When to Use ORB
Use ORB when
Your workload is a persistent process that needs to stay alive across tasks.
An AI agent needs a full environment — files, packages, running processes, network. It works on tasks, waits for new ones, and picks up where it left off. ORB gives each agent its own computer that costs nothing when not actively working.
Good fits:
- A coding agent that works on a repo for hours, calling LLMs between edits
- A sales agent that runs 24/7, managing prospect pipelines, mostly waiting
- A support agent that handles conversations with long pauses between messages
- An orchestrator managing multiple sub-agents, each idle between tasks
- Any long-running process that calls external APIs and waits for responses
Don't use ORB when
Your workload is short-lived or stateless.
If you run a script, get a result, and you're done — you don't need a persistent computer. A simpler tool is better:
| Workload | Better option | Why |
|---|---|---|
| Run a script, return output | Docker, Lambda, Modal | No state to preserve. Run and discard. |
| Stateless API call | Cloud function | No persistent process needed. |
| One-shot code execution | E2B, Docker | Sandbox spins up, runs, tears down. |
| Batch processing | Lambda, Fargate, Modal | Scales horizontally, no per-process state. |
| CI/CD pipeline | GitHub Actions, Buildkite | Purpose-built for build/test workflows. |
The key question: does your process need to stay alive between tasks?
If yes — ORB. Your agent keeps its state, sleeps when idle, wakes when needed.
If no — use something simpler. Docker, Lambda, or a plain VPS.
ORB vs alternatives
| ORB | VPS | Docker | Lambda | E2B | |
|---|---|---|---|---|---|
| Persistent process | Yes | Yes | Yes | No | No |
| Sleeps when idle | Yes | No | No | N/A | No |
| State survives idle | Yes | Yes | Yes | No | No |
| Cost at 1,000 idle agents | ~$500/mo | ~$30,000/mo | ~$30,000/mo | $0 (but no state) | ~$10,000/mo |
| Full Linux environment | Yes | Yes | Yes | No | Limited |
| Sub-second wake | Yes | N/A (always on) | N/A | Cold start 1-10s | ~200ms boot |
| API-managed | Yes | No (SSH) | Docker API | Yes | Yes |
The exec trap
If you're only using ORB's exec endpoint to run one-off commands (run script → get output → done), you're not getting ORB's value. That's just a remote shell — any VPS does it.
Deploy your agent as a persistent process inside the computer. That's where ORB's advantage kicks in.
Wrong: exec "python3 review.py --pr 123" → get result → computer sits idle
Right: deploy review agent → agent receives tasks → agent sleeps between tasks → wakes instantly