Think tool

The think tool gives an agent an explicit checkpoint between gathering information and choosing its next action. It is useful when a run requires comparison, planning, or a deliberate pause before another tool call.

Explore the think tool

Page	Learn how to
Add the tool	Register createThinkTool() and customize its name.
How it works	Understand the tool call, echoed result, and extra model turn.
When to use	Decide whether a workflow benefits from a checkpoint.
Instructions	Tell the model when and what to record.
Privacy and visibility	Handle think calls as transcript data.
Production checklist	Review cost, policy, observability, and tests.

Add a checkpoint

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

const agent = new AgentBuilder('incident-review', model)
  .instructions([
    'Investigate the incident using the available evidence.',
    'Use think after gathering evidence and before proposing remediation.',
    'Keep the checkpoint concise and do not include secrets.',
  ].join('\n'))
  .tools([
    createThinkTool(),
    searchLogsTool,
    getDeploymentTool,
  ])
  .build()

When the model calls think, Anvia accepts { thought: string } and returns the same text as the tool result. Nothing is retrieved and no external side effect occurs.

model requests evidence
        ↓
tools return observations
        ↓
model calls think with a short checkpoint
        ↓
think echoes the checkpoint
        ↓
model chooses another tool or answers

Use the right mental model

Think is a normal tool, not a hidden scratchpad. Its call and result can appear in runtime events, traces, and stored messages according to the same rules as other tool activity.

It can improve deliberate tool use, but it does not enforce approval, authorization, or business policy. Keep those controls in tool handlers, hooks, and application code.