While browser sessions let your agents interact with the web, computer sessions give them the power to create. In a computer session, your agent can generate code, process data, build websites, and produce files of any type. It's like giving your agent access to a development environment where it can write and execute code to accomplish tasks that pure web browsing can't handle.
Computer sessions run in isolated environments—sandboxed spaces where your agent can work without affecting anything else. This isolation means your agent can experiment, generate files, and execute code safely, with the results flowing back to your workspace as finished deliverables.
What Computer Sessions Can Do
Computer sessions unlock a range of capabilities that go beyond web navigation. They're the engine behind any task that requires generating new content or processing data programmatically.
Capabilities include:
- Build websites - Create complete HTML, CSS, and JavaScript projects from scratch
- Process data - Transform, filter, and analyze datasets using Python and popular libraries
- Generate files - Create documents, spreadsheets, scripts, and any other file type
- Run calculations - Perform complex computations, financial modeling, or statistical analysis
The key distinction from browser sessions is that computer sessions create things rather than interact with existing things. If you want your agent to visit a website, that's a browser session. If you want it to build a website, that's a computer session.
How It Works
When a task requires code execution or file generation, your agent spins up a computer session automatically. You don't need to explicitly request this—your agent determines the right approach based on what you've asked for.
The process follows a natural flow:
- Your agent recognizes that the task requires code generation or execution
- An isolated session is created in the cloud
- Any input files you've provided are made available within the session
- Your agent generates and executes code to accomplish the task
- Output files are saved to your agent's file storage
- The session cleans up automatically when work is complete
From your perspective, this happens seamlessly. You ask for something, and files appear in your agent's storage—the technical orchestration happens behind the scenes.
Visual Verification for Websites
When your agent generates a website, it doesn't just create code and hope for the best. The system includes a verification loop that ensures the output actually works.
After generating website files, your agent takes a screenshot of the rendered page and analyzes it for problems. If it detects issues—a blank screen, visible errors, broken layouts—it automatically revises the code and tries again. This cycle continues until the website renders correctly.
The verification process works like this:
- Website code is generated and rendered
- A screenshot is captured
- The agent analyzes the visual result for problems
- If issues are found, the code is revised automatically
- The process repeats until the output looks correct
This means you get working websites rather than code that might work. The agent takes responsibility for the end result, not just the code generation step.
File Management
Files flow naturally through computer sessions. Your agent has access to the files you provide, and any files it creates are automatically saved where you can access them.
The file flow works in both directions:
- Input files you share with your agent are available within the computer session
- Generated outputs are automatically saved to your agent's Files section
- You can download, preview, or reference these files in future conversations
This means you can hand your agent a CSV file and ask for analysis—the input flows into the session, and charts or reports flow back out. Or you can describe a website, and the complete project files appear in your agent's storage ready for download or deployment.
Related: Building Websites | Understanding Deliverables