Agents

An agent is an AI helper that decides which jobs to run to finish a task. Where a workflow follows a fixed recipe, an agent is given a goal and a set of tools and works out the steps itself. In the Builder you can inspect every agent your project has — what it's for and exactly what it's allowed to do — before it ever goes live.

Annotated screenshot of the Builder's Agents tab listing an AI agent with its description and the tools it can use, callout: "Agents are AI helpers you build — see what each one can do here."

What an agent can decide

An agent is handed a task and a toolbox. The "tools" are your own functions — the small jobs your backend can do. The agent reads the task, picks which jobs to run and in what order, and stops when the goal is met. For example, a support agent might decide on its own to look up an order, then issue a refund, then send a confirmation email.

Because it makes its own choices, it's worth knowing exactly what's in its toolbox — which is what this tab is for.

Seeing what each agent can do

The Agents tab lists every agent in your project, before any of it is live.

• Read its purpose. Each agent has a plain-English description of what it's meant to do.
• See its tools. You can see exactly which functions the agent is allowed to reach — so you know its limits.
• Confirm before it ships. This is a design-time view: you're inspecting what an agent can do, not watching it work.

Trying an agent in the playground

Reading what an agent can do is one thing; seeing how it actually behaves is another. Open the playground to try an agent interactively — give it a task and watch it work: which tools it reaches for, in what order, and what it answers. It's a safe place to probe an agent before it's live, without affecting any real users.

Annotated screenshot of the agent playground: a chat-style input where you give the agent a task on the left, and a panel on the right showing the steps it took (tools it called, in order) and its final response. Callout: "Give an agent a task and watch how it works it out."

Agents use AI, so they may cost

An agent thinks using an AI model, so each time it runs it consumes some AI usage. That's billed like the rest of your app and shows up on your Usage page.

You choose what powers it: use Pikku Studio Cloud's built-in AI, or bring your own provider key. Either way you stay in control of what gets spent — see AI providers.

Seeing agents actually run

This page is about inspecting what an agent is allowed to do. Once your app is deployed, you'll want the live picture — its endpoint, a history of every time it ran, and who triggered it. That lives in Agents in production.

Where to go next

• Workflows — fixed-recipe jobs, the simpler cousin of agents.
• Agents in production — live endpoints, invocation history, and who ran what.
• AI providers — choose what powers your agents and control AI cost.