Click the New Workflow button to open the workflow editor.
3
Define your workflow
Fill in the workflow details:
Name (optional): Give your workflow a descriptive name like “Extract Patient Data”
Main Prompt: This is the instruction that tells AI what to do. For example:
Navigate to the patient portal and extract the demographics informationfor patient ID: {patient_id}. Look for patient name {patient_first_name}{patient_last_name} to confirm you have the right record.
You can also add images to the prompt from the dashboard editor (click Add
Image). Images help the agent understand tricky icons and UI elements in
legacy apps.
Input Schema (optional): Define expected run inputs with JSON Schema. This validates the merged run payload (input_values + sensitive_input_values + machine/session values) before execution. In input_schema, sensitive root keys use a $ prefix (for example, "$api_key").
Output Schema (optional): Define the structure of data you want back using JSON Schema
Variables in Workflows:
Use {variable_name} for input variables - values you pass when starting the workflow
Use {{runtime_variable}} for runtime variables - values discovered and set during execution by focused_action
Use {$variable} for sensitive variables - pass values at run creation via sensitive_input_values. Sensitive values are stored in a secure vault during the run, never logged or sent to LLMs, resolved only during actual computer actions, and deleted immediately after the run completes.
Example: "Search for {patient_name} and save their ID as {{patient_id}} for later use"
AI Assist Feature: Instead of writing prompts manually, you can use our AI assist feature:
Type a natural language description of what you want to automate
Click Apply
AI will create a detailed prompt for you
You can send follow-up instructions to refine the prompt - AI uses the current field contents as context
The same AI assist is available for both Input Schema and Output Schema fields to help you define JSON schemas quickly.
4
Save your workflow
Click Create Workflow to save. You’ll be redirected to the workflow details page where you can find your workflow ID.
Cyberdriver connects your desktop to Cyberdesk, enabling remote automation.
Cyberdriver will automatically request administrator privileges when it runs.
Make sure you have the necessary permissions or administrator password
available, as this is required for automating certain desktop applications.
1
Run the installer in PowerShell
# Create tool directory$toolDir = "$env:USERPROFILE\.cyberdriver"New-Item -ItemType Directory -Force -Path $toolDir# Download cyberdrivertry { Invoke-WebRequest -Uri "https://github.com/cyberdesk-hq/cyberdriver/releases/download/v0.0.41/cyberdriver.exe" -OutFile "$toolDir\cyberdriver.exe" -ErrorAction Stop} catch { Write-Host "ERROR: Failed to download Cyberdriver. If Cyberdriver is already running, run 'cyberdriver stop' first. Otherwise, check your internet connection and try again." -ForegroundColor Red return}# Verify installationif (Test-Path "$toolDir\cyberdriver.exe") { $fileSize = (Get-Item "$toolDir\cyberdriver.exe").Lengthif ($fileSize -gt 34MB) { # Add to PATH if not already there $userPath = [Environment]::GetEnvironmentVariable("Path", "User") if ($userPath -notlike "_$toolDir_") {[Environment]::SetEnvironmentVariable("Path", $userPath + ";" + $toolDir, "User")}Write-Host "Cyberdriver installed successfully! You may need to restart your terminal for PATH changes to take effect."} else {Write-Host "ERROR: Download appears incomplete (file too small). Please try again." -ForegroundColor Red}} else {Write-Host "ERROR: Download failed. Please try again." -ForegroundColor Red}
2
Restart PowerShell
Close and reopen your PowerShell terminal for the PATH changes to take effect.
Now let’s execute your workflow programmatically using the SDK.
TypeScript
Python
1
Install the SDK
npm install cyberdesk
2
Create and run your workflow
import { createCyberdeskClient } from 'cyberdesk';async function runWorkflow() { const client = createCyberdeskClient('YOUR_API_KEY'); const { data: run, error } = await client.runs.create({ workflow_id: 'your-workflow-id', // From Step 1 machine_id: 'your-machine-id', // Optional: omit to auto-select an available machine input_values: { patient_id: '12345', patient_first_name: 'John', patient_last_name: 'Doe' } }); if (error || !run) { console.error('Failed to create run:', error); return; } console.log('Run created:', run.id); let currentRun = run; while (currentRun.status === 'scheduling' || currentRun.status === 'running') { await new Promise((resolve) => setTimeout(resolve, 5000)); const { data: updatedRun, error: refreshError } = await client.runs.get(currentRun.id); if (refreshError || !updatedRun) { console.error('Failed to refresh run:', refreshError); return; } currentRun = updatedRun; console.log('Status:', currentRun.status); } if (currentRun.status === 'success') { console.log('Results:', currentRun.output_data); } else { console.error('Run failed:', currentRun.error?.join(', ')); }}runWorkflow();
1
Install the SDK
pip install cyberdesk
2
Create and run your workflow
from cyberdesk import CyberdeskClient, RunCreateimport asyncioasync def run_workflow(): client = CyberdeskClient('YOUR_API_KEY') # Create a run run_data = RunCreate( workflow_id='your-workflow-id', # From Step 1 machine_id='your-machine-id', # Optional: omit to auto-select an available machine input_values={ 'patient_id': '12345', 'patient_first_name': 'John', 'patient_last_name': 'Doe' } ) response = await client.runs.create(run_data) if response.error: print(f'Failed to create run: {response.error}') return run = response.data print(f'Run created: {run.id}') while run.status in ['scheduling', 'running']: await asyncio.sleep(5) response = await client.runs.get(run.id) if response.error: print(f'Failed to refresh run: {response.error}') return run = response.data print(f'Status: {run.status}') if run.status == 'success': print('Results:', run.output_data) else: print('Run failed:', ', '.join(run.error or []))asyncio.run(run_workflow())
If your workflow defines an Input Schema and provided inputs do not match, the API returns a 422 validation error with path-level details so you can fix the payload before retrying.
Output Data: The extracted/processed data based on your output schema
Message History: Complete conversation between AI and your desktop
3
Generate and approve workflow trajectories
From the run details panel:
Click Generate Trajectory (when available) to promote that run’s captured path into your workflow trajectory library
Then navigate to your workflow’s Trajectories tab to:
Review step-by-step actions with screenshots
Approve trajectories to enable fast cached execution on future runs
Trajectories are Cyberdesk’s intelligent caching system. Capture happens in the background during runs, generation makes a trajectory visible/reviewable, and approval enables replay. Learn more in Trajectories 101.
