Computer Use Component

Use the Computer Use component to give your agent a powerful, AI-driven virtual computer. It can execute complex, multi-step tasks described in natural language without writing a single line of code.

Why this matters

This component allows you to automate tasks that would normally require a human operator or complex custom code. When you provide a natural language prompt, the AI formulates an execution plan. This plan may involve using various tools like web search, data calculation, or text analysis to fulfill your request. The final result is then structured and returned as an output. This component is exclusively available for Enterprise accounts. Contact us to get access.

Step 1: Define the Task

This is the core of the component. You provide a clear, natural language instruction that tells the AI computer what you want it to accomplish.

Setting	Description
Prompt	The primary instruction that guides the AI computer's operations. You can connect this to outputs from other components to dynamically generate tasks.

Example Prompts:

• "Search for the current weather in New York and extract the temperature and conditions."
• "Find the latest stock price for Microsoft and calculate the 7-day average."
• "Look up population data for the top 10 most populous cities in California and create a summary table."

Step 2: Provide Inputs

You can make your prompts dynamic by adding custom inputs.

Input	Description
Prompt	The default input for your main instruction.
Custom Inputs	Click the + icon to add new inputs (e.g., location, company_name). You can then reference these in your prompt using {{variable}} syntax to make your tasks reusable and dynamic.

Step 3: Handle Outputs

The component provides a single, comprehensive Output and allows you to define custom outputs to easily extract specific pieces of data.

Output	Description
Output	The default output containing the complete, structured results of the executed tasks.
Custom Outputs	Click the + icon to add new outputs that extract specific data from the main Output (e.g., temperature, stock_price). SmythOS automatically understands that these custom outputs are parsing the default Output, so no complex expression is needed.

Easily Parse Results

By creating custom outputs with clear names, you can easily route different parts of the result to other components. For example, a temperature output can be sent to a notification, while a summary output can be saved to a database.

Best Practices

• Be Specific: Provide clear, detailed instructions for best results. Instead of "find company info," try "Find the founding date and current CEO of Microsoft."
• Start Simple: Begin with straightforward, single-step tasks and gradually increase complexity as you get a feel for how the component interprets prompts.
• Test Iteratively: Run your prompt in Debug Mode and refine it based on the output to improve accuracy.
• Create Dedicated Outputs: Set up specific custom outputs for each data element you need. This makes your workflow much cleaner than trying to parse a single large JSON object downstream.

Limitations

• Tasks are limited to operations that can be performed in a secure, isolated environment (primarily web-based research and data manipulation).
• The component cannot access private systems, local files, or networks without proper authentication provided through other components.
• Complex or ambiguous instructions may require refinement for optimal results. Processing time will vary based on task complexity.

Troubleshooting Tips

If your task isn't executing correctly...

• The result is inaccurate or irrelevant: Your prompt is likely too ambiguous. Add more specific details, context, and constraints to guide the AI.
• The component fails to extract data: Ensure the data you are asking for is publicly available on the web. If parsing fails, inspect the main Output in Debug mode to see the raw data structure and adjust your prompt or custom outputs accordingly.
• The task is too complex: If a very complex prompt fails, break it down into smaller, sequential tasks using multiple Computer Use components or other tools.

What to Try Next

• Feed the raw Output into a GenAI LLM Component to have it write a narrative summary of the structured data.
• Use this component as a more powerful alternative to the basic Web Search Component for complex research tasks that require multiple steps.
• Use a JSON Filter Component to further refine or clean the structured Output before passing it to other parts of your workflow.