Update a hundred appliances without visiting one.
A release goes to a handful of machines first and only widens when the ones before it stayed healthy. If the failure rate crosses the line you set, Reamstead stops the rollout by itself and puts the appliances that took it back on the version they were running.
7 rollout stages · 9 command types · rollback to the previous slot

Six rungs from the platform to the tray
Everything in the fleet hangs off one chain, and every list you open is filtered by where you sit on it. A dealer sees their own clients. A business owner sees their own locations. An operator sees the counter they stand at.
The reseller rung is optional: a business that buys direct hangs straight off the platform. Everything below it is the same either way, which is why a dealer and a direct customer can be supported with the same console and the same commands.
Platform to Printer
01Platform
Us, or whoever runs this deployment.
02Reseller
A dealer with their own client list. Optional.
03Tenant
The business that owns the print room.
04Location
One shop, lobby, library or print centre.
05Appliance
The box on the shelf at that location.
06Printer
Each machine the appliance drives.

How a release reaches the fleet, and how it retreats
A download is not an update. Nothing counts as installed until the new agent on that appliance has answered its own health check, and nothing widens to more machines until the batch before it has sat quietly for a while.
- 01
Sign the release
Artifacts are signed with a key listed in a root-signed keyring, and each one carries its own checksum and size. A version that is not signed cannot be deployed, because the appliance would refuse it anyway.
Ed25519 over a canonical manifest
- 02
Choose who gets it
A deployment binds the release to everything, to one dealer, to one business, to one location, or to a list of named appliances.
- 03
Walk the stages
Internal machines first, then a widening share of the fleet. Membership of a stage is worked out from the deployment and appliance identifiers rather than drawn at random, so each stage contains the one before it and no appliance is updated twice or quietly skipped.
internal, 1, 5, 10, 25, 50, 100 percent
- 04
Let it soak
Each stage carries a soak window, a minimum number of successes and a maximum failure rate. The evaluator advances only when nothing is still in flight, enough installs have succeeded and the window has elapsed.
Evaluated every minute
- 05
Pause without being asked
When the failure rate for a stage crosses its threshold the deployment pauses itself, records why, raises an alert and notifies. It does not keep rolling into a fleet that is failing.
AUTO_PAUSED
- 06
Roll back to known good
The agent keeps two slots and installs into the one it is not running from. If the new agent cannot report healthy within two minutes it puts the previous slot back and says so. A rollback across a deployment offers the previous known-good version to every appliance that took the update, and is refused rather than guessed at when no such version exists.
120 s health check
- 07
Stop trying, and say so
Three failed boots and the appliance stops applying updates and runs a recovery agent that updates never touch. It keeps its identity, its printer configuration and its job database, including money taken while the line was down, and it keeps reporting so the console shows a machine needing attention rather than one that went dark.
Safe mode
7stages
Internal, then 1, 5, 10, 25, 50 and 100 percent
That ladder is the default and every stage on it is yours to change. Because stage membership is computed rather than sampled, the same appliance lands in the same position every time the rollout is evaluated, which is what makes a paused deployment safe to resume.

What you can do without a van
Three sets of controls, all of them narrow on purpose. There is no shell, no arbitrary script and no way to become one of your customer's users.
Remote commands
- Restart the agent
- RESTART_AGENTEvery command needs a reason and an expiry, is delivered at most once, and is written to the audit log on send and again on result.
- Restart printing
- RESTART_PRINTING
- Re-read a printer
- REFRESH_PRINTER
- Run diagnostics
- RUN_DIAGNOSTICS
- Check for updates
- CHECK_UPDATES
- Install an approved update
- INSTALL_APPROVED_UPDATE
- Restart the appliance
- RESTART_APPLIANCE
- Collect a support bundle
- GENERATE_SUPPORT_BUNDLE
- Clear the local job cache
- PURGE_JOB_CACHE
Alerts that fire
- Appliance stopped reporting
- APPLIANCE_OFFLINE
- Appliance fell back to recovery
- APPLIANCE_SAFE_MODE
- Appliance low on disk
- APPLIANCE_DISK_LOW
- New appliance waiting for approval
- APPLIANCE_AWAITING_APPROVAL
- Printer not answering
- PRINTER_OFFLINE
- Paper jam
- PRINTER_PAPER_JAM
- A job could not be handed over
- JOB_DISPATCH_FAILED
- A rollout paused itself
- OTA_AUTO_PAUSED
- Device certificate expiring
- DEVICE_CERTIFICATE_EXPIRING
Support sessions
- Longest session
- 240 minutesOpened against one business, with a reason typed in at the start.
- Impersonation
- NoneThe supporter acts as themselves with a temporary grant. There is no becoming another user.
- What the operator sees
- A bannerThe console at the counter shows the session while it is open, and the business owners are notified when it starts.
- What is recorded
- Every actionEach audit row written during the session carries the session id, so the whole visit reads back as one block.
Commands, alerts and sessions are the same on an appliance you sold and an appliance you own, which is what makes a hundred of them supportable by one person.
Watch a rollout before you place the first box
Tell us how many locations you run or sell into and we will walk you through the fleet view, a staged update and what a support session looks like from both ends.
fleet · stages internal/1/5/10/25/50/100 · rollback to previous slot