Compute Blade: Your rack-mountable ARM cluster

Update 57: The Finale (Part 2, the Orchestrator)
about 1 month ago – Thu, Aug 27, 2026 at 10:52:47 AM

Uptime Orchestrator

Over the past few months, the project has undergone significant refinements and has been deployed to a cloud server. It is now ready for a full-scale beta test.

But first, let me explain what it has become. The original plan was to use it to manage a Kubernetes cluster based on Compute Blade. Now, it’s essentially an MDM server for Raspberry Pi devices. 
Its unique feature is a device recovery mode that uses a trusted OS, which is booted over the network into the device’s RAM. In this mode, you have full control over your device and can completely reinstall or restore the OS on the primary storage. The main thing is that, for recovery mode, the primary storage doesn't need to have an operating system or even work at all. The primary storage device can be a USB drive, SD card, NVMe, SSD, or even an HDD. Perhaps in the future, we won't need a main storage system at all, but let's not get ahead of ourselves.

I’ll briefly describe how it works: The Orchestrator flashes your Raspberry Pi’s EEPROM with a signed image; then, when the Raspberry Pi boots up, it contacts the server to request the image - if the server provides it, the Raspberry Pi boots into recovery mode. If not, it boots into the local OS (Of course, things are much more complicated under the hood, but I won’t go into the details right now).

And that's just the tip of the iceberg

I’ll share a few things I want to highlight in this announcement, but these are by no means all of the orchestrator’s capabilities. What’s more, development is ongoing, and the backlog is full of new features. Stability and security are our top priorities. That said, I’ll be honest: sometimes things stop working due to the rapid pace of development and constant AI-powered audits designed to detect bugs and security vulnerabilities. But that’s what the beta is for.

Let's start by looking at how to add a node to the system.
There are several options available; the simplest is to add an existing device with a single command. Alternatively, you can create a master image to deploy on an SD card, and all new devices will appear in your admin panel, awaiting approval.
You can also simply flash the Raspberry Pi (in Bootloader mode and connected via USB) directly from your browser.

A device that’s already been set up with the agent offers a wide range of metrics and possible actions.

Here, I’d like to highlight just a few things:

The ability to specify the platform or HATs being used (in the case of the Raspberry Pi 4/5).
Right now, the focus is on the Compute Blade, but more is to come.

There are also more technical tasks that can be done with literally a single click. For example, overclocking the processor or enabling Watchdog, which will reboot your device if the conditions you specify are met.

Oh, the interface shown above is just a cropped version.
Actually, we have a whole lot of stuff

When developing this product, I drew on my experience as a system administrator and my familiarity with various user interfaces. However, I don't think this is yet the final form of Orchestrator in terms of the UI.
We have classic OUs here, with the ability to assign policies to them. Overall, assigning policies, permissions, credentials, and utilities based on OUs is a great practice. It makes it easy to add new nodes or replace existing ones.

You can use traditional passwords, SSH keys, and even keys issued by a CA

Store secrets, variable environments, and node attributes. Attributes are needed to make a node unique and to store that uniqueness on the server. This way, you can replace one device with another and transfer the attributes. This might seem like a narrow use case for now, but I hope it will become a little clearer later on why this is necessary

The Operations tab is a whole world of setting up blank devices and backups. And what’s even more interesting is creating golden images. Why? In the world of CI/CD, you sometimes need to completely reinstall a device - and ideally, you want it to have the latest OS.
The orchestrator can take a snapshot of a node on a schedule - creating a hardware-independent image from it - and then deploy it to specified devices. Even to the entire fleet. And the attributes mentioned earlier allow you to automatically customize the nodes in your farm after deployment.

All operations are based on playbooks. I've added a few templates and plan to upload more to GitHub in the future.

You can run playbooks on either a single device or an entire group of devices.

And here’s another feature I wanted to share here (though it’s far from the last on the list)
the utilities.
With Playbook, you can flash an image or even install software. Using the “Utilities” tab, you can more flexibly manage the software or settings on the devices themselves. 

Here’s a specific example: 
kiosk mode with a specific link in the browser. If you want to display a web page on several of your devices, you can do it in just a few clicks.

The displayed link can be specified explicitly or can be an attribute; in this way, it can be unique not only for the OU but also for each node.

Another useful tool is Tailscale; with a single command (you'll need to add your key), you can add all your devices to your Tailscale dashboard—all you have to do is approve them.

And, of course, there are logs and device status monitoring. Right from here, you can update the agent or even the bootloader across your entire fleet.

You’re probably looking at all this, and I’m sure some of you are wondering, “What’s the point?” Especially if you only have 1 or 2 devices in your setup. 
Well, from my perspective, the recovery functionality alone will be useful to everyone.
Whether it’s a Raspberry Pi serving as a smart home server or a garage door controller. If its storage fails, you’ll still be able to connect to it and perform basic debugging via SSH, and I think that covers half the cases (if not more) - where you’d otherwise be forced to physically access the device, connect a monitor, and so on. And when it comes to a remote device, this is a real lifesaver.

All right, I hope I’ve piqued your interest a bit. Let’s move on to the numbers and details. 

1. This is usually the last thing people cover, but here I’ll start by talking about monetization. I haven’t fully figured out what this will look like yet. Most likely, it’ll involve paid plans for 10 or more devices. Or maybe there will be limits on image storage on the server (which is already in place for beta users). And of course, premium plans for enterprises with custom pricing.

2. Open source? The project turned out to be incredibly complex, seriously. There’s a massive backend; images need to be signed, and devices need to know where to go. In short, I don’t plan on making it open source for now. However, the agent itself - the one that gets installed on devices - will be open source in the future.

3. How do I join the beta test?

The most important thing is that the beta version is the best way to improve the product. Please join us at this stage if, after reading all of the above, you're interested and would like to try it out on at least one device. Well, just be prepared for the possibility that some things might not work at all—because development is moving so fast, not everything is covered by tests, and some things (such as adding new devices) are extremely difficult to cover with tests right now, so we have to test them manually. Anyway, I hope everything will be okay.

So, if you're okay with that, go to https://app.orchestrator.sh and sign up with the beta invite code: beta-d0e09b99 (works for both email sign-up and "Continue with Google" - enter the code in the "Beta invite code" field). The number of invitations is currently limited. If it doesn't work, please try again in a few days. I'll gradually increase the limit

To connect a device: Nodes → Add Node → "One command" tab, generate the command and run it on the device as a user with sudo. It installs the agent and registers the device — it then appears in your account pending approval. The device only needs outbound internet; no port forwarding or VPN.
Approve the node, and you're set. Heads-up: to get SSH access to the device through the platform, add a user with your SSH key under Credentials first.
The one-command install works on Ubuntu/Debian-based ARM devices (Raspberry Pi 4/5, CM4/CM5, and similar). There are also USB-flash and SD-card flows for network-first provisioning — happy to walk you through those if you want them.
Even though I try to make the product intuitive, documentation is still necessary, and it will be added. Someday. But not today.

Feedback is the whole point of the beta:
-  Bugs and feature requests: GitHub link
-  Questions and chat: Discord 


Leaving the beta / cleaning a device:

1. In the UI, delete the node first (Nodes → select the device → Delete). If it's still online, the platform also tells the agent to drop its adoption identity.
2. On the device, run:
curl -fsSL https://app.orchestrator.sh/uninstall.sh | sudo sh
This disarms the watchdog first (so nothing reboots mid-uninstall), then stops the agent and removes the package, its identity, and all its state.
3. SSH users and keys the platform configured are left in place by default, so you don't lose your way in. To remove those, too:
curl -fsSL https://app.orchestrator.sh/uninstall.sh | sudo sh -s -- --purge-access
It will refuse if that would leave no working login on the device.
4. When you're done with the beta entirely: Account → Request account deletion in the UI, and we'll purge the account and its data on our side.
One caveat: if your device was provisioned network-first (EEPROM flashed via USB or the SD card flow), the uninstaller doesn't touch the bootloader — reflash a stock EEPROM with rpi-eeprom-update afterward if you want the boot order fully back to default.

_________

Maybe for the last time here. With the deepest respect, and I tip my hat to you,
Ivan

Update 57: The Finale (Part 1)
about 1 month ago – Thu, Aug 27, 2026 at 10:40:28 AM

Hello!

I think we’ve reached the final chapter of the Compute Blade story on Kickstarter.
I’ve decided to break this down into two updates, one after the other. The first part (below) deals specifically with shipping the current devices and fulfilling (in one way or another) our obligations to you.

The second part covers the current stage of the Orchestrator and beta access.

So. With the support of our US partner, PiShop.us, we have completed another production batch of ComputeBlades. We now have enough inventory to fulfill all remaining Kickstarter pledges and outstanding campaign orders.

This is a major milestone for the project, and we are very grateful for your patience while we worked through an unusually difficult and unpredictable production environment.
There are, however, a couple of important matters we need to explain before the remaining orders are shipped.

If you are one of the backers in the US or Canada who has not yet received your ComputeBlade, please watch for a joint email from us and PiShop.us, sent from [email protected]. You should receive it on August 27–31, and your ComputeBlades should be shipped a day or two later.

However, it is very important that you receive, review, and reply to that email, as it will contain information we need to complete your fulfillment.
With sincere appreciation for your support and enormous patience!
 

For backers in Europe, the UK, Switzerland, Australia, and other regions, fulfillment is being handled by our European partner pi3g. The remaining orders are being processed, and we expect to complete the majority of them over the next few weeks. Some shipments may take slightly longer due to export requirements or outstanding components. pi3g will contact affected backers directly as their orders move toward shipment.

If you have any issues with your Kickstarter devices, please email [email protected]. If you’ve already sent an email and haven’t received a response from me, please send it again, and I’ll respond personally or contact the distributor to resolve the issue.
 

What's next for the project?

The chip crisis (this time involving memory, though not exclusively) has hit everyone hard, with renewed force. Do you still remember the last time?
 
One thing is certain: prices will rise. Right now, we’re looking for ways to lower the production cost of the Compute Blade and are actively discussing a new batch of devices that will be available for general sale in the future.
With the price of the Raspberry Pi rising, the situation doesn’t look promising there either. Nevertheless, there is demand in the market; it has likely shifted quite significantly toward large clients and businesses. And I very much hope that, thanks to these clients, we’ll be able to sustain our current operations and offer solutions to a broad audience at reasonable prices.


Thank you for putting up with me all this time.
With faith in a bright future,
Ivan

Update 56
4 months ago – Sat, Jun 20, 2026 at 06:37:12 AM

Hello,

It's hard to believe after everything that's happened, but we've received most of the devices from that complicated shipment.

• 2,536 Basic Blades
• 471 TPM Blades
• 1,000 Standard Fan Units
• 998 Smart Fan Units
   

Because of the missing parts, there were many minor replacements, and testing took a long time. But it looks like everything is fine.

Next week, we’ll ship part of the batch to the U.S. and keep the rest in Europe. We’ll continue to ship orders. However, at this time, there are orders that do not include DEV blades (which we expect to receive from the manufacturer by the end of the month).

I remember that I promised the Orchestrator a long time ago. An early version of it is running in a test environment, but a lot of minor fixes have piled up that need to be implemented before release, and unfortunately, that hasn’t been done yet.

There are also other accessories that require attention (the AI module and the HSM module). I’m in touch with the manufacturer (another one) that can produce them, but I haven’t received the final figures yet.

I'm sorry I haven't been writing as often as I should have,
Ivan

Update 55. A new (final?) wave of shipments is coming soon
4 months ago – Wed, May 27, 2026 at 02:32:54 PM

Hi, over the past few weeks, my correspondence with the production team has consisted of exchanging emails every 1–3 days. We’ve already exchanged over 85 emails on this topic.

First, I’ll describe the current situation. For once, there’s been a positive development.
So here’s the update:
- Fan units 1000 Basic + 998 Smart – shipped and received
- 1,215 Basic blades are in logistics on manufacture; 1,348 Basic blades and 468 TPM units are currently at the factory awaiting packaging.

- 1,994 pcs of DEV Blades are blocked by 600 pcs of LinkPP (which were “lost” by the factory earlier and reordered). Regarding the timeline for these, I was told “shipment from HK 05-27 + shipment ~1 week + production ~2 weeks.”

Now, here are a few highlights from this crazy story:

Ordering the antistatic bags 76x305mm (to replace the lost ones) took much longer than expected, so I asked them to change the bags to 254x305mm from stock. There was also the option of 400x400mm ESD bubble bags, which we almost ended up choosing, but fortunately, we managed to avoid them.

The second “amusing” story involved serial numbers. I quote: "Our operator got mixed up, and engraved DEV Blade SN from 0001615 to 0002301"

So the numbers ended up shifted like this (we couldn't just throw away the circuit boards):
Plan for DEV blades was: start from SN0000013
But we have 686 pcs. : SN0001615 to SN0002301  
So remaining 1312 pcs. will be: SN0000303 - SN0001614 
 

Unfortunately, it took days of back-and-forth correspondence to clarify the details before we reached these conclusions. Then there was a pause of nearly two weeks during which I received no replies at all; it turned out that the connectors hadn’t arrived. I’m now waiting for a response with tracking information on when the devices will be shipped from the logistics warehouse. 
Once we receive the blades at the Pi3G warehouse, we’ll ship them to the US; last time, there were no issues with this.
At the moment, I’m trying to process incoming messages regarding changes to shipping addresses and problematic devices (fortunately, there aren’t many of those).

P.S. I apologize for the delay with this update; once again, I really wanted to share some great news. But it turned out to be just okay.

Sincerely,
Ivan

Update 54
6 months ago – Fri, Apr 17, 2026 at 02:22:50 PM

Hello!

Overall, I just wanted to let you know about another delay. However, we do have the final figures as of now.
Fan units (1000 STA + 1000 Smart) are expected to be finalized (packaging) on 04.27. The assembly was completed in early April. 
They also started production with Blades basic 2536 pcs – now waiting for packing material to arrive at the end of April.

Blades DEV 1994pcs and TPM 468pcs will start production in the coming week, still waiting for the new Link-PP Ethernet connectors (originals lost by the factory)

The original plan was to manufacture the TPM and DEV units and ship them in early April. But due to delivery delays, the schedule has shifted, and it looks like I’ll be receiving the entire batch together

As soon as they receive the packaging materials (which were also lost by the factory), they’ll send them to Pi3G, and things will get moving. Unfortunately, they hadn’t received the materials due to “logistical issues”.

As I said, they “lost” the packaging as well, so we ordered new boxes based on the old designs. The box will be gray - don’t worry - the sticker on the box will have the latest information. We won’t confuse it with the earlier batch, which had power supply issues.

   

I understand that you are tired of updates (regardless of whether you received the Blades or not). With every update, I start to believe more and more in some kind of evil fate. It’s just unbelievable that so many problems keep piling up one after another, honestly.

But don’t want to end the update on such a note. I believe that the project will see it through to the end and go down in history as one of the longest-running SUCCESSFUL projects. 

Best Regards,
Ivan