Documentation

Quickstart

Install the Runta SDK, write a policy file, and wrap your first agent with sandbox isolation in under ten minutes.

Prerequisites

You need Python 3.9 or later and a Runta API key. If you do not yet have an account, request access on the Runta website.

Step 1: Install the SDK

Install the runta-sdk package from PyPI. It has no compiled dependencies and runs on Linux, macOS, and Windows.

terminal
$ pip install runta-sdk
Successfully installed runta-sdk-0.9.4

Step 2: Set your API key

Set the RUNTA_API_KEY environment variable. The SDK reads it automatically at startup.

terminal
$ export RUNTA_API_KEY=rnta_live_xxxx

Step 3: Write a policy file

Create a runta.yaml file in your project root. This minimal policy caps memory, CPU seconds, and declares an egress allowlist.

Policy file
version: "1"
sandbox:
  memory_mb: 512
  cpu_seconds: 30
  wall_clock_s: 120
egress:
  allow:
    - api.openai.com
    - api.anthropic.com
audit:
  retention_days: 90

Step 4: Wrap your agent

Import runta and wrap your agent invocation in a sandbox() context manager. Your existing agent code runs unchanged inside the sandbox.

agent_run.py
import runta
from my_agent import ResearchAgent
 
policy = runta.Policy.from_yaml("runta.yaml")
 
with runta.sandbox(policy=policy) as sb:
  agent = ResearchAgent(sandbox=sb)
  result = agent.run("Summarise Q2 contract risks")
 
# Sandbox torn down; audit log committed
print(sb.audit_log().summary())

Next steps

Running into an issue?

The Runta team responds to support tickets and GitHub issues. For technical issues, we typically turn around a first reply the same day.