Skip to main content
Novita is a cloud sandbox backend that runs your agent’s code in an isolated remote environment — no local Docker, no host access.

Quick Start

1

Install and set your key

2

Run an agent in a Novita sandbox

The user asks the agent to run code; the agent hands it to a Novita cloud sandbox, which executes it in isolation and returns the result.

How It Works

Each run starts a fresh Novita sandbox, executes the code, and cleans up. NovitaSandbox implements the standard SandboxProtocolstart, stop, execute, run_command, read_file, write_file, list_files, get_status, cleanup, and reset — so it works anywhere the other cloud backends work.

Configuration

SandboxConfig.novita() sets sandbox_type="novita" and reuses the shared sandbox settings. The sandbox honours the shared resource_limits.timeout_seconds from SandboxConfig when creating the remote environment.

Use the Backend Directly

Import NovitaSandbox when you want the isolated runtime without an Agent.

Best Practices

NovitaSandbox.is_available returns False when the key is missing, so start() fails fast with an install-and-key hint. Set the key in your environment or CI secrets.
pip install "praisonai-sandbox[novita]" pulls novita-sandbox>=2.0.0 and nothing else — keep the dependency tree slim by avoiding [all].
Novita runs code off your host, so it’s a safe default for code you don’t fully trust — unlike the local subprocess backend.
Each execute auto-starts the sandbox; call cleanup() (or rely on the agent’s auto-cleanup) so remote sandboxes don’t linger.

Sandbox Backends

All cloud and local backends and how to choose between them

praisonai-sandbox Package

Standalone package and the full backend registry