Daytona
ARCHIVED OSS: sandbox platform for running AI-generated code — open-source dev ended June 2026, core now closed-source
About
Daytona is a sandbox infrastructure platform for running AI-generated code: isolated environments ('sandboxes') with a dedicated kernel, filesystem, network stack and allocated CPU/RAM/disk, controllable via Python, TypeScript, Go, Ruby and Java SDKs, a CLI and a REST API. It offers container, Linux VM, Windows, macOS (via use.computer) and NVIDIA/AMD GPU sandbox classes, plus snapshots, volumes, an MCP server and secrets/governance tooling.
Status warning: the flagship open-source repo (daytonaio/daytona, 71.5k stars) was archived in June 2026 — Daytona states core development moved to a private codebase. The archived code remains usable as-is under its final license, but receives no updates or security fixes; new users should evaluate the commercial platform, not the OSS repo, as the product.
What it can do
- • Sandbox lifecycle management (create, stop, archive) via SDKs, CLI and REST API
- • Container, Linux VM, Windows, macOS (via use.computer) and NVIDIA/AMD GPU sandbox classes
- • Code execution, filesystem operations, process management, PTY and language-server support
- • Snapshots, volumes and persistent agent state across sessions
- • Secrets vault, audit logs, network limits and organization-level governance controls
- • MCP server for wiring sandboxes into AI agent clients
Setup
- Sign up at https://daytona.io and open the dashboard (app.daytona.io).
- Create an API key under Settings → API Keys.
- Install the SDK:
pip install daytona(TypeScript, Go, Ruby and Java SDKs also exist per the docs). - Set the
DAYTONA_API_KEYenvironment variable. - Create a sandbox and run code:
```python
from daytona import Daytona
daytona = Daytona()
sandbox = daytona.create()
result = sandbox.process.code_run('print("hello from daytona")')
print(result.result)
```
Prerequisites
Daytona account + API key; Python 3.8+ or Node.js for the SDKs. GPU sandboxes require a paid plan (provider-claimed).
Use cases
Works with
Trust score — 4/10
Scored 4/10 (metrics tier, metrics-v1). Evidence: 71,559 GitHub stars (log-scaled vs catalog peers); blended with ~8,365,992 monthly package downloads; repository is archived — no future fixes expected; capped at 4: OSS archived Jun 2026 — historical reference; commercial platform continues. Manual review: ARCHIVED Jun 2026 — core dev moved to private codebase; commercial platform continues; stale self-host claims. License: AGPL-3.0 (final archived tag; relicense point unverified).
★ 71,559 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.3/10
- Maintenance · recency, releases, issue load
- 2.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 4.0/10
- · 71,559 GitHub stars (log-scaled vs catalog peers)
- · blended with ~8,365,992 monthly package downloads
- · repository is archived — no future fixes expected
- · manual review: ARCHIVED Jun 2026 — core dev moved to private codebase; commercial platform continues; stale self-host claims
- · capped at 4: OSS archived Jun 2026 — historical reference; commercial platform continues
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.