Skip to main content
Version: 3.6.1-rc.2

What are tools and preset agents?

What is a tool?​

A tool is an action a model can ask to run, such as searching your documents or creating a ticket in another system. Every tool the platform offers comes from a server that uses the Model Context Protocol (MCP), an open standard for connecting models to tools. There are two kinds:

  • MCP servers you connect. Servers that run elsewhere and give the model access to other systems. Some ask each user to sign in with OAuth.
  • RAG endpoints. Each RAG endpoint has a built-in MCP server with tools to search its documents. In Chat, these appear as knowledge bases.

What is a preset agent?​

A preset agent is a preconfigured AI assistant built for a specific use case, such as answering questions from your HR policies or helping your support team with tickets. Someone sets it up once, and everyone it is shared with can use it like a model. It combines:

  • a base model or model alias, which writes the answers
  • instructions, which tell it how to behave
  • a set of tools, such as RAG endpoints and MCP servers, which it uses on its own
  • optionally, locked parameters such as temperature, and output guardrails that check its answers

With a preset agent, you can:

  • Chat with it. Pick it in Chat, next to the models. It already knows which documents and tools to use.
  • Call it from your application. Name the agent in the model field of a Responses API request.
  • Use it without extra access. You only need access to the agent, not to its model or tools.

See Creating preset agents.

Why use tools and preset agents​

Tools​

  • You don't write the tool loop. Send one request to the Responses API. The Model Gateway calls the tools for the model and gives you back the final answer.
  • Connect a tool once, use it everywhere. The same tools work in Chat, through the API, and in MCP clients such as coding agents.
  • Tools only do what you are allowed to do. Every tool call runs as you, so a tool only reaches data that you, or the agent you are using, can access.

Preset agents​

  • Share an agent without sharing everything behind it. Users only need access to the agent, not to its model or tools.
  • Check the answer before users see it. An output guardrail asks another model to review the final answer first. With a guardrail on, the agent's tool calls and in-between steps stay hidden.

How the gateway runs tools​

client ──► POST /v1/responses ──► Model Gateway ◄──► model
│
└──► tools: RAG endpoints, MCP servers
  1. A client sends a request that names a model and the tools it may use, or names a preset agent, which brings its own model and tools.
  2. The gateway keeps only the tools the caller may use and sends the request to the model.
  3. When the model asks for a tool, the gateway runs it and passes the result back. It repeats until the model has a final answer, then returns the response.

More resources​