API reference overview

Choose the Python SDK or daemon API surface for your application.

The typed Python SDK runs over the daemon HTTP API. Choose the entry point that matches lifecycle ownership.

Choose an entry point

Need	Entry point	Ownership
Run the optimized placed trajectory	AsyncComputerSandbox.create() and ComputerSessionHandle.borrow_async()	The async owner keeps the Sandbox. The placed Function borrows it.
Use an explicit unplaced compatibility path	AsyncComputerSandbox.create_unplaced()	The caller owns the Sandbox and topology.
Use synchronous Modal primitives	ComputerSandbox	Created desktops are owned. Attached desktops are not.
Connect to an existing daemon	DaemonClient or AsyncDaemonClient	Closing the client closes its connections.
Manage several desktops	ComputerSandboxManager	Lifecycle actions remain explicit.

The sync and async borrowed clients both expose computer.step(actions). The async path is the primary model-loop surface.

Reference map

Configuration

Review public models, defaults, and environment ownership.

Namespaces

Find desktop operations and the stable Step method.

Errors and models

Identify result models and recovery failures.

OpenAPI

Inspect the checked-in daemon schema and route groups.

Import boundary

Import stable SDK types from modal_computer_use.

from modal_computer_use import (
    AsyncComputerSandbox,
    ComputerConfig,
    ComputerStepResult,
    Screenshot,
)

Core imports stay provider-free. Install only the Modal, OpenAI, or Anthropic extras your application uses.

Daemon boundary

The daemon owns desktop primitives, native input, capture, batching, receipts, and cleanup. The application owns model calls and provider state. Modal orchestration stays in the SDK owner and handoff surfaces.

Use OpenAPI for the HTTP contract. Use Namespaces for Python methods. The application API guide records the complete lifecycle and recovery contract.