CPU Sandbox gives an AI agent an isolated, programmable computer for running code, installing tools and carrying out multi-step work, without running that code directly in your application environment. It is part of the Telnyx Agent Platform and is in beta.
What's new
- Isolated runtime. Each Sandbox has its own environment, isolated from your main app and from other customers.
- Work that persists. Keep a working environment across multiple agent steps, instead of rebuilding it every time.
- Pause and resume. Checkpoint running work and resume it later, including machine state, not only saved files.
- Fork. Clone a working environment to try several approaches in parallel.
- Programmatic control. Create a Sandbox, execute code, inspect files and manage its lifecycle through the Telnyx API or the telnyx-edge CLI. Agents built with the Agent SDK or deployed through Agent Control Plane create and manage Sandboxes the same way.
Why it matters
Models can generate code. Running it is the risky part. Your agent lives on Telnyx Edge Compute, always on and holding its memory and credentials. A Sandbox is a separate computer the agent creates on demand: install a library, open a file it did not write, run the script, return the result. Keep it available across tasks, and terminate it when it is no longer needed. If the code fails, it does so in an environment isolated from your agent's runtime.
Getting started
- Create a Sandbox: telnyx-edge sandbox-cpu create --image docker.io/library/python:3.12-slim --name doc-quick-start --timeout 1800
- Run code in it: telnyx-edge sandbox-cpu exec doc-quick-start -- python3 /workspace/primes.py
- Pause it, resume it, then terminate it: telnyx-edge sandbox-cpu pause doc-quick-start, telnyx-edge sandbox-cpu resume doc-quick-start, telnyx-edge sandbox-cpu terminate doc-quick-start
You need a Sandbox-enabled account and the telnyx-edge CLI. Every step also has an API call in the quick start.
Read the CPU Sandbox docs.












