salesforce when someone provides their info
  • Send a Slack message to notify your team about an interested lead
  • Call your own API to check order status or update a database
These actions are called tools. A tool is something your agent can do when the conversation calls for it.
If you used Bland before agents, you may know Custom Tools. They are now legacy. Tools replace them, and you attach a tool to an agent by tagging it in a prompt.
For the full reference on building tools in the agent builder, see Tools. This page is a short orientation.

​
Three kinds of tool

Walkthroughs:

​
Where tools live

Set the sidebar selector to All agents, then open Tools under Resources. You get four tabs: Inside an agent, Tools under Build lists only that agent’s tools. See Where tools live for how the two scopes differ. You connect an account once under Connections, then build as many tools on it as you need. One Salesforce connection can power a tool that creates a contact, another that creates a lead, and a third that creates an opportunity.

​
How a tool gets its inputs and outputs

Inputs. In a webhook tool, write a variable like {{email}} anywhere a value should change per call. For each variable, choose Add input variable to have the agent fill it from the conversation. Leave it alone and it resolves from variables already on the call instead. See Build a webhook tool. Outputs. Output fields name the parts of the response the agent can use. Every tool starts with ok, message, and data. Keep ok checked: it is the only signal that the call failed. Checking at least one output field makes the tool wait for a response. Check nothing and the tool fires and moves on. That suits logging and notifications, and it is wrong for anything the agent needs an answer from. To use an output in a prompt, write its bare name, like {{booking_id}}.

​
Give a tool to an agent

Building a tool does not give it to an agent. In the agent’s Prompt, type @ and pick the tool by name. The tool then appears under Tool references below the prompt. Tools attach per prompt. A tool tagged in the agent’s prompt does not follow the agent into a pathway, so tag it there too if the pathway needs it. Tagging also attaches a copy of the tool. See Tag a tool into a prompt before you edit a tool an agent already uses. The tool’s description is what the agent reads to decide when to call it. Write the trigger, not the mechanism:

​
Monitoring

The Analytics tab at All agents shows call volume and failure rate per tool. Watch the failure rate. A tool that fails quietly leaves the agent improvising mid-call. To pull execution logs programmatically, including logs for a single call by call_id, use the Tool Execution Logs endpoint.

​
API reference

Manage tools programmatically using the Tools API:
  • Create Tool: Create a new integration tool connected to Slack, a custom API, or any supported integration.
  • List Tools: Retrieve a paginated list of your tools.
  • Update Tool: Update any field on an existing tool. All fields are optional.
  • Tool Execution Logs: Query per-execution logs across your tools.
  • Execution Stats: Aggregate execution metrics grouped by integration, status, date, and more.
Docs for agents: llms.txt