What we log
Everything homemade records about your VMs and why, and everything it never records.
Disposable VMs raise a fair question: what does the platform see, and what does it keep? This page lists everything we record, why, and what we never record.
What we record, and why
Domain facts only: enough to bill accurately, debug a broken boot and catch abuse.
- Lifecycle events, with who and why. Created, booted, asleep and more. Each one is tagged with the user, admin, PR, ticket or system that triggered it.
- Which hostname served traffic, and when you attached a terminal. A host-level match off the proxy's access feed: hostnames only, never URLs, paths or request bodies.
- Egress bytes and new-connection counts. Per-VM counters for billing and abuse detection. We never log packet contents. The scan-target set is an in-kernel counter that expires in 60 seconds, not a log.
- Outbound connection metadata. Which VM opened a connection to which destination IP and port, per minute, and the source port it left from on the shared host address. Never what was sent. Kept for 7 days, so we can answer an abuse report about a shared host IP.
- Boot-step outcomes and capped install logs. Which setup step succeeded or failed, plus secret-redacted, size-capped install and build output, so you can debug a broken boot.
- Live resource gauges. CPU is sampled on the host; memory and disk are read from the guest on demand, not stored. Just enough to draw the runtime panel.
- An audit log of account and admin actions. Actor, action, resource, IP and user agent, with secret-redacted diffs, for account security.
- Operational wide events. Domain fields only, sent to our logging backend, kept for 30 days. Your email address is never logged.
What we never record
These never reach a log, a database row or our monitoring backend.
- Terminal contents. A reattach replay is serialized inside your own VM's memory. It dies with the VM and never touches us.
- AI prompts and conversations. Chat transcripts live on your VM's own disk, and your subscription's AI traffic goes straight from the VM to the provider. It never passes through us.
- Your code or file contents. We orchestrate the VM; what's inside it is yours.
- Secret values. Encrypted at rest, write-only through the API. We can store them but never read them back.
- Email bodies. We send the email; we don't keep a copy of what it said.
Where it's not that simple
Deep sleep storage and workspace-recovery archives
When a VM goes into deep sleep, its disk is stored in object storage so your work survives. When a VM breaks because of a problem on our side, we save its workspace for 7 days so you can download it through a presigned, owner-only link. We don't read either. When you delete a VM, all of this is deleted with it and cannot be recovered.
Network rules every VM gets
Outbound mail on port 25 is blocked, so no VM can send spam directly. Ports 465 and 587 stay open: send through your mail provider there. Known crypto-mining pools, botnet command servers and the Tor network are blocked too, from public lists we refresh daily. A VM that keeps trying to mine is put to sleep, and we email you why. These rules block traffic in the kernel and only count blocked attempts. Your VM's DNS lookups go through a resolver on its host that refuses mining-pool domains; it counts refused lookups and never logs the names you look up.
7-day outbound connection log
Every VM on a host shares that host's public IP, so when a provider reports an attack from it, the only way to find the VM responsible is to know who connected where. We keep the destination IP and port, the minute and the source port for 7 days, then delete them. Contents are never recorded.
30-day operational log retention
Wide events about what our systems did (job outcomes, timings, error rates) are kept for 30 days, domain fields only, then deleted.
That's the whole list. No hidden analytics, no session recording, no reading your code or your agent's conversation with you.