Suna — Open-Source Autonomous AI Agent Platform

Introduction

Suna gives AI agents a real computer to work in — a full Linux Ubuntu sandbox with bash, a filesystem, a browser, and network access. Instead of chat-only interactions, agents can execute multi-step tasks by writing scripts, browsing websites, manipulating files, and calling external services.

What Suna Does

• Provides AI agents with an isolated Linux sandbox for executing arbitrary tasks
• Includes a built-in browser for web research, scraping, and form filling
• Supports file creation, editing, and management within the agent workspace
• Offers 60+ built-in skills and 3000+ tool integrations via connectors
• Enables scheduling via cron jobs and external triggers via webhooks

Architecture Overview

Suna consists of a Next.js frontend, a FastAPI backend, and Docker-based sandboxes. Each agent session spawns an isolated container with a full Ubuntu environment. The backend orchestrates LLM calls (supporting multiple providers), tool execution, and persistent memory storage. The agent loop follows a plan-act-observe cycle: the LLM decides on actions, the sandbox executes them, and results feed back into the next planning step. Persistent memory across sessions is stored in a PostgreSQL database.

Self-Hosting & Configuration

• Requires Docker, Node.js 18+, and Python 3.11+
• Configure LLM provider API keys (Anthropic, OpenAI, or others) in the .env file
• Set up Supabase or PostgreSQL for persistent storage and authentication
• Customize agent behavior by editing skill definitions and system prompts
• Deploy behind a reverse proxy (Nginx, Caddy) for production HTTPS access

Key Features

• Full Linux sandbox gives agents capabilities beyond text generation
• Persistent memory lets agents recall context across sessions
• Multi-channel access via web UI, API, and webhook triggers
• Built-in web browser with screenshot capture for visual tasks
• Open-source and self-hostable with no vendor lock-in

Comparison with Similar Tools

• OpenHands — coding-focused AI agent; Suna provides a broader sandbox with browser and infrastructure skills
• AutoGPT — autonomous goal-driven agent; Suna emphasizes a real computing environment over prompt chaining
• CrewAI — multi-agent orchestration framework; Suna is a full platform with UI, persistence, and sandboxing
• E2B — secure sandboxes as a service; Suna bundles the sandbox with the agent framework and UI
• Open Interpreter — runs code via natural language locally; Suna isolates execution in Docker containers

FAQ

Q: What LLM providers does Suna support? A: Suna supports Anthropic Claude, OpenAI, and other providers via configurable API endpoints.

Q: Is the agent sandbox secure? A: Each session runs in an isolated Docker container with no access to the host system.

Q: Can Suna handle long-running tasks? A: Yes. Tasks can run for extended periods, and cron-based scheduling supports recurring jobs.

Q: What license is Suna released under? A: Suna is released under the Apache 2.0 license.

Sources

• https://github.com/kortix-ai/suna
• https://suna.so/