Runtime monitoring
See your Agent's open ports, CPU and memory usage, and sandbox status — set its sleep timeout and restart it when needed.
The Agent's cloud computer is a real machine, and sometimes you need to look under the hood: which ports are open, how busy it is, whether it's awake — and the ability to restart it. You'll find this under the Agent's Settings → Computer, plus port discovery built into the workbench.
Forwarded ports
When the Agent starts a server, Buda detects the listening port so you can open it. In the workbench, the Web Preview flow lists open ports and turns a sandbox localhost link into a reachable preview URL. Open ports refresh as servers start and stop.
Status, CPU, and memory
The Computer settings panel shows the live sandbox:
- Status —
running, or sleeping when idle (see below) - Resource spec — the vCPU / RAM tier the sandbox runs on
- CPU and memory usage — current utilization while running
- Volume — storage type, size used, and file count for the Agent's Drive
Workspace directories
The sandbox mounts a few well-known paths:
| Path | Purpose |
|---|---|
/agent | The Agent's persistent working directory — survives restarts |
/space | Shared Space storage |
/system | System files managed by Buda |
/tmp | Scratch space — cleared on restart |
Idle timeout (sleep)
To save resources, the sandbox sleeps after a period of inactivity and wakes automatically the next time you or the Agent use it — your terminal, files, and processes pick up where they left off. Set how long it waits in Settings → Computer → Sleep timeout, from 5 minutes up to 8 hours.
Longer sleep timeouts are gated by plan. If an option is locked, an Upgrade prompt appears next to it.
Restarting the sandbox
If the environment gets into a bad state, click Restart sandbox in Settings → Computer. Buda restarts (or recreates) the machine while keeping the Agent's persistent /agent Drive intact. Anything in /tmp and any running processes are lost, so restart only when you need a clean slate.
Related pages
Environment variables
Give an Agent the API keys and config its sandbox needs — set per-Agent environment variables once and they're available everywhere it runs.
Create an agent
Hire a new agent — name it, pick its model and reasoning effort, give it skills — then put it to work in its own cloud computer.