Skip to main content

Getting Started

The Agent Builder lets you create custom AI agents by combining tools and instructions together. Think of it like assembling a specialist — you define what it knows (instructions), what it can do (tools), and how it behaves (model settings).

Core Concepts​

An agent is made up of three building blocks:

  • Instructions — Guidelines that shape how the agent responds. These set the tone, rules, and priorities for the agent's behavior.
  • Tools — Defines specific capabilities for your agents. A tool gives your agent the ability to do tasks.
  • Agent — The final composition. An agent combines a model with your chosen tools and instructions into something you can interact with.

The relationship is simple: Instructions + Tools = Agent.

Tools and instructions are independent, reusable resources. One instruction set can be shared across many agents, so changing a persona is a single edit. One tool can be attached to any number of agents. An agent can hold several of each.

Finding Your Way Around​

Open the Agent Builder from your app launcher. The sidebar has five entries:

  • Agents — your agent library.
  • Tools — your tool library.
  • Instructions — your instruction library.
  • Getting Started — an in-app version of this overview.
  • Creating Tools — an in-app reference for authoring tools, mirroring the Tool Authoring Reference.

The three resource entries each open their own landing page: a searchable grid of cards for that resource type. Opening a card takes you to a full-page editor for that one resource. Every resource has its own address, so an editor can be linked to and opened directly.

Landing Pages​

Every landing page has the same shape:

  • A search box that filters by name and description.
  • A New Agent / New Tool / New Instruction button that opens an empty editor.
  • Cards grouped into ownership sections: Your items, Shared with you, your organization's items, and Third Loop items. A section filter above the grid narrows the view to one group, and sections you have nothing in are hidden.
  • A ⋯ menu on each card you own, offering Edit, Share, and Delete. Cards shared with you offer Leave instead. Organization and Third Loop cards are read-only and have no menu.

Editors​

Each editor is a full page with a consistent header: a back arrow to the landing page, the resource name, a Draft — unsaved changes hint while you have edits pending, and the actions for that resource (Share, Save, and on agents, Test agent).

Editors guard your work. If you navigate away — sidebar, back arrow, or closing the tab — with unsaved changes, you are asked whether to save, discard, or stay put.

Saving creates a new version

Saving changes to an existing agent or tool creates a new version of it rather than overwriting the old one. The editor follows the new version automatically and existing shares move with it, so there is nothing for you to do — but it is why the address changes after you save.

Creating Your First Agent​

We recommend creating an instruction set and a tool or two first, then combining them into an agent. You can create them in any order, and you can start a new tool or instruction without leaving the agent editor.

  1. Open Instructions and click New Instruction. Give it a name, a description, and one or more rules describing how the agent should behave. Save.
  2. Open Tools and click New Tool. Name it, write a purpose, and define its schema — or upload a Python handler. Save.
  3. Open Agents and click New Agent. Give it a name and description, adjust the model settings if you want, then attach the tool and instruction you just made in the Attachments section.
  4. Click Save, then Test agent to try it out in the side panel.

Quick Start​

If you want to jump right in:

  1. Open Agents and click New Agent.
  2. Give it a name and description.
  3. Leave the model settings at their defaults.
  4. Click Save.
  5. Click Test agent and send it a message.

That's it — you now have a working agent you can test right from the builder, and you can attach tools and instructions to it whenever you are ready.

For more details on each building block, see the dedicated pages for Agents, Tools, and Instructions. For advanced tool authoring — schemas and Python executable tools — see the Tool Authoring Reference.