Single agent
Tool use
Give the model real actions — search, code, APIs — not just text.
1 agentCore
How it works
- 1Expose each capability as a clearly-described tool with typed inputs.
- 2Let the model choose tools and arguments based on the request.
- 3Execute the call, return the result, let the model continue.
- 4Validate tool outputs before trusting them downstream.
Use it when
The task needs live data, computation, or to actually do something — write a file, hit an API. The model decides when and how to call.
Reach for something else when
A static answer suffices. Every added tool widens the surface for errors and needs guardrails.
Where you stay in the loop
You decide which tools exist and what they're allowed to touch; the agent decides when to call them. Keep oversight on state-changing tools and validate outputs before they're trusted downstream.
In the wild
A research agent calls a web-search tool, reads the results, then calls a calculator tool to total the figures it found.
Hand this to your agent
Act as a tool-using agent.
Tools you can call:
<list each tool: name, what it does, its inputs>
For my request, reason briefly about which tool to use, then output the tool
call as JSON: {"tool": "...", "args": {...}}. Stop and wait for me to paste
the result, then continue — call more tools if needed. Only give a final
answer once you have enough.
Request: <...>Replace the <…> placeholders, paste it into your agent, and it'll scaffold the workflow with you.