Agent Studio / Guides / Build and manage agents

Studio Guide

Build

Build and manage agents

An agent is an autonomous assistant you build in Studio. This guide covers how to build, preview, deploy, schedule, deactivate and delete agents.

Ways to start: 3 options
Agent parts: 11
Knowledge files: CSV, PDF, Markdown, TXT

1.What is an agent?

An agent is an autonomous assistant you build in Studio. Each agent has a language model, a role, a goal, instructions, and capabilities you configure, such as tools, skills and access to your knowledge base. Agents are saved in your project alongside your workflows.

Use an agent when the work is too open-ended for a fixed workflow and you need an assistant that can reason about requests, pick the right tools and adapt to responses.

2.What you can build with agents

Use agents to:

Answer questions using your uploaded files and connected services.
Take actions using the tools you attach.
Trigger or coordinate workflows to complete larger tasks.

Agents can also delegate work to other agents and run on a schedule.

3.How agents work

An agent runs a reasoning loop. When you send it a message, or a schedule triggers it, the agent reads your instructions, looks at the request and decides what to do next:

Call a tool.
Search your knowledge base.
Hand off to another agent.
Ask a follow-up question.

It repeats this loop until it produces a final response.

4.Parts of an agent

Configure these parts of an agent in the Agent Builder.

Model: The language model that reasons and generates responses. Choose a provider and model when you set up the agent.
Role: Who the agent is, for example a support specialist or a research analyst.
Goal: What the agent is trying to achieve for the user.
Instructions: The system prompt that describes the agent's tone, response format and constraints.
Tools: Actions the agent can take: built-in tools such as Python, MCP servers, and workflows, including other agents.
Web search: Lets the agent search the web for current information, using the model's native search tool or a fallback service.
Skills: Reusable behavior bundles that package instructions with the tools needed for a specific task.
Sub-agents: Other agents you attach to this agent so it can delegate work to them.
Schedules: Tasks the agent runs on a recurring basis once it is deployed.
Knowledge base: Files the agent can search and read for context.
Memory: Session memory keeps the context of the current conversation. It is on by default.

5.Build an agent

Build agents in the Agent Builder. Start with a name and a model, add the role, goal and instructions, then attach the tools, skills, knowledge and sub-agents the agent needs. Studio saves your changes to the draft automatically.

Choose how to start

There are three ways to create an agent.

Option	How it works	Best for
New agent	Fill in the Behavior form (Role, Goal and Instructions) and choose a model. Select Generate to have the behavior written for you from a short description.	A guided setup (this guide)
Generate with AI	Describe the agent and Studio builds a first version for you.	A quick first version
Blank canvas	Start from an empty workflow and build it yourself.	Full manual control

Create the agent

Select your project from the left menu, then open the Agents tab.
Select New Agent. Studio opens the Agent Builder with a blank draft.
In the Agent tab, enter a name for the agent.

Note: Use Studio Assistant

Describe what you want the agent to do, and the Studio Assistant suggests instructions, tools and skills to add. Refine the suggestions in the Agent Builder as you go. This is different from Generate with AI, which creates a complete first version of the agent.

Choose a model

In the Agent tab, open the Model field and pick a provider and model. If the provider needs credentials, follow the prompt to add them.

Write the role

In the Role field, describe who the agent is, for example "a customer support specialist for an insurance company".

Write the goal

In the Goal field, describe what the agent should achieve, for example "resolve policy questions and route anything else to a human".

Write instructions

In the Instructions field, describe:

The agent's tone and response format.
What it should and shouldn't do.
Which tools or skills it should prefer for common tasks.

Tip

Keep instructions specific. If the agent doesn't behave as expected, refine the instructions first before adding more tools.

Add tools

In the Tools section, select Add tool. The tool picker lists every tool you can attach. Search by name, or use the All, Built-in, MCP and Workflows tabs to filter by type.

Built-in: Tools that come with Studio, such as Python.
MCP: Select MCP server by URL to connect any MCP endpoint, with optional authentication.
Workflows: Agents and workflows built by users. You can call agents you have already built. If you are using Studio for the first time, select one of the built-in template agents.

AI Suggest

Not sure which tools you need? Select AI Suggest, describe what you want the agent to do, and Studio suggests the tools that fit your needs.

The agent decides which tool to use based on your instructions and the task, using the credentials you attach when you add the tool. For sensitive tools, you can require approval before the agent runs them.

Sub-agents

To let the agent delegate work, attach another agent from the Workflows tab. The attached agent becomes a sub-agent that this agent can hand tasks to.

Enable web search

Turn on Web search to let the agent look up current information on the web. The agent uses the model's native search tool when the model has one, or a fallback search service when it does not.

Bundle capabilities with skills

A skill packages instructions with the tools needed for a specific task. Use skills when the agent handles several distinct tasks, each with its own steps.

To add a skill

Open the Skills section and select Add skill.
Give the skill a name and a short description.
Pick the tools the skill can use.
Write the instructions the agent should follow when it runs the skill.

The agent picks the right skill based on the description and the current request.

Upload knowledge

Add files the agent can search and read from the Knowledge tab.

Supported file types: CSV, PDF, Markdown and TXT

Once you upload a file, the agent can search your knowledge base to answer questions and pull in context.

Memory

Session memory keeps the context of the current conversation. It is on by default and needs no setup.

6.Preview your agent

Select Playground in the header to test the agent without deploying it. Studio starts a chat session with the current draft. Use the preview to check the agent's behavior, tool choices and responses before you deploy.

Playground disabled?

If Playground is disabled, the draft has configuration errors. Studio lists the items you need to resolve.

7.Deploy an agent

Deploy an agent to make it available to users, channels and schedules. Only the deployed version runs in production, so you can keep editing the draft without changing what people see or what your schedules use.

Open the agent in the Agent Builder.
Check for errors. If Deploy is disabled, Studio lists the configuration items you need to resolve.
Select Deploy.

Update a deployed agent

Once an agent is deployed, edits go into the draft. The deployed version keeps running until you deploy again.

To roll out your changes

Edit the draft in the Agent Builder. Studio saves your changes automatically.
Test the changes with Playground.
Select Deploy again to release a new version.

8.Run agents on a schedule

Schedules let a deployed agent run on its own at a set frequency. Open the Scheduler Configuration for the agent to set it up. The Executions list shows the schedule's past runs; it reads "No schedules found" until you create one.

To create a schedule, select Create a New Schedule and fill in:

Scheduler name: A name for the schedule.
Frequency: Minutes, Hours, Days, Months or Cron.
Interval: How often the schedule runs, in the unit you chose. For Minutes, the interval must be between 30 and 60.
Cron expression: Used with the Cron frequency, for schedules the other options can't express.
Max Retries: How many times a failed run is retried. The maximum is 10.
Retry Delay (minutes): How long to wait between retries. The maximum is 60.
Scheduler Input: The task or message the agent receives on each run, written the same way you would send it in chat.

Select Cancel to close the form without saving.

Note

Schedules run only against the deployed version of the agent. Deploy the agent before you expect a schedule to run.

9.Deactivate an agent

Deactivating takes a deployed agent offline without deleting it. It stays deactivated until you activate it again. There are two ways to do this:

Go to Deployments, find the deployed agent and toggle it off.
In the Agent Builder, open Settings from the Deploy menu and toggle the agent off.

To bring the agent back online, toggle it on again in either place.

10.Delete an agent

Deleting is permanent

Deleting removes the agent permanently. If you only want to take it offline for a while, deactivate it instead.

Go to Deployments and find the agent.
Select Delete.
Confirm the deletion.

11.Manage agents through MCP

You can also build and manage agents from an MCP client, such as Claude Desktop or Claude Code, by connecting it to Studio's MCP server. The server provides tools to search for, create and configure agents, in addition to its tools for working with workflows.