A private Linux computer for every chat
The sandbox in Definable is a private Linux computer that each chat gets when you turn it on. The agent runs code and commands, works with files and installs what the job needs. The computer stops after 15 minutes idle and is deleted 30 days after it was last used.
How the sandbox works
Turn on the sandbox and each chat gets a private Linux computer, powered by Daytona, where the agent runs code, works with files and hands results back.
At a glance
- Small, Medium or Large
- Debian · Python 3.11 · Node 22 · git
- Internet access
- Kept 30 days after last use
- 01
One machine per chat
It starts in about a second and stops after 15 minutes idle. Files and installed packages stay between messages, for 30 days after the chat's last use; running processes don't.
- 02
Real work, not snippets
Shell commands and background jobs, reading, writing and searching files, opening what people attach, installing tools with apt, a preview link to a running server for an hour, and files handed back as downloads.
- 03
Set up once
Ask Builder to install what the agent's chats need. It's saved as the agent's environment, publishing freezes it, and Reset to base undoes it.
- 04
Secrets stay secret
Keys like GH_TOKEN reach your team's chats and scheduled tasks as environment variables, never people on channels or the API, and any output shows them as [secret NAME].
Sandbox FAQ
What is the sandbox?
Each conversation gets its own Linux computer where the agent can run code and commands. It's kept between messages, stops after 15 minutes idle and is deleted 30 days after it was last used. Secrets you add become environment variables, hidden in the output, and only your team's own chats and schedules get them.
What uses credits?
Credits pay for what runs on our side: Builder, and the sandbox computer by the second. Model calls on your own keys never use credits.
What are skills?
A skill teaches an agent how your team does a task: instructions, and sometimes scripts, in a folder. Add one from the open catalog at skills.sh, upload your own, write one in the editor, copy one from a public GitHub repo or ask Builder to draft it. Catalog skills show what five security auditors found, and the agent's owner approves every skill.