What is SuperAgent

SuperAgent is a platform for building and running personal AI agents. You create agents, give them instructions, connect them to your accounts, and let them work autonomously in the background -- inside secure, containerized sandboxes.

How it works

Each agent you create in SuperAgent runs inside its own isolated container (Docker, Podman, or other supported runtimes). Inside that container, the agent is powered by Claude, Anthropic's large language model. The agent has access to a set of tools -- a shell, file system, web browser, and any external services you connect -- and uses them to carry out the tasks you describe in natural language.

You interact with agents through a chat interface. You can send messages, ask questions, assign tasks, and watch the agent work in real time. Agents remember context across sessions, learn your preferences over time, and can be scheduled to run tasks on their own.

Key capabilities

Containerized execution

Every agent runs in its own sandboxed container. The agent can execute code, install packages, read and write files, and run shell commands -- all without direct access to your host machine. This keeps your system secure while giving agents the freedom to get work done.

Connected accounts

SuperAgent integrates with dozens of external services through OAuth -- Gmail, Slack, GitHub, Linear, Salesforce, Google Drive, and many more. You connect an account once, and any agent you authorize can use it. API calls are proxied outside the container, so agents never see your OAuth tokens. You get an audit trail of every action taken on your behalf.

Browser automation

When no API is available, agents can open a web browser inside their container to interact with websites directly. You can watch the browser session in real time, and the agent can ask you for input when it encounters CAPTCHAs or login screens.

Scheduled tasks and triggers

Agents can schedule recurring or one-time tasks using cron expressions or specific dates. You can also set up webhook triggers that start an agent session when an external event occurs. This lets agents serve you autonomously in the background without manual prompting.

Dashboards and artifacts

Agents can build interactive dashboards -- small web apps rendered inside SuperAgent -- to surface information, charts, or controls. Agents can also deliver files and other artifacts directly to you during a conversation.

Multi-agent orchestration

Agents can discover, create, and invoke other agents in the same workspace. This lets you decompose complex workflows across specialized agents that collaborate to complete a task.

Skillsets

As agents develop reusable capabilities, you can package them into skillsets -- shared skill libraries backed by Git repositories. Install a skillset to give any agent instant access to a curated set of skills, and publish your own to share with your team.

Deployment options

SuperAgent runs in two modes:

• Desktop app -- Download for Mac or Windows and run locally. The desktop app bundles everything and manages containers through Docker Desktop, OrbStack, or Podman.
• Web app / server -- Run SuperAgent as a Docker container or from source and access it through your browser. Supports multi-user auth mode with role-based access control for team deployments.

Prerequisites

To run SuperAgent, you need:

1. A container runtime -- Docker Desktop, OrbStack, or Podman.
2. An Anthropic API key -- Get one from the Anthropic Console.
3. A Composio API key (optional) -- Required for connected accounts (OAuth integrations). Get one from Composio.

Next steps

• Quickstart -- Create your first agent and start a conversation.
• Core Concepts -- Understand the key abstractions: agents, sessions, containers, tools, and policies.