Sandbox use cases
This page lists six kinds of agent work: coding agents, a code interpreter, computer use, browser automation, data analysis and running untrusted code. Each one has a page with what it needs from a sandbox and a table of all 79 providers.
Updated
- 79providers
Coding agents
A coding agent works for hours, keeps files between steps and often needs to go back to an earlier state.
+73 more$0.001from, per vCPU-hour - 79providers
Code interpreter
A code interpreter runs many short pieces of code that a model writes, one request at a time, and returns the result to the chat.
+73 more$0.001from, per vCPU-hour - 79providers
Computer use
A computer-use agent looks at a screen and controls a mouse and a keyboard, so it needs a sandbox with a graphical desktop.
+73 more$0.001from, per vCPU-hour - 79providers
Browser automation
Browser automation runs a headless browser that an agent drives through pages, forms and downloads.
+73 more$0.001from, per vCPU-hour - 79providers
Data analysis
Data analysis loads files into memory, runs code a model writes over them and returns tables and charts.
+73 more$0.001from, per vCPU-hour - 79providers
Running untrusted code
Untrusted code is code you did not write and cannot review first, such as code from a model or from a user of your product.
+73 more$0.001from, per vCPU-hour
Also browse by
Other ways to group the providers