Code Execution

What is Code Execution?

Code Execution lets your AI agents run scripts in 50+ programming languages with a single API call. It is designed for quick, stateless computations — data transformations, format conversions, calculations, and validation logic.

Under the hood, Transactional runs a self-hosted Piston engine with Linux namespace isolation to securely execute untrusted code.

Key Features

• 50+ languages — Python, JavaScript, TypeScript, Go, Rust, Ruby, Java, C, C++, and many more
• Fast execution — Pre-warmed language runtimes for near-instant starts
• Secure isolation — Linux namespace and cgroup isolation per execution
• Configurable limits — Set time and memory caps per request
• Stdin/stdout — Pass input data and capture output programmatically
• No infrastructure — No VMs to manage, no containers to orchestrate

When to Use Code Execution

Code Execution is ideal for quick, stateless tasks:

For long-running processes, package installation, or stateful workflows, use Sandboxes instead.

Code Execution vs. Sandboxes

How It Works

Your Agent
    |
    v
tx.code.run({ language: 'python', code: '...' })
    |
    v
+---------------------------+
|  Piston Engine             |
|  - Linux namespace         |
|  - cgroup resource limits  |
|  - Pre-warmed runtimes     |
|  - Stdout/stderr capture   |
+---------------------------+
    |
    v
{ stdout, stderr, exitCode, executionTime }

1. Submit — Send code and language to the API
2. Execute — Piston runs the code in an isolated namespace
3. Return — Stdout, stderr, exit code, and execution time are returned

Resource Limits

Security Model

Each execution is isolated using Linux kernel features:

• Namespaces — Separate PID, network, mount, and user namespaces per execution
• Cgroups — Hard limits on CPU time, memory, and processes
• No network — Executions cannot make outbound network requests
• No filesystem persistence — All writes are discarded after execution
• Process limits — Maximum number of spawned processes is capped

Pricing

Next Steps

• Quickstart — Run your first script
• Languages — Full list of supported languages
• Architecture — How Code Execution fits into the agent platform