Skip to main content
An agent in OpenSwarm can do anything you can do on your computer — read and write files, run commands, search the web, control a browser, send emails, manage your calendar — and handle long-running, multi-step tasks autonomously. Think of each agent as a teammate you can brief on a task and let loose, while you watch it work in real time.

​
What makes an agent different from a chatbot

A regular chatbot takes a prompt and returns text. An OpenSwarm agent takes a prompt and acts on it — it plans, uses tools, reads results, adapts, and keeps going until the job is done or it needs your input. Agents can:
  • Read, create, and modify files across your entire filesystem
  • Browse the web via a real browser — navigating, clicking, filling forms
  • Use integrations — Gmail, Google Calendar, Drive, Sheets, X/Twitter, Reddit, and any custom integrations you connect
  • Spawn sub-agents to parallelize work or delegate specialized tasks
  • Generate interactive apps that render live on the dashboard

​
Agents in the context of OpenSwarm

OpenSwarm isn’t about running one agent — it’s about running many at once. Each agent lives as a card in a workspace canvas. You can have infinite agents working simultaneously. The workspace gives you a bird’s-eye view of everything that’s happening.

​
The swarm model

  1. You launch an agent from the toolbar and give it a task
  2. That agent can spawn sub-agents, browsers, and execute actions as needed.
  3. Permissions are set globally so you control what exactly an agent can or can’t do.
  4. When human approval or insight is needed, a popup appears within that agent’s chat.
In the center of the Open Swarm application’s header, you can also find the floating island bubble. This gives you a quick overview of all agent states. If human input is needed, this floating island will morph into an input field that you can interact with.

​
Modes

Agents run in a mode that controls their behavior — which tools are available, what system prompt they use, and how they approach tasks. OpenSwarm ships with built-in modes:
  • Agent — Full access to all tools. The default general-purpose mode.
  • Ask — Conversational only, no tool use. For quick questions.
  • Plan — Thinks through a plan before acting. Read-only tools only.
  • App Builder — Specialized for generating interactive Views.
  • Skill Builder — Specialized for creating reusable skills.
You can also create custom modes with your own system prompts and tool restrictions. Switch modes at any time from the chat input.
99% of the time, the default “Agent” mode is all you need. Modes are helpful when you want holistic, re-usable, presets for how agents behave.

​
Models

Each agent is powered by an Ai Model which you can select. Here is a brief overview of the models:
ModelBest for
Sonnet 4.6Fast, capable, good default for most tasks
Opus 4.6Most capable, best for complex reasoning and long-horizon tasks
Haiku 3.5Fastest and cheapest, good for simple tasks
You can switch models mid-conversation — the new model takes effect on the next message.

​
Integration approvals

When an agent wants to use an integration, it may need your permission depending on how that integration is configured:
  • Allow — runs automatically, no approval needed
  • Ask — pauses and shows you exactly what it’s about to do (the command, the file, the API call) so you can approve or deny
  • Deny — blocked entirely
You can set these permissions in the Integrations page.

​
Cost tracking

Every agent session tracks token usage and cost in real time. A context window ring in the chat input shows how much of the context window has been consumed.