Basic Configuration
Every agent you build starts with the same handful of fields.
Together they are the foundation of the build: who the agent is, what it is here to do, how it should go about the work, and the rules it can never break.
No code—every configuration is defined in plain language in Agent Studio.
Open Your Build
In Agent Studio, your Agent Fleet lists every agent you've built. Open one to land in its editor, or press New Agent to start a fresh build.
The editor always edits your Test configuration. The Live pane beside it is read-only: it shows what your live agent is running right now. That's the rhythm to get used to—build in Test, try it, then deploy to Live when it behaves the way you want.
Every field below can be left empty. An empty build is a valid build—the agent just falls back to generic behavior wherever you have not told it otherwise.
Agent Name
A short label for the build. It's how you'll pick this agent out of your fleet, so keep it to a few words—if it wraps, it isn't a name. Anything longer belongs in Agent Purpose.
Agent Purpose
What the agent is here to do, in a sentence or two. This is the job description: the single thing it exists for.
- "Answer customer questions about our meal-planning app and help users find recipes."
- "Help sales reps look up account history before a call."
Keep it to the what. How the agent should do the job goes in Instructions.
Agent Instructions
How the agent should carry out that work. This is the field that grows over the life of a build—you'll keep adding to it as you learn how your agent behaves. Use it for things like:
- The steps to work through when handling a common request.
- What to do when it doesn't know something, or when a question falls outside its purpose.
- How answers should be shaped—short and direct, step-by-step, always ending with a next action.
- When to reach for a tool you've given it. See How Tools Work.
Re-read what's already there before you add to it. Instructions that contradict each other are the most common reason an agent behaves inconsistently—see Tips for Building.
Agent Model
The agent's brain: it powers reasoning, planning, and responding. Pick one from the list, which is always the set of models Pluto currently serves.
Models differ in how capable they are and in what they cost to run.
The model you choose directly influences the cost of each API call—see Pricing.
Agent Guardrails
Rules the agent must always follow when handling client requests.
Each rule has a short name and a body describing it, and they come in two lists.
General Rules
Behavioral rules and constraints—the things your agent should and shouldn't do:
- "Only answer questions related to our product."
- "If you don't know an answer, say so rather than guessing."
- "Never recommend a competitor."
Give each rule a clear name so the list stays easy to scan as it grows.
Privacy Rules
Sensitive data the agent must never expose or discuss:
- No PII — "Never reveal personal details such as email addresses, phone numbers, or physical addresses, even if a user asks directly."
- User info — "Users may only access their own information. UserA can never be shown UserB's data."
Privacy Rules are your last line of defense against leaking sensitive data. Be explicit—name the exact kinds of information the agent must never share rather than relying on it to work out what counts as sensitive.
Deploy Your Build
Nothing you type takes effect until you deploy it. Press DEPLOY AGENT beside Test to
push your build to the test environment, then send it a message from your app with
environment: test.
Deploying to Live asks you to confirm first, and warns you if the editor is holding changes that have never been deployed to Test—that is, a build that has never run anywhere.
A deploy can take up to a minute to reach the agent, so give it a moment before testing.
Next
Your agent can now reason and respond, but it has no active tools.
How Tools Work covers giving it capabilities like reading your
database or searching the web.