Clusterfudge Sandboxes
Computers for AI agents
Give agents secure, scalable computers of their own. Full UI control, terminal access, file operations, and process management in an isolated environment.
New!See Clusterfudge Sandboxes in actionPersistent
Maintains state across sessions
Sandboxed
Each AI gets its own sandboxed instance
Burstable
Scales from 1 to 1000s of sandboxes in seconds
Customisable
Pre-configure your image, apps, and dependencies
Complete control
Everything an agent needs to get work done
Clusterfudge Sandboxes provide full-featured computers for AI agents to work autonomously and securely.
UI interaction
Full keyboard and mouse control for agents to navigate applications just like humans. Support for clicks, drags, scrolling, and complex key combinations.
Terminal access
Secure command execution with isolation between agents. Execute any command with persistent history and full output capture.
File operations
Create, view, edit, and download files. Perfect for coding tasks, data analysis, and document creation with full file system access.
Process management
Launch and control processes with stdin/stdout streams. Monitor exit codes and kill processes when needed.
Interactive process communication
Write to process stdin, read from stdout/stderr, and monitor exit codes in real-time. Perfect for interactive application control and script execution.
File transfer capabilities
Download files/folders from sandboxes and upload content directly. Enables seamless data exchange between sandboxes and your systems for training.
Panopticon
Monitor all your sandboxes in real-time with the Panopticon dashboard. Have visibility into every action taken by your agents, with full audit trails and the ability to take direct control when needed.
Enterprise-ready
Secure, scalable, and manageable
Built with security, performance, and observability as core principles. Deploy with confidence in production environments.
Persistent state
Sandbox state persists across sessions. Your agents can pause work and resume later without losing progress.
Security isolation
Each agent gets its own isolated environment. Strong authentication with token-based access control for secure operations.
Burstable scaling
Scale from 1 to 1000s of sandboxes in seconds. Handle peak load with automatic provisioning and shutdown to optimise costs.
Full customisation
Pre-configure your image with apps, dependencies, and starting state to give agents exactly what they need from the start.
Complete audit trail
Full history of all actions taken by agents. Review terminal commands, file changes, and UI interactions for transparency.
Human oversight
Take control at any time through the Panopticon dashboard. Observe active sandboxes and intervene when necessary.
Simple to integrate
- Powerful Python SDK with intuitive API
- Control UI, terminal, file system, and processes
- Seamless integration with existing AI pipelines
- Authentication tokens for secure access control
- Persistent storage across sessions
Endless possibilities
Clusterfudge Sandboxes enable AI agents to accomplish complex tasks that require real computer interaction.
Data analysis
Let agents use data science tools to analyze complex datasets, generate visualisations, and run simulations.
Software development
Enable agents to write, test, and debug code in real development environments with proper tools and dependencies.
Web research
Agents can browse the web, download content, and synthesize information from multiple sources.
Document processing
Process and create documents, spreadsheets, and presentations using real office applications.
Automation testing
Test your applications with AI agents that can interact with your UI just like human testers would.
Content creation
Generate images, videos, and other media using professional creative applications.
Start a sandbox in seconds
Join leading AI companies building agents on top of Clusterfudge Sandboxes.