Tools

Tools are actions a bot can invoke — either as an explicit flow node (API Call, for example), or through an LLM Response node with tool calling enabled (agentic bots).

Where nodes describe conversation logic, tools describe what the bot can do. Common examples: look up an account balance, check order status, schedule an appointment, process a refund, book a demo.

Concepts

Tool — a named, reusable action. Configured once at the tenant level; referenced by multiple bots.

Parameters — the inputs a tool needs. Each parameter has a name, type, description, and required flag. Parameters are the contract between the bot and the tool.

Action — what the tool actually does when invoked. Usually an HTTP call to your systems, but can also be an MCP tool call, a Salesforce operation, or an internal platform action.

Tool library — the collection of tools available to your tenant. Managed in Operations Portal → Tools.

Where tools are used

LLM Response with tool calling — the model sees the tool's name, description, and parameter schema. It decides autonomously when to call which tool. Best for agentic bots that handle a variety of user requests.

Flow node explicit — you can wire a specific tool into a specific spot in a flow. Less flexible, more deterministic. Use when you want the tool to fire unconditionally.

Multiple tools per LLM Response — a single agentic node can have access to a dozen tools; the model picks whichever fits the current turn.

Getting started

• Creating Tools — the walkthrough for defining a new tool
• Best Practices — patterns that keep tools reliable + safe

System tools

Some tools ship with the platform as is_system tools — you can't edit or delete them, but they're available to every tenant:

• Handoff to Agent — same behavior as the Handoff node
• End Conversation — cleanly close the session
• Get Current Time — returns the current timestamp (useful for time-sensitive routing)
• Get Tenant Info — surface tenant metadata to the LLM

System tools appear alongside your custom tools in the tool library, marked with a system indicator.

Practice

• Tool descriptions matter — the LLM reads the description to decide when to call the tool. Vague descriptions produce inconsistent behavior.
• Fewer, more powerful tools > many small ones — a get_customer_context that returns account + orders + subscriptions is easier for the model to use than three separate tools.
• Test extensively — a small change to a tool's parameter schema can break every bot using it. See Best Practices.

See also

• Creating Tools
• Best Practices
• LLM Response with tools — agentic patterns
• API Call — often the underlying action of a tool