How Custom Agent Functions Work
From caller question to live system lookup in seconds.
Caller asks a question requiring live data
"Do you have the XR-500 in stock?" or "What's the status of my order #12345?"
AI recognizes the intent and selects the right tool
Based on the conversation context, the AI decides which of your configured tools to call — your inventory API, order lookup endpoint, or CRM system.
Tool is called with extracted parameters
The AI extracts relevant values from the conversation (product name, order number, customer email) and calls your API with those parameters. Template variables like {{param}} are substituted automatically.
AI uses the response to answer naturally
Your system responds in under 30 seconds. The AI reads the result and answers the caller naturally: "Yes, we have 4 units of the XR-500 available for pickup today."
Two Ways to Connect. Unlimited Possibilities.
HTTP tools for simple REST APIs. MCP for richer, dynamic integrations.
HTTP Tools — Any REST API
Define name, description, URL, method, headers, and input parameters using a visual JSON Schema builder. No code required. Your AI can call any HTTP endpoint during a live call.
MCP Server Support
Connect Model Context Protocol servers over SSE or HTTP transport. Tools are discovered dynamically — add new capabilities to your MCP server and the AI picks them up automatically.
Shared or Agent-Specific
Define tools once and share them across all your agents, or assign specific tools to specific agents. A sales agent can look up pricing while a support agent checks ticket status.
Works in SMS Too
Custom functions are not limited to voice calls. Your AI can query external systems during SMS conversations as well — same tools, same data, consistent experience across channels.
What Businesses Connect
E-commerce & Retail
Shopify, WooCommerce, custom stores
Field Service & HVAC
ServiceTitan, Jobber, Housecall Pro
Healthcare & Clinics
EHR systems, scheduling platforms
Law Firms & Professional Services
Clio, PracticePanther, custom CRMs
Property Management
Buildium, AppFolio, Yardi
SaaS & Tech Support
Internal APIs, Jira, Zendesk
Frequently Asked Questions
What kinds of APIs can I connect?
Any REST API that accepts HTTP requests. This includes your CRM (HubSpot, Salesforce, Pipedrive), inventory management systems, order tracking APIs, booking platforms, payment processors, and custom internal APIs. If it has an endpoint, the AI can call it.
What is MCP and why would I use it instead of HTTP tools?
Model Context Protocol (MCP) is an open standard for connecting AI agents to external data sources and tools. If you already run an MCP server, you can connect it directly and the AI will discover all available tools automatically. For simpler use cases, HTTP tools with manual configuration are usually easier to set up.
How do template variables work?
In your URL, headers, or request body you can write {{variableName}} as a placeholder. When the AI calls the tool, it substitutes the actual value extracted from the conversation. For example, a URL like /orders/{{orderNumber}} becomes /orders/12345 when the caller mentions their order number.
What happens if the API is slow or times out?
Each tool call has a 30-second timeout. If the API does not respond in time, the AI gracefully tells the caller it could not retrieve that information right now and offers to take a message or try again. The call does not drop and the caller is never left in silence.
Can the AI update records, not just read them?
Yes. HTTP tools support any method — GET, POST, PUT, PATCH, DELETE. You can configure a tool that creates a support ticket, updates a CRM contact, logs a callback request, or marks an order as flagged. The AI can write to your systems, not just read from them.
Is my API data secure?
Tool calls happen server-side. Your API keys and secrets live in the tool configuration headers and are never exposed to callers. All requests are made over HTTPS. You can also restrict what parameters the AI can pass to limit the scope of each tool.