API reference overview

Choose the modal-computer-use v1.1.0 Python SDK or daemon API surface.

Version 1.1.0 provides a typed Python SDK over the daemon HTTP API. Choose an entry point based on who owns the desktop lifecycle.

Choose an entry point

Need	Entry point	Lifecycle behavior
Create or attach to a Modal desktop from synchronous Python	ComputerSandbox	Created desktops are owned. Attached desktops are not.
Create or attach without blocking an event loop	AsyncComputerSandbox	Async contexts apply the same ownership rules.
Connect to an existing daemon	DaemonClient or AsyncDaemonClient	Closing the client closes connections only.
Borrow an owned desktop inside a deployed Modal Function	ComputerSessionHandle.borrow_async()	The original owner keeps the Sandbox.
Manage several Modal desktops	ComputerSandboxManager	Lifecycle actions remain explicit.

from modal_computer_use import ComputerConfig, ComputerSandbox

with ComputerSandbox.create(config=ComputerConfig()) as computer:
    computer.mouse.move(100, 120)
    screenshot = computer.screenshots.full()

Use the async surface when your application already uses asyncio:

from modal_computer_use import AsyncComputerSandbox, ComputerConfig

async with AsyncComputerSandbox.create(config=ComputerConfig()) as computer:
    await computer.mouse.move(100, 120)
    screenshot = await computer.screenshots.full()

The sync and async entry points expose parallel namespace methods. Async methods add await but keep the same operation names and result models.

Reference map

Configuration

Review public SDK models, defaults, and environment ownership.

Namespaces

Find each desktop operation by namespace.

Errors and models

Identify stable result models and failure types.

OpenAPI

Inspect the checked-in daemon schema and route groups.

Import boundary

Import stable top-level SDK types from modal_computer_use.

from modal_computer_use import (
    AsyncDaemonClient,
    ComputerConfig,
    ComputerSandbox,
    Screenshot,
)

Core imports do not require Modal, OpenAI, Anthropic, or provider credentials. Install the modal extra before you enter a Modal provisioning or attachment context.

Daemon boundary

The daemon exposes health, readiness, version, capability, and /v1/* primitive routes. The Python clients own connection pooling and typed serialization. They do not add a provider-owned model loop.

Use OpenAPI when you need the HTTP contract. Use Namespaces when you write Python.

The v1.1.0 API guide records lifecycle, handoff, action, and provider-adapter contracts.