Skip to main content

Agent Swarms

Take on big, multi-part projects by letting a team of AI agents work together in parallel.

An Agent Swarm turns a single complex request into a coordinated team of AI agents. Instead of one agent doing everything one step at a time, a Swarm splits the project into independent pieces and works on several of them at once — so you get higher-quality outcomes, faster.

This guide covers:

  1. What Is an Agent Swarm?
  2. When to Use an Agent Swarm
  3. When Not to Use an Agent Swarm
  4. How to Trigger an Agent Swarm
  5. Credits: Limits and Consumption

What Is an Agent Swarm?​

An Agent Swarm is a way of running a complex request with multiple AI agents working together instead of just one. There are two roles:

  • Master Agent — the coordinator. It understands your goal, breaks the project into smaller tasks, works out which tasks depend on which, hands each task to an agent, keeps track of how everything is progressing, passes information between agents when one needs another's output, and brings the finished pieces together into a final result for you.
  • Worker Agents — the specialists. Each Worker Agent is a full, independent Abacus AI Agent that takes on one task — for example, building an app, producing a presentation, or writing a research report. Because Workers run at the same time rather than one after another, the whole project finishes sooner.

Think of it like a small project team: a manager who plans, delegates, and reviews, plus specialists who each focus on their own part of the job. The Master Agent handles all the coordination behind the scenes — you only need to describe what you want and approve the plan.

When to Use an Agent Swarm​

An Agent Swarm is the right choice when your request naturally breaks into multiple independent deliverables that can be built side by side. In broad terms, that means:

  • Multiple applications. When you want two or more separate apps built at once — for example, a web app and a mobile app.
  • Multiple independent deliverables of any kind. Any request that splits into several distinct pieces that don't depend on one another — such as multiple presentations, or an app together with a separate research report, data analysis, or slide deck.

In short: if the work amounts to several distinct things that don't depend on one another, a Swarm lets a team of agents tackle them in parallel for faster, higher-quality results.

When Not to Use an Agent Swarm​

For everything else, a single Abacus AI Agent is the better fit — it's quicker to start and uses fewer credits. You generally don't need a Swarm when your request is:

  • A single application — no matter how complex. One app is still one app, even if it has its own backend, API, database, several user roles, and many pages. A single agent builds the whole thing end to end.
  • A single presentation — no matter how many slides. One slide deck is handled by one agent, regardless of length or number of sections.
  • Research or data analysis on its own. Even multiple research studies or analyses are handled efficiently within a single agent — a Swarm isn't needed.
  • A research report plus a single presentation. One agent can do the research first and then produce the deck.
  • Everyday tasks like chat and Q&A, writing or document generation, image or media generation, code editing, review, or debugging.
  • Design-only work such as wireframes, mockups, screen designs, or landing-page visuals — even when they mention multiple platforms or screen sizes. These are design tasks, not multiple applications.

If you're unsure, just start in a normal chat — Abacus AI Agent will suggest a Swarm if it thinks your request would benefit from one.

How to Trigger an Agent Swarm​

There are two ways to start a Swarm.

1. Accept the automatic suggestion​

When you begin a new conversation and your first request looks complex enough to benefit from teamwork, Abacus AI Agent shows a card titled "Recommended Approach: Use an Agent Swarm." It explains the benefit and gives you two choices:

  • Start an Agent Swarm — run your request as a coordinated multi-agent project.
  • Ignore, continue in this chat — skip it and let a single agent handle the task as usual.

The suggestion is always optional. If you ignore it, your task is still completed by a single agent.

2. Select Agent Swarm yourself​

You don't have to wait for the suggestion. On the home screen, choose the Agent Swarm option from the list of categories. This starts your next request in Agent Swarm mode directly, so a coordinated team of agents takes on the work.

Credits: Limits and Consumption​

Because a Swarm runs several agents in parallel, it uses more credits than a single-agent chat. Keep the following in mind:

  • Typical consumption. How much a Swarm uses depends on how many deliverables are involved and how complex they are. Simpler projects with fewer Worker Agents cost less.
  • You need enough balance to start. A Swarm requires a sufficient credit balance to launch. If you ask for one but don't have enough credits, you'll be prompted to use a simpler agent mode or top up your credits first.
  • If credits run out mid-project. If your balance is exhausted while a Swarm is running, it stops and lets you know that you need to purchase more credits to continue.

Still Need Help?​

If you're still stuck, contact support — we're here for you.