code.sh — Agent-Native Execution

code.sh

Sandboxed code execution.

Run untrusted code snippets in isolated sandboxes. Short-lived, metered, observable. Perfect for tools that need a tiny VM, not a VPS.


# Run a snippet
$ curl -X POST https://code.prim.sh/v1/run \
    -H "X-402-Payment: $TOKEN" \
    -d '{"language": "python", "code": "print(1+1)"}'

{ "stdout": "2\n", "stderr": "", "exit_code": 0 }

Firecracker-style VMs

Per-run billing

CPU + memory limits

Part of agentstack

What agents use it for

User plugins

Run user-provided transforms on data without trusting their code in your main process.

Inline tools

Compile and execute small programs on demand — compilers, linters, formatters.

Secure sandboxes

Give agents a scratchpad for experiments without risking your long-lived infrastructure.

Education & demos

Let agents spin up ephemeral coding environments for tutorials and explanations.

API reference

POST   /v1/run           # Run snippet
GET    /v1/runs/:id    # Check status
DELETE /v1/runs/:id    # Cancel run

Pricing

Action	Cost	Notes
Run start	$0.001	Includes 100ms
Extra 100ms	$0.0005	Capped at 10s
Storage	Free	Ephemeral filesystem

Execution is an API call.

Agents don't SSH. They submit jobs.

Read the docs →