Skip to main content
Looking for full control over the agent loop? See the Quickstart — use any LLM, any framework, and control desktops via the Tools API.

​
What are Managed Workflows?

With Managed Workflows, Nen handles the agent loop for you. You write Python automation scripts using the Nen SDK, and Nen executes them inside a managed desktop — controlling applications via keyboard, mouse, and VLM-powered vision. 
from nen import Agent, Computer, Secure

Agent

VLM-driven controller. Execute actions, verify screen state, extract structured data.

Computer

Low-level keyboard, mouse, and filesystem access inside the desktop.

Secure

Type-safe secret references. Values never enter the desktop.

run()

Entry point function. Accepts typed Pydantic params, returns typed results.

​
Core Concepts

Every Nen workflow is a Python file with a run() function. The SDK provides four building blocks:
ConceptPurpose
run()Entry point. Receives validated Pydantic params, returns validated results.
AgentAI controller that sees the screen and performs actions via natural language.
ComputerDirect keyboard, mouse, and file system access.
Secure[str]Type-safe secret handling — real values never touch the desktop.

​
How It Works

1

Write a workflow

Define a run() function with Pydantic input/output models. Use Agent for AI-driven actions and Computer for precise control.
2

Deploy to a managed desktop

Your workflow runs inside an isolated desktop environment. Nen orchestrates VLM calls to interpret the screen.
3

Get structured results

Results are validated against your Pydantic schema and returned via API. Files are packaged as assets.zip.

​
Quick Example

workflow.py
from nen import Agent
from pydantic import BaseModel, AnyUrl

class Params(BaseModel):
    website: AnyUrl

class Result(BaseModel):
    title: str

def run(params: Params) -> Result:
    agent = Agent()
    agent.execute(f"Open the browser at {params.website}")
    data = agent.extract("What is the page title?", Result.model_json_schema())
    return Result.model_validate(data)

​
Next Steps

Quickstart

Get a minimal workflow running in minutes

Agent

Learn execute, verify, and extract

Full Example

See a complete end-to-end workflow