How AI Studio works

AI Studio has two ways to build the same kind of thing. This page explains what they are, what they are made of and where the credits go.

Agents and workflows#

An agent is a form with a test chat beside it. You choose a model, write instructions, add tools and switch on memory. It suits the common case: someone asks something in a chat, an AI agent works on it with its tools, and the answer comes back.

A workflow is a canvas. You place steps on it and wire them together, so you decide the order, add branches, and call several models or services in a fixed sequence. Choose it when the process has to run the same way every time or needs more than one agent step.

In the sidebar they are two separate lists: Agents and Workflows. They are the same thing underneath: an agent is a workflow whose graph is one chat trigger, one AI Agent step and one Chat Output. That is why an agent has everything a workflow has, including sessions, chatbots, schedules, integrations and version history.

An agent stays an agent only while its graph keeps that shape. If you want to add steps the agent form has no field for, use Convert to workflow in the builder. See convert to workflow.

Steps#

A step is one unit of work: read a document, call a model, send a message. Every step has inputs and outputs, and steps are grouped into categories such as triggers, agents, retrievers and outputs. The step reference lists them all. In the agent builder you rarely see steps directly; the form sets them for you.

Triggers#

A trigger is the step that starts a run. Every agent has the Chat trigger, which covers the playground, chatbots and the API. In the builder’s Triggers section you can add others, such as a Slack message, a schedule or a webhook. In a workflow you place triggers on the canvas. See workflow triggers.

Runs and sessions#

Each time a trigger fires, FlowHunt starts a run. For chats, a run belongs to a session: the whole conversation with one user. The builder keeps every test conversation as a session so you can reread it. Conversations that happen through a chatbot show up under Chatbots history in the sidebar’s History page. See chat history.

Draft and published#

Every agent and workflow has two versions at once.

• The draft is what you edit. The builder saves it automatically, and the test chat runs it.
• The published version is the live one. Publish changes copies the draft to it.

Versions are numbered, and the History tab of the agent builder keeps earlier versions you can restore. See publishing and versions.

Where credits are spent#

AI models cost credits. Before an AI Agent step starts, FlowHunt checks the balance of the workspace. A run in the test chat is a real run, so it uses credits like any other. When the workspace is out of credits, schedules created for a flow are deactivated, and the home page shows You are out of credits. Creating a flow also counts against your plan’s flow limit. See credits and plans and limits.

Next steps#