The facts behind SwengleIO's security and reliability: where data is kept, how it is encrypted, who can reach it, how it is backed up and what happens when something fails. Where something is not in place, we say so.
Documents and contacts
Sub-processors
Every provider that processes customer data for us, what it does and where.
SwengleIO runs in Swengle's own accounts with the cloud providers below; nothing runs in your cloud account. You pick an endpoint's region when you create it, and it stays there for its whole life.
Region
Cloud
Messages and backups
Encrypted off-site backup copies
aws-eu-central-1 (Frankfurt, Germany)
AWS
In the region
European Union, guaranteed
aws-eu-west-2 (London, UK)
AWS
In the region
European Union, guaranteed (in the EU, not the UK)
aws-sa-east-1 (São Paulo, Brazil)
AWS
In the region
Eastern North America, a preference, not guaranteed
aws-us-east-2 (Ohio, USA)
AWS
In the region
Eastern North America, a preference, not guaranteed
gcp-asia-northeast1 (Tokyo, Japan)
Google Cloud
In the region
Asia-Pacific, a preference, not guaranteed
gcp-europe-west2 (London, UK)
Google Cloud
In the region
European Union, guaranteed (in the EU, not the UK)
gcp-europe-west3 (Frankfurt, Germany)
Google Cloud
In the region
European Union, guaranteed
gcp-us-east1 (South Carolina, USA)
Google Cloud
In the region
Eastern North America, a preference, not guaranteed
oci-eu-frankfurt-1 (Frankfurt, Germany)
Oracle Cloud Infrastructure
In the region
European Union, guaranteed
oci-uk-london-1 (London, UK)
Oracle Cloud Infrastructure
In the region
European Union, guaranteed (in the EU, not the UK)
oci-us-ashburn-1 (Northern Virginia, USA)
Oracle Cloud Infrastructure
In the region
Eastern North America, a preference, not guaranteed
Messages and backups stay in the region. An endpoint's messages are stored in its region's database, and that database's backups in the same region, with the same cloud provider.
Off-site copies. An encrypted copy of the backups of every region we offer is also kept in Cloudflare R2, so they survive the loss of the whole region or of its cloud provider. For regions in the UK and the EU, the copies are kept under Cloudflare's EU jurisdiction, which keeps them in the European Union (for London, that is the EU, not the UK). For other regions Cloudflare is given a location hint, which it treats as a preference: copies for Tokyo, for example, can be kept outside Japan.
Account and Console data (accounts, organizations, members, endpoints' settings, key IDs, usage counts, the audit log and Assistant chats) is kept in Google Cloud's us-east1 region (South Carolina, USA), with its backups in Google Cloud's nam4 dual region (Iowa and South Carolina). It does not store your endpoints' messages.
Broadcasts and presence are never stored. There is no replication or failover between regions: to serve from several regions, create an endpoint in each.
We don't offer a residency guarantee beyond what is stated here and in the Privacy Policy. If you need one, for example copies kept only in Brazil or only in the UK, talk to us before relying on it.
Encryption
In transit
Endpoints and the Console are served over TLS only (https:// and wss://), and the client libraries always use it.
Inside a region, our servers reach the databases, relay hosts and each other over TLS, verified against a certificate authority of that region's own.
At rest
What
How it is encrypted
Disks on Google Cloud, Oracle Cloud Infrastructure and Microsoft Azure
Encrypted by the cloud provider, always, with keys the provider manages.
Disks on AWS
Every disk we create asks for encryption with the region's AWS-managed key, and each region also encrypts new disks by default. This was put in place in November 2026: disks created before it were not encrypted, and AWS cannot encrypt a disk in place, so they are encrypted by replacing their machines.
Backups
Encrypted before they are written to storage (libsodium), with a key per database that a machine is given only for the database it backs up or checks. The cloud's storage also encrypts them at rest.
Off-site backup copies (Cloudflare R2)
Copies of the already-encrypted backups, so Cloudflare holds only ciphertext and never the key. Each region's key is kept with the copies only sealed to a recovery key that Swengle holds offline.
Secret keys
A secret key is shown once, when it is created; nobody can show it again, Swengle included.
Not offered
Customer-managed or customer-supplied encryption keys.
End-to-end encryption of message payloads. Encrypt payloads in your application before publishing if Swengle must not be able to read them.
Encryption of database contents beyond the disk's own encryption above.
Access control
Your organization
Every member has Read; Write, Billing and User admin are granted separately. Removing a member ends their access at once.
The organization's audit log shows who changed what, from which address and when, kept a year.
Sign in with Google, with an email and password, or with your organization's single sign-on (SAML 2.0 or OpenID Connect). Signing in with Google uses your Google account's protection, such as 2-Step Verification.
Two-factor sign-in with an authenticator app's code, set up on the Console's Account page. An organization can require it of every member.
Secret keys stay on your servers; browsers and apps get short-lived client tokens that your server mints.
Swengle staff
Operators run the platform. Their tools call one administrative API, which accepts only Google-signed identity tokens of the accounts on its list; there is no static admin password or token. Operators' changes to regions, clusters and hosts are written to an audit log with their identity.
Admins are Swengle staff named in the Console's configuration. In the Console they see Swengle's own organization, with its shared clusters and test endpoints, and a fleet overview. Being an admin does not make anyone a member of a customer's organization.
Production access. Deploys of our servers run from our repository's automation, only when started by a person, with short-lived cloud credentials; the stored exceptions are an Oracle Cloud token-exchange secret and a Cloudflare DNS token. This website publishes automatically, and only what is already deployed. Our machines accept no SSH from the internet. When a person needs a shell on one, it goes through the cloud provider's own access service (Session Manager on AWS, Bastion or Run Command on Oracle Cloud Infrastructure, Run Command on Azure, Identity-Aware Proxy on Google Cloud), under that cloud account's access control.
Access to customer data within Swengle is limited to the staff who operate the Services, for operating, supporting and securing them, for example restoring an endpoint at your request. Staff may read Assistant chats to follow up on a support request or investigate abuse; other members of your organization cannot.
Nobody at Swengle will ask for your password, a secret key or a client token.
Backups and recovery
Dedicated and Cluster have an availability level that you choose; Shared always has the highest. What each can lose (recovery point) and how long it takes to serve again (recovery time) are by design, not commitments:
Level
If the database host is lost, it loses
Serving again
No backups Cowboy mode the default for a new Dedicated or Cluster
Everything: the endpoint comes back empty
Minutes, on a new host
Daily backups Safety net recommended
Everything since the newest full backup, up to about a day
Minutes, restored onto a new host
Continuous backups Flight recorder
At most about the last minute
Minutes, restored onto a new host
Replicated database Co-pilot
Nothing acknowledged, while the replica is in sync
What is kept. Daily and continuous backups take a full backup every day and keep the newest 7; continuous backups also archive every write, at most about a minute behind. Shared endpoints are always backed up continuously, besides their replica in another zone.
Restores are checked every week. Each database a region backs up is restored on a separate machine and compared with its source, checking that rows and checksums match what was marked before the restore. A check that fails is retried after an hour, then at growing intervals until it passes, and alerts our operators.
Point-in-time restore. With backups, you can restore an endpoint's channels and messages as of an earlier time into a new endpoint, from the Console. The original is not changed.
Deleting. Deleting a Dedicated or Cluster, or turning its backups off, deletes its backups. Deleted backups are kept 30 days, then gone for good; their off-site copies are deleted 14 days after the backup, and never sooner than 30 days after they were made. A Shared endpoint's messages already in its shared cluster's backups stay until those backups age out.
Losing a region. There is no automatic failover to another region. If a region or its cloud provider were lost, Swengle staff could restore its databases from the off-site copies, by hand; we don't state a time for that.
The Console. Its database has a synchronous standby in another zone and is backed up continuously, so it can be rebuilt in another region, losing about the last minute of changes. Running endpoints don't depend on the Console: they keep serving while it is down.
Availability and failover
The five availability levels decide what a failure does. The numbers below are what we have measured in our tests, not commitments; design clients to ride out a pause, which the libraries do on their own (they retry publishes with the same idempotency key and resume subscriptions from their cursors).
Database failover
A replicated database host that stops reporting for 15 seconds is failed over to its synchronous replica, losing nothing acknowledged. On a Dedicated cluster at Zone outage protection, publishes and deliveries pause for about 15 to 25 seconds when its host alone is lost. On a Dedicated cluster at Replicated database, the standby host takes over and clients reconnect to it. A zone outage that also takes Swengle's regional control database in that zone, which fails over first, can pause publishes and deliveries for up to about 80 seconds, and a Dedicated cluster at Replicated database, whose clients reconnect, for up to about two minutes: in zone outage tests on our staging regions the longest pause in publishes was about 76 seconds at Zone outage protection and about 95 seconds at Replicated database. On a Cluster and on Shared, relay hosts stay up and clients stay connected while the database switches.
Without a replica
A lost host is replaced automatically and, with backups, restored: minutes of downtime, during which the endpoint shows Degraded.
A zone outage
With Zone outage protection, on every Cluster that has it, on Dedicated where offered and on every Shared endpoint, the other zone keeps serving.
Planned changes
Updates, operating system patching (on our regular schedule, during our maintenance windows) and new host sizes lose nothing at any level. Clients of an updated relay reconnect at once and resume; when a Cluster's relay host is replaced, deliveries pause for well under a second. Moving an endpoint up a plan pauses its publishes for about a second while it switches.
The Console
Changes in the Console pause for a few seconds (about 5) when its database switches to its standby, for example for patching. Running endpoints are not affected.
We don't publish or promise an uptime figure. A committed service level needs a separate agreement.
Incidents and status
Monitoring. Alerts about every region reach our operators by email, and the most serious ones page the operator on call. Each region is also checked end to end, publishing and receiving messages.
The status page shows the Console and each region as we see them now. Some problems show there on their own, in fixed words; incidents are posted by our staff, with their updates as they happen, through Investigating, Identified, Monitoring and Resolved. Past incidents stay for 90 days, and an Atom feed carries every update.
How we run one. One person leads each incident, and severity is set by what customers feel. For an outage or a partial outage, we aim to post on the status page within 15 minutes and to update at least every 30 to 60 minutes. For a large incident in a region, we may also email every member of the organizations with endpoints there.
After. We write a blameless review of every serious incident. After a major one, we send affected customers a summary of what happened and what changes, and to anyone who asks.
Security incidents are handled at the highest severity. If one affects your Customer Content, we notify you as our Terms describe.
Vulnerability management
Our automated checks scan our Go code, our JavaScript dependencies and the binaries we build for known vulnerabilities. One that affects us and has a fix we can take fails the check until it is fixed, and every release is scanned before it ships.
We patch the operating systems of our machines ourselves, on a regular schedule, during our maintenance windows, one machine of a cluster at a time. Our machines never update themselves, so updates never restart a machine outside a window.
No certifications or audit reports. SwengleIO does not hold SOC 2, ISO 27001 or any other certification or attestation, and is not offered for HIPAA or PCI-regulated data. The cloud providers we run on publish their own.
Card details go to Stripe on its own pages and never reach us.
Data processing agreement. We don't publish one yet. Write to privacy@swengle.com if you need one.
Our commitments are the ones in our Terms of Service and Privacy Policy. This page describes how SwengleIO works today and is not itself a contract.