Skip to content

Think Tool

Think Tool is a reasoning tool for an agent. It works like a notepad: the agent writes its reasoning in it, the tool hands it back unchanged and the agent carries on with its thoughts in order. It doesn’t look up external information or take any action.

Connect its Tool output to the agent’s Tools input. The agent decides when to use it, following the instructions in Tool Description.

When to use it

  • The agent gets complex questions that need to be broken into parts.
  • The agent has to follow several rules at once, such as return policies with exceptions.
  • You want the run trace to keep the agent’s reasoning so you can review it.
  • The agent rushes and answers before checking all the data it has gathered.

Configuration

FieldWhat it does
Tool Name (required)The name the agent sees for the tool. Default: thinking_tool.
Tool Description (required)Instructions for the agent: what the tool is for, when to use it and what to write in it. They come filled in with a complete example you can use as it is or adapt.

Output

Tool: the reasoning tool. Connect it to the agent’s Tools input. Each time the agent uses it, it gets back the same text it wrote.

Example

An insurer uses an agent that answers whether a claim is covered by the policy.

  1. Add Think Tool and connect it to the agent’s Tools input.
  2. In Tool Description, add: “Use it before answering about coverage to go over the conditions and exclusions that apply”.
  3. When a customer asks whether water damage in the garage is covered, the agent first notes which conditions apply and which exclusions might affect it, then answers.

Good practices

  • Explain in Tool Description in which situations you want the agent to use it. The more specific it is, the better the agent uses it.
  • Each use is one more step for the agent: it adds a little time to the answer.