Security
The rule behind keel.run's design: no customer can see, reach or slow down another customer. This page explains how, and where your data lives.
Where your data is
- Servers: Hetzner data centers in Germany and Finland. No US hyperscaler is in the data path.
- Backups: Hetzner Object Storage in the EU.
- Payments: Stripe processes card payments; keel.run never sees your card number.
- E-mail: sign-in links and billing notices are sent through a transactional e-mail provider, named in the data processing agreement.
The data processing agreement lists every subprocessor.
Tenant isolation
- Sandboxed containers. Every app runs in a gVisor sandbox, a user-space kernel between your code and the host. Pods run without host networking, host paths or privileges, and without access to the cluster API.
- One network per environment. Each environment is its own Kubernetes namespace with network policies: services accept traffic only from their own environment and the platform's ingress, and they can reach their own environment and the internet – not other customers and not internal platform services. Outgoing mail (port 25) is blocked.
- Resource limits. Each environment has quotas for CPU, memory, storage and pods derived from your plan. RAM is reserved, never overcommitted, so a neighbor cannot push your app out of memory.
- Isolated builds. Each build runs in its own short-lived job with rootless BuildKit in a user namespace. The repository token exists only inside the clone step and is deleted afterwards. Builds can reach the internet, but not private networks.
- Private registry per workspace. Images are pushed with a credential that is valid for one build and one service, and pulled with a credential of your workspace only. Images are deployed by digest.
- Separate domains. Customer apps run under
keelrun.app, which is on the Public Suffix List, never underkeel.run. Apps cannot set cookies for each other or for keel.run. - Databases. Each database is its own Postgres cluster with its own credentials and no superuser. It accepts connections only from its own environment and from keel.run's database proxy, which requires TLS.
Secrets and encryption
- Variables and database passwords are encrypted with a key per workspace (envelope encryption with a platform master key) and bound to the row they belong to, so a value cannot be moved to another place in the database.
- Secret values are hidden in listings. Revealing them needs the
config:writescope and is recorded in the audit log, as is every read of a database password. - Tokens, OAuth codes and refresh tokens are stored only as hashes.
- In transit: TLS 1.3 for keel.run's own endpoints, HTTPS with automatic certificates for every app, TLS for every external database connection.
- At rest: databases live on local NVMe volumes on dedicated servers; backups are written to object storage.
Accounts, tokens and agents
- Sign-in with GitHub, GitLab, Codeberg or a one-time e-mail link; no passwords to steal.
- Sessions are HttpOnly cookies; the API checks the origin of every state-changing browser request.
- API tokens belong to one workspace, can be limited to projects and scopes, and never have more rights than the person who created them. Agent tokens always expire.
- Destructive actions by agents need a second call with a short-lived confirmation token.
- Every change is recorded in the workspace's audit log with the person, token and client that made it.
Details: Agents and MCP and API authentication.
Backups
Databases are backed up daily, and the write-ahead log is archived continuously, so you can restore to any point in your plan's recovery window. Restores always create a new database.
Reporting a vulnerability
Please report security issues and abuse of the platform to abuse@keel.run, with enough detail to reproduce the issue. Do not access other customers' data while testing.