This guide walks you through building one working AI Content Agent from a blank start. You will create the agent, attach a small set of tools, write its instructions, set a trigger, and test it on a single project before you turn it on for real work.

You do not need to write code. If you can describe a task in plain sentences, you can build this. The example below builds an agent that adds a short draft outline note to a project when it reaches a review stage, but the same steps apply to any task you want to automate.

Before you start

Have these ready:

A workspace you can safely test in, with at least one project inside it.
A clear, one sentence description of the task you want the agent to do.
A rough idea of which tools the task needs. Query tools read your data, Project tools create and update projects, and Asset tools handle files, docs, and notes. If you are unsure, start with Query tools plus one action tool.

Step 1: Create a blank agent

Go to the AI Content Agents area and create a new blank agent. A blank agent is a valid starting point, so you do not need a template.

Give it a clear name and a short description that says what it does, for example "Adds a draft outline note when a project enters review." A plain name makes it easier to find later and easier for teammates to understand.

Step 2: Attach the right tools

Open the Tools picker and attach only the tools the agent needs. An agent can only take actions with the tools you give it, so keep the set small and specific.

Relato groups its own tools by category. For this example you would attach a Query tool (to read the project) and an Asset tool (to create the note). Other categories are available when you need them, including Google Workspace (Docs, Sheets, Slides, Drive), Task, Workspace, Comment, Reporting, Data and Visualisation, Memory, Web Research, Email, Sub-Agent, and Human-in-the-Loop.

A good rule: if the task does not clearly need a tool, leave it off. You can always add more later.

Step 3: Write the instructions

The Instructions field is the agent's prompt. Write it precise and boring. State the goal, the exact condition to act on, the steps to take, and when to stop.

Here is a small worked example you can adapt:

You add a draft outline note to a project when it enters the Review stage.
First, check the project's current stage. If the stage is not Review, stop and do nothing.
If the stage is Review, read the project title and summary.
Create one note on the project with a short outline: a working headline, three section ideas, and one open question.
Do not change the project stage, and do not create more than one note per run.

Because a stage trigger fires on every stage change in the scoped workspace, the stage check inside your instructions is what keeps the agent from acting at the wrong time. Put that check first.

Step 4: Set the trigger

Open the agent's Triggers settings and choose how it runs. You have three options:

Project stage changed: the agent runs on every stage transition in the workspace it is scoped to. Use your instructions to act only on the stage you care about.
Schedule: the agent runs on a set time or interval.
Manual: the agent runs only when you start it. This is the safest choice while you test.

Set the agent's Scope to your test workspace so it only acts there. For this first build, start with a manual trigger so you stay in control.

Step 5: Test on one project

Pick one safe project in your scoped workspace and run the agent manually against it. Open the agent thread and watch the run. The thread shows each tool call and its payload, so you can follow every step: which project it read, and what note it created.

Check the result on the project. If the note is wrong, missing, or in the wrong place, adjust the instructions and run again. Small edits to the instructions usually fix behavior faster than adding tools.

Step 6: Enable it

When the manual runs look right, switch the trigger to the one you want for real use, for example Project stage changed, with the scope still set to your workspace. Watch the first few live runs in the thread the same way you did in testing, so you catch any surprises early.

Tips

Keep instructions precise and plain. Vague prompts produce vague runs.
Attach the fewest tools the task needs.
Always test on one project before you enable a broad trigger.
Put your "only act when stage equals X" check at the top of the instructions.
Use the agent thread to read tool calls when a run does something you did not expect.

FAQ

Do I need to write code to build an agent?

No. You describe the task in plain sentences in the Instructions field, and you choose tools and a trigger in the settings.

My stage trigger fires too often. What is wrong?

The Project stage changed trigger fires on every stage transition in the scoped workspace. Add a check at the start of your instructions so the agent stops unless the stage matches the one you want.

How do I stop the agent from touching projects outside my test area?

Set the agent's Scope to a single workspace. It only acts within that scope.