Daytona — Secure, Elastic Infrastructure for Running AI-Generated Code
Daytona is an open-source infrastructure project that lets teams securely create, control, and run isolated sandboxes for executing AI-generated or otherwise untrusted code. Built with developer ergonomics and security in mind, Daytona exposes SDKs and APIs that make it straightforward to programmatically create sandboxes, manage files and Git state, run code, and interact with language-server features.
Key features
- Fast sandbox creation: reported sub-90ms turnaround from code to an executable sandbox, enabling low-latency workflows and rapid iterations.
- Strong isolation: separated runtime and filesystem isolation to ensure AI-generated code cannot compromise host infrastructure.
- Programmatic APIs: file, Git, LSP, and execution APIs allow tight integration into automated pipelines and developer tooling.
- SDKs: first-class SDKs for Python and TypeScript (examples in the repo) to simplify integration from common stacks.
- OCI / Docker compatibility: supports arbitrary OCI/Docker images as sandbox bases so you can run code in reproducible containerized environments.
- Persistence & parallelism: sandboxes can be persisted indefinitely and the platform is designed to support high parallelization for concurrent AI workflows.
- Open-source license: distributed under the GNU Affero GPL v3 (AGPL-3), with contributor guidelines in the repository.
Typical use cases
- Safely executing model-generated code (e.g., code produced by LLMs) without risking production infrastructure.
- Running automated tests, dynamic code evaluation, and grading for coding assistants or educational platforms.
- Integrating with agent systems that need to execute arbitrary code as part of decision-making.
- Building CI-like flows where ephemeral, but fast, isolated environments are required.
Architecture & security model (high level)
Daytona centers around lightweight sandbox instances that encapsulate process, filesystem, and network controls. Isolation boundaries allow the host and other sandboxes to remain protected while executing potentially untrusted code. The platform exposes APIs to create, run, inspect, and delete sandboxes; these APIs can be called from SDKs or integrated directly into backend systems. Compatibility with OCI images means teams can bring custom runtime environments while retaining Daytona's sandboxing semantics.
Getting started & integration
Developers can sign up at the Daytona web app, create API keys, and use the provided Python or TypeScript SDKs. Common flows include creating a sandbox, uploading code or a Git repo snapshot, executing commands or scripts, and collecting outputs and logs. The repo README includes quick-start code samples for both Python and TypeScript.
Licensing & community
Daytona is open-source under AGPL-3. The project includes contribution guidelines, a developer certificate of origin, and links to documentation and community channels (Slack, X). The repo contains badges linking to docs and issue templates to streamline reporting and contribution.
For more details, follow the project documentation and the repository README (https://daytona.io/docs and the GitHub repo).
