Add the think tool

Create the built-in tool and register it with the concrete tools available to the agent.

1. Use the default tool

import { Agent, createThinkTool } from '@anvia/core'

const agent = new Agent({
  id: 'support-investigator',
  model,
  instructions:
    'Use think when you must compare multiple tool results before answering.',
  tools: [
    createThinkTool(),
    searchTicketsTool,
    getAccountTool,
  ],
  maxTurns: 6,
})

Registration makes the definition available to the model. The model still decides whether to call it, so give a specific instruction describing the useful decision boundary.

2. Customize its definition

const evidenceCheck = createThinkTool({
  name: 'review_evidence',
  description:
    'Record a concise assessment of the evidence before choosing the next action.',
})

name defaults to think. description defaults to a built-in explanation that the tool records a thought without retrieval, memory storage, or external state changes.

Changing either option does not change the input and output contract:

type ThinkInput = {
  thought: string
}

type ThinkOutput = string

3. Keep it static when needed

const agent = new Agent({
  id: 'operations',
  model,
  tools: [
    createThinkTool(),
    operationsToolIndex,
  ],
})

A normal tool is visible on every turn. A ToolIndex contributes only dynamically selected definitions, so this arrangement keeps the checkpoint available while retrieving operational tools.

4. Add it selectively

Use think for agents that genuinely compare evidence or make multi-step decisions. Adding it to every tool-capable agent increases prompt surface and may encourage unnecessary turns.

Next, understand how it works.