Runloop is winding down its services. Its documentation says support for existing customers runs through November 13, 2026 at 11:59 PM PT. If your agents run in Runloop devboxes, they need a new home before then.
This guide maps each Runloop piece to SuperHQ and walks through the move.
Two ways to run on SuperHQ
Nanite is a sandbox API. Each sandbox is a microVM with its own Linux kernel: Debian 13 with root, 2 vCPU, 2 GiB of memory and a 20 GiB disk, created over plain HTTP from any language. Docker runs inside it. A sandbox stops after 30 idle seconds and wakes on the next request, with its disk as it was.
Durables are agents we run for you, each on its own Nanite sandbox. A durable has a private chat at its own address, an API, Telegram, schedules and webhooks. Every turn is written to its disk, so after a sleep it picks the conversation up where it left off. It runs Claude, GPT or any model on OpenRouter with your own key, and we do not charge for tokens.
If you built your own agent loop on devboxes, Nanite is the direct replacement. If a devbox was the home of one long-lived agent, look at durables.
What each Runloop piece becomes
Runloop does not transfer devboxes, blueprints, snapshots or Axon data, so plan to rebuild rather than copy.
| Runloop | SuperHQ | Notes |
|---|---|---|
| Devbox | Nanite sandbox | A microVM with its own kernel, arm64, Debian 13 as root. |
| Command execution | Exec | Over HTTP, or as a session over a WebSocket that you can attach to again. Stdout, stderr and the exit code. |
| File read and write | Filesystem routes | Read, write and watch files in the sandbox. |
| Suspend and resume | Stop and wake | A sandbox stops when idle and wakes on the next request with its disk. Processes start fresh, and a start command can bring your server back. |
| Disk snapshot | Checkpoint | Save the disk of a sandbox and restore it when something goes wrong. |
| Tunnel | Preview URL | Every sandbox has an HTTPS address on nanite.page, open to your workspace only by default, or to anyone. |
| Credential gateway | Secrets | The sandbox only sees a stand-in. The proxy puts the real value in on the way out, and only to the hosts you name. |
| Blueprint | A setup script | Run your setup on create. If installs are slow, pack the prepared environment once and unpack it on create. |
| Axons | Your own event layer | Run a broker such as Redis or NATS in a sandbox, or use a hosted one. Preview URLs take webhooks and wake the sandbox, schedules cover timed events, and exec sessions stream output. |
| Benchmarks | Your own harness | Point it at the Nanite API: a sandbox per task, run the task, collect the results. Docker runs inside, so harnesses built on containers work. |
Pricing
Prepaid credits, with no seats, plans or minimums. A sandbox is billed for the time it runs and for its disk while it sleeps.
| What | Price |
|---|---|
| CPU | $0.072 per vCPU hour |
| Memory | $0.0432 per GB hour |
| Stored disk | $0.15 per GB month |
| Data sent out | $0.15 per GB |
A durable is billed as the Nanite sandbox it runs on. Model calls go from the sandbox to your provider on your key.
How to move
- Export before November 13. Download the files you need from your devboxes, and write down your blueprint steps, environment variables and which secrets go to which hosts.
- Create a workspace at app.superhq.ai and buy credits.
- Rebuild the environment once. Start a sandbox, install your toolchain, and keep the steps as a script you run on create.
- Swap your calls. Replace create, exec, file and lifecycle calls with the Nanite API. Your agent logic stays as it is.
- Check before you cut over. Compare exit codes, file round trips, what runs after a wake, and network rules.
- Move your traffic.
We can help
Write to support@superhq.ai with the subject “Runloop migration”, or book a call. Bring your blueprint and your workload, and we will help you rebuild them on SuperHQ.