Execution Sandbox
Run agents in a secure sandbox with streaming output, keyboard shortcuts, and execution history
Execution Sandbox
The Execution Sandbox provides a secure, isolated environment for running AI agents. This guide covers the sandbox interface, keyboard shortcuts, and execution management.
Running an Agent
Open the sandbox
Navigate to any agent's detail page and click Run Agent.
Enter your input
Type or paste your input in the message field at the bottom of the sandbox.
Submit and watch
Press Enter or click Send. The agent processes your input and streams the response in real-time.
Streaming Output
The sandbox uses Server-Sent Events (SSE) for real-time output streaming:
- Responses appear as they are generated, token by token
- A progress indicator shows when the agent is processing
- Error messages appear inline if something goes wrong
Keyboard Shortcuts
Power users can use keyboard shortcuts for efficient interaction:
| Shortcut | Action |
|----------|--------|
| Ctrl + Enter | Send message |
| Ctrl + L | Clear conversation |
| Ctrl + / | Show shortcut help |
| Escape | Cancel current execution |
| Ctrl + K | Focus input field |
Press Ctrl + / anytime to see the full list of available shortcuts.
Environment Profiles
Save and switch between different execution configurations:
- Default — Standard execution settings
- Debug — Verbose output with intermediate steps
- Production — Optimized for speed
Profiles store:
- LLM provider and model selection
- Temperature and other generation parameters
- Timeout settings
- Environment variables
Execution History
View past executions for any agent:
- Full input/output logs for each execution
- Execution timestamp and duration
- Success/failure status
- Cost attribution (if pricing is configured)
Access history from the agent detail page under the Analytics tab, or from your Personal Dashboard.
Security
Each execution runs in an isolated Docker container:
- Container Isolation — No access to host system or other users' data
- Resource Limits — CPU and memory bounds prevent runaway processes
- Timeout — Configurable time limits (default: 5 minutes)
- Audit Logging — All executions are logged for security review
Next Steps
- Pipelines Guide — Chain multiple agents together
- Collections Guide — Organize your favorite agents