Suna

Suna is an open-source generalist AI agent that browses the web, runs code, manages files, and completes real tasks through a conversational interface.

Overview

Suna pairs a capable agent runtime with an unusually complete toolset: a sandboxed computer the agent can control, browser automation, file management, code execution, and web search, all orchestrated through natural conversation. You describe an outcome — research competitors and build a spreadsheet, plan a trip with bookings, crawl and analyze a site — and Suna decomposes it, works through each step, and shows its actions transparently. Being open source, the whole stack can be self-hosted, extended with custom tools, or integrated into other products. The project has attracted a lively community that shares agent recipes and improvements, and its clean architecture makes it a popular base for teams building their own specialist agents on top of a proven generalist foundation. Its agent operates a real virtual computer rather than a narrow set of APIs, which means it can handle messy multi-application jobs — comparing spreadsheets, filing downloads, cross-checking websites — that brittle workflow tools cannot express.

Key Features

• Sandboxed computer control including browser, shell, and file system
• Transparent step-by-step execution with interruptible task plans
• Self-hostable open-source stack with custom tool integration
• Community-shared agent recipes for research, ops, and analysis jobs
• Conversation interface that accepts files, links, and follow-up steering
• Full virtual computer control that survives messy multi-application jobs

Pricing

Suna is free and open source, so self-hosting costs only infrastructure. A hosted cloud option charges for agent runtime minutes and model usage, with free trial credits for evaluation.

Who Is It For

Developers and power users who want a genuinely autonomous generalist agent they can inspect and extend, and teams seeking a proven open foundation for domain-specific agents. It is well suited to technical users evaluating open agent frameworks, and to teams that would rather adapt a working generalist than build an agent stack from zero.