Overview
Daytona is infrastructure purpose-built for running AI-generated code. Each sandbox is a fully isolated environment — its own kernel, filesystem, and network stack — that boots in well under a second, so an agent can create one per task, run code, and discard it.
flowchart LR code[LLM code] --> daytona[Daytona Sandbox] daytona --> exec[(isolated env)] daytona --> out[stdout / results]
It exposes Python and TypeScript SDKs for spinning sandboxes up and down, with state and filesystem access inside. The core is open (AGPL-3.0) and there’s a hosted offering for teams that don’t want to run it themselves.
When to use it
Choose Daytona when an agent runs many short-lived, untrusted code executions and you want strong per-task isolation with sub-second startup — a code interpreter, a data-analysis step, or an autonomous coding agent’s workspace.