UK sovereign LLM inference
relax.ai107 points by benjamintnorris 2 days ago
107 points by benjamintnorris 2 days ago
Could we have a bit more "who and where" on this please? "Relax.ai by Civo", great, who are Civo? Where's the datacenter? What's the corporate structure? UK resident founders?
relaxAI is a spin-out from Civo — UK-incorporated cloud provider, been around since 2018, founded by Mark Boost. Probably best known for managed Kubernetes (fmajid’s got it right downthread).
We spun out in 2026 to focus on sovereign inference and UK workloads that need in-country processing. UK-resident founders. Not a UK subsidiary of a foreign hyperscaler or model provider — wholly UK-owned and operated, running in Civo’s LON1 datacentre.
ICO-registered, NCSC-aligned, on G-Cloud via CCS. UK law and UK courts only, no cross-border data flow at any layer. The majority of our customers are UK orgs where in-country processing is a hard requirement such as UK healthcare, legal and education companies.
This should be on your "About" page, ideally coupled with direct link(s) to Companies House.
You need to be flaunting your sovereignity, not merely wielding it.
Civo is a UK cloud provider known mostly for its low-cost Kubernetes hosting service (albeit with fairly expensive storage).
what's preventing you from going to their website? op has linked directly to the documentation, so that info is not necessary expected to be there
OP should have linked to the website rather than the documentation; that would have provided more immediate context for the discussion.
> what's preventing you from going to their website?
Lack of links.
It's a common annoyance when subsections of a site fail to link back to the parent.
I did actually try looking for an "about" on the linked page and googling, neither of which was immediately helpful.
5 minutes to load and it just dumps me to a documentation site with no useful information about that this is, who made it, what it can do, etc.
Apologies - we had a lot of sudden traffic from Hackernews! We’re looking at it right now. Thanks for flagging.
Congrats, its a small step in the right direction.
The UK it seems has dropped the ball on the whole training and building models part, although we are punching up in other areas now.
We really need to get our own equivalent to Mistral, and fast!
Well, it surely did not help that the government has been drip-feeding us computational resources. First we had about 16 GPU nodes to share between the whole country for over a decade. Then just before Isambard-AI came online they made one open call for about a week where you could get nearly enough to train a sizeable model, but the call was poorly advertised and in the middle of high vacation season. After this, the only big call explicitly cut out training large language models by its scope and the general calls have been peanuts and less than me and the UK-LLM team had access to during the Isambard-AI beta phase! When I gave a talk at White Hall recently my message was clear: We have the team, the knowledge, the data, etc. We just need an open call for enough compute to train the darn thing! Here is to hoping that they listened.
Completely agree - we’re really eager to work with any UK model lab that wants to make a difference!
Just my curiosity. Is (insert country) sovereign X is an efficient marketing strategy these days?
Yes.
Some people might interpret this comment as political commentary, but it’s actually just the reality of what people are saying and doing.
There’s a lot of data to suggest that America’s recent policy of reducing its soft power around the World & decoupling itself from alignment with interests of allies is causing increased interest and prioritisation of sovereign capability across tech, defence, public health and policy programs.
This was a campaign strategy/promise for the US President. I’m not going to comment on whether it’s good for the US or for the allies, but I will note it could have been better anticipated by all: the only real surprise is the speed and depth.
It raises some interesting questions - it’s one thing to say you don’t want Microsoft or Starlink in your infra tech stack, or don’t want to use AWS or GCP, but where does the line stop? Does the UK get out of Trident? Does the UN General Assembly get out of New York? No idea, but the fact these are conversations probably happening right now is remarkable.
>This was a campaign strategy/promise for the US President.
I don't remember seeing "if you elect me I'll destroy NATO, threaten allies, and make sure even USA's oldest allies hate us" as part of the campaign.
Perhaps you could link that promise from the time before the election?
As to your questions, I think people are hoping that rule of law returns and there is an outbreak of common sense. No-one really expected US president will align with Russia and a cult of pseudo-Christian white-supremacist nationalists will worship him as the second coming as an expectation. That the Senate, SC, and tech leaders have fallen in place behind that (the latter literally paying fealty to their god-king) is complete insanity.
> I don't remember seeing "if you elect me I'll destroy NATO, threaten allies, and make sure even USA's oldest allies hate us" as part of the campaign.
You weren’t paying attention, or spent to much time in wrong echo chamber.
> No-one really expected US president will align with Russia and a cult of pseudo-Christian white-supremacist nationalists will worship him as the second coming as an expectation. That the Senate, SC, and tech leaders have fallen in place behind that (the latter literally paying fealty to their god-king) is complete insanity.
I think it's completely expected that the Senate Republicans would do what they were told, and the multi-decade project to stack the Supreme Court with extremists to overthrow Roe v. Wade was something that was very obvious while it was happening.
> No-one really expected US president will align with Russia and a cult of pseudo-Christian white-supremacist nationalists will worship him as the second coming as an expectation. That the Senate, SC, and tech leaders have fallen in place behind that (the latter literally paying fealty to their god-king) is complete insanity.
I would like to remind you of Project 2025. While Trump claimed to know nothing about it prior to being elected, so far he's appointed every editor/contributor to the document/project to his administration and achieved more than half of the stated goals of the document.
> No-one really expected US president will align with Russia and a cult of pseudo-Christian white-supremacist nationalists will worship him as the second coming as an expectation.
I'm sorry but everyone who was paying attention expected exactly that.
>No-one really expected US president will align with Russia and a cult of pseudo-Christian white-supremacist nationalists will worship him as the second coming as an expectation.
QAnon happened during Trump's first term. The support for Trump by the alt-right and white supremacists were so well known they were written about in the mainstream media.
Trump's links with Russia have been well known for decades. He literally started his construction business doing deals with the Russian mob.
Hell, even Trump's sexual abuse issues were right there for everyone to see. "Grab'em by the pussy" would have ended anyone else's entire political career, it just made Trump more popular.
>That the Senate, SC, and tech leaders have fallen in place behind that (the latter literally paying fealty to their god-king) is complete insanity.
Why do you think Trump packed the court with candidates hand-selected by the Federalist Society?
Did people miss the creeping right-wing shift in the Republican Party after 9/11? The ascent of Evangelical Christianity and Christian nationalism? The normalization of authoritarianism and fascism within the Conservative movement has been happening since the days of the John Birch Society. The dark enlightenment shit in SV goes back to Curtis Yarvin in the early 2000s.
All of this was seen coming from miles away.
All of this was warned about.
Shouted about from the rooftops and on the streetcorners.
For years.
No one cared because it was women and gays and people of color, "leftists" and "liberals" doing the warning. They were dismissed as delusional. Americans only trust in the authority of white men, and half the country swept Trump into power specifically because they wanted a white supremacist Christian revolution purging "DEI" and "feminism" and "progressivism" and "secularism" from culture and government. Even Trump's own supporters warned us about their intent. They literally published it in a document, with bullet points and everything, and Americans trusted the white guy with his fingers crossed behind his back - despite being an infamous liar, con-man and holding views that align perfectly with Project 2025 - because the alternative would mean trusting that a "leftist" could be right about something, and having a black woman president. Can't have that. We had a half-black man in the White House who was at best center-right and Americans lost their goddamn minds about him being the Marxist Black Panther Antichrist.
Americans don't get to say they didn't see it coming. This was the most easily telegraphed arc of fascism ever in the history of civilization. The US has been playing it on easy mode and still lost.
Suspect it depends on the sentiment.
Don't think you'd have much luck convincing say a German that they shouldn't use Mistral because it isn't German sovereign. But you might have luck with that line against china or america.
Or put differently depends more on the fault lines in public perception than strict borders
Well put.
Which goes to show maybe it's less about literal sovereignty than wanting to rely on parties less prone to exploit the reliance for stabbing you in the back.
Considering it took decades of absolutely no meaningful action against the US tech sector's cavalcade of object lessons in 'They "trust me". Dumb fucks', and orgs were still eagerly ripping out homegrown system solutions to get all their communication and storage locked into Azure and AWS at the most four years ago, I'm not sure whether to feel relieved or exasperated that there are some signs of real reactions this time.
Yes. My first-hand experience is that there's huge demand for it in the UK at the moment, for data sovereignty in particular.
LLM inference is currently a secondary concern - lots of conversations, some pilots and PoC implementations but not much beyond that (yet). I'd expect that market to grow substantially over the next couple of years, though.
Note that "sovereignty" isn't seen as a binary thing in this context. Rather, organisations are using a risk planning approach: mapping out potential sovereignty concerns, then building strategies to mitigate them. Think of it as the start of a wave that will likely continue to build for the rest of the decade.
Big-bang onshoring operations aren't really on anyone's radar at this stage, but it's great to know that there are more UK-based options are becoming available.
It is if your country isn't in the US and (a) GDPR requires data residency in UK/EU; (b) you're concerned about capricious actions by the US govt cutting off access to US-controlled services (cloud, payments systems, etc).
I’d say it’s more than a marketing strategy, it’s reflecting real demand. Countries have legal requirements (or increasingly strong preferences) for the kind of guarantee that data/inference sovereignty gives.
I'm not sure if you are asking for hard numbers, but I would say it's definitely "a thing" for people to reduce their reliance on certain countries.
I mean, it's pretty rich for coloniser like the British empire to be talking about soverign anything
US tech is currently being weaponized against the ICC and its member judges in Europe[1], and the US is threatening to annex Greenland, as a result all (former) US allies are scrambling to get rid of their strategic dependency.
[1]: https://www.lemonde.fr/en/international/article/2025/11/19/n...
> as a result all (former) US allies are scrambling to get rid of their strategic dependency.
In a very limited way. Some countries are moving some government systems off US suppliers, but its very limited.
Somethings are going the other way. For example (there was a post about this on HN a few days ago) the new EU ID/age verification app depends on Google or Apple attested phones.
There is no real effort to push the private sector off American systems. Your government might technically function but its going to be crippled if your private sector has ground to a standstill.
It's still limited, but the shift has been dramatic over the past few month alone. It's hard to tell what the situation will look like in a few years. Unless the midterms are a democrats landslide (which the GOP has tried to prevent as much as possible with gerrymandering) I don't see the trend stopping anytime soon.
I can't see any information here about prompt caching (AKA prefix caching). That's absolutely required for any serious API usage with pay-per-token pricing. Anyone from Civo or relaxAI here able to clarify that? Thanks!
btw, you can claim "relax" name instead of "relaxai" on pypi
pypi.org/project/relax is abandoned library, which owner registered via email with expired custom domain, so you can claim this domain and reset owner's account by email.
It's probably a better idea to follow the process documented in PEP 541 [1] and contact the PyPI admins to request a transfer of the name. Taking over the domain to impersonate the original owner would look indistinguishable from a supply-chain attack.
[1] https://peps.python.org/pep-0541/#how-to-request-a-name-tran...
The obvious approach to me for a country seeking sovereign ai would be to hire as many of the recent qwen core team as possible.
If you're talking about UK sovereign LLM inference you need to mention Doubleword... very serious inference optimization lab in london with public endpoints for OS models
> Civo isn’t just another cloud and AI platform, it’s a whole new way of thinking.
come on now
Classic LinkedIn copy and paste line you see on someone with "Founder, CEO and Cereal Entrepreneur" in the job description.
Nice. All for seeing more geographically diverse options.
BTW don’t see opencode in the docs yet much less known tools are?
Hi! We support OpenCode, as well as any other coding tools accepting OpenAI compatible endpoints. For example, take a look at: https://relax.ai/docs/integrations/developer-productivity-to...
Why would smaller and worse models not be 80% cheaper?
If I can run those models on my consumer hardware, I'd better believe they are 80% cheaper than the models that need 1 TB of RAM.
For the very hardest reasoning tasks GPT-5 and Opus are still ahead, no argument there. But what we see in practice is customers dropping in an open-source model and getting very similar results on 80-90% of real-world use cases — with significant cost savings and end-to-end UK data residency (which matters a lot for our enterprise and institutional customers). And on the consumer-hardware point: these are Blackwell GPUs in a UK datacentre, in a token factory architecture.
Comment from poster says they are offering Deepseek v4-Pro. Cannot find any details on website.
click "Models and Pricing" in the left menu https://relax.ai/docs/getting-started/pricing
Its written
Input Price: £1.17 Output Price: £2.33
So, slightly cheaper than Fireworks AI
Personal take: terrible name. RelaxAI feels like you trawled for available .ai domains with dictionary words and landed on this. But it doesn't work, unless it's a relaxed AI. Is it slower, but cheaper, we'll process your requests when we get to them, so relax!
You could have bought languagemodels.co.uk off me and used that!
Personal take: terribly disguised pitch to get someone to buy that way too long domain name from you.
What about those in the UK worried about their data when using AI... Perhaps they could, I dunno, relax, knowing they're using a service in their local jurisdiction?
And more generally, let the AI do the thinking/coding/whatever... Just relax until it's ready.
I will say, I find it interesting that the world relax has such a negative connotation in your mind.
Anyhow, Is it the best name ever? No.
Is it a hell of a lot better than languagemodels.co.uk? 1000%.
> You could have bought languagemodels.co.uk off me and used that!
this is a joke, right?
This looks very interesting.
I have no idea why you got downvoted so much.
As The Crown is sovereign of the United Kingdom, is this running in Buckingham Palace or in City of London?
Can the user choose which sovereign is doing the computation?
I'd personally prefer not to have the weird uncle do the computation, maybe the younger ones living abroad can do it.
;)
Parliamentary sovereignty is a cornerstone of the UK constitution, fwiw. Parliament has been de facto sovereign since the late 1600s.
The server is under the woolsack!
Hi HN, I'm Ben, founding engineer at relaxAI.
We built a UK sovereign inference provider for developers who are either paying too much for OpenAI/Claude tokens or can't use US hyperscalers due to data residency requirements.
The short version: drop-in OpenAI-compatible API, latest open source models (Kimi K2.6, DeepSeek V4 Pro, Nemotron 3 Super, GPT OSS 120b), running on NVIDIA Blackwell GPUs in the UK. Zero code changes to switch from OpenAI. Up to 80% cheaper per token cost saving!
We built it on fully UK sovereign cloud infrastructure, so data never leaves UK jurisdiction. For anyone building in regulated sectors — finance, legal, health, defence — that matters a lot. But honestly, most of our early users just came for the huge cost savings.
We're looking for developers to kick the tyres. Check out our API docs at relax.ai/docs. I'd love your feedback and happy to answer any questions.
Firstly, congrats! As a Brit this looks cool, and I'm happy to see it. I wish you every success.
Secondly: I get that 'sovereign' is probably an important sales term for your company. But this, in common with the government's 'sov/ai' fund, does not deserve to be described as sovereign. This is other countries' models served on chips designed and manufactured abroad, powered by a grid which imports 44% of its power.
Of course this isn't your company's fault. Last week I went to an event where the sovereignai.gov.uk people presented. In a very Keir Starmer way (spiritually, he wasn't there), they said in as many words 'oh but I'm sure all reasonable people would agree _really_ sovereign AI would be too hard. So let's all agree to pretend that just popping a bit more money into the AI startup ecosystem is a sovereign AI strategy'.
I'm unsure if the UK does need to be sovereign in anything; it certainly doesn't seem to want to be. But I will continue to poke fun at anything using the pompous phrase 'sovereign' for anything that isn't.
If sovereign AI is a problem you're in earnest about, I hope you go after it seriously, and fix the rest of the stack. I'll cheer you on!
Sovereign capability just means that no foreign government can pull the plug. Who cares where it was copied from?
If it were somehow legal for a company to provide MS Office (not a clone) fully in the UK with no control from Microsoft, that would also count as a sovereign capability, even though none of the code was written in the uk.
Maybe that's not how you like the term to be used but it's widely used that way and widely understood.
> Sovereign capability just means that no foreign government can pull the plug
OK, fair enough on 'pull the plug ~instantly'. But models and chips age fast. If another country can stop you getting new models and chips, this means you're sovereign in state-of-the-art AI for only a window of a year or two (maybe this will widen if model progress tails off).
If it is a short window, strategically, that doesn't seem worth much given the timelines of: a) inter-state conflict, or trade wars b) cold-start time to be able to make your own models and chips
> Maybe that's not how you like the term to be used but it's widely used that way and widely understood.
Noted. But as a data-point, the audience at the event I mentioned (various AI builders and founders) made it clear from their questions to the speaker that the 'sovereign' that sov/ai was aiming at was hollow, for exactly the reasons I've stated.
> But models and chips age fast. If another country can stop you getting new models and chips, this means you're sovereign in state-of-the-art AI for only a window of a year or two
OK, that's a fair point.
The UK grid does not import 44% of its energy.
Gell-Mann suggests I should treat the rest of your post with skepticism.
EDIT: maybe you meant the UK total primary energy? I feel like that's extending the boundary a bit far. Should we start digging for uranium? Or stick to renewables, but only with locally sourced silica and rare earths?
Import dependency. The UK government put it at 43.5% in 2025 and 43.8% in 2024: https://assets.publishing.service.gov.uk/media/69cd1451b5210...
> EDIT: maybe you meant the UK total primary energy? I feel like that's extending the boundary a bit far.
If we're talking about 'sovereign', is it too far? I think the history of conflict in the modern era, hot and cold, shows that your rivals' ability to disrupt your energy imports matters a lot.
If I could give prizes for comments you'd get one. Too much fart sniffing goes on in these parts, it's always a pleasant change to see dissent
Hey Ben. I find communication like this fairly off-putting. In so far the 80% cheaper per token (or any part of it) is something of your own making/ingenuity, by all means, do tell, but it requires comparing token cost fairly with comparable models on i.e. OpenRouter and not across different models and pretending it's the same thing.
Hi jstummbillig, I appreciate the feedback. We’re careful to state only that users can expect up to an 80% cost reduction when switching away from OpenAI. Our DeepSeek V4 Pro model is a good example of this.
https://relax.ai/privacy-policy
"We may share personal information to third parties outside of the UK"
"to find out how long your information is being retained, please see 'additional information'". Additional information is an email address.
https://relax.ai/terms-of-service
non-committal will not share customer data "except[...]with the consent of the Customer". 'see DPA'. there is no DPA on this page.
otherwise,
my use on novita with zero data retention [in, out, cache]: [65.9, 2.4, 424.6][1.74, 3.48, 0.13] = $178.22
i couldn't see cache price on relax, so [65.9, 2.4, 424.6][1.17, 2.33, 1.17] = 579.48gbp = $774.45
Hi Luke - both points are fair and worth answering properly.
On the policy: you’re right, that wording is too loose for what we actually do. The inference path is UK-only - no model inference, no prompt or response data, leaves UK jurisdiction.
The “third parties outside the UK” language covers things like payment processing and standard SaaS sub-processors, not customer inference data, and it should say that. We'll be re-writing this next week!
Marketing aside, why are you using the term "UK sovereign"?
I assume UK based DCs, so why not just say that, UK based LLM inference.
Is it a DC owned/ran by HM Gov? Is that why it's sovereign?
Not a criticism, more of a critique.
Much the same way the word "patriotism" is more common in US national discourse, "sovereignty" is very common in UK national discourse.
You're thinking of when chavs used to wear sovereign rings... it's fell out of fashion now tbh
Data Sovereignty as a term is now fairly well established term that doesn't have specific government connotations e.g. https://events.linuxfoundation.org/kubecon-cloudnativecon-eu...
He does say that? "running on NVIDIA Blackwell GPUs in the UK" is in there and that's pretty unambiguous.
The problem is if those GPUs are running on an AWS server (or any other American provider), even if it the server is in the UK the sovereignty claim is null and void.
Doesn't "We built it on fully UK sovereign cloud infrastructure, so data never leaves UK jurisdiction" cover that?
In theory it should, but I've seen that language describing Azure "sovereign cloud" servers before. The data might indeed be stored in the UK, the problem is the CLOUD act which supersedes it.
Civo isn’t a US company and isn’t a subsidiary of one. UK-incorporated since 2018, UK-resident leadership, no US parent, entirely UK founder owned. We also share the concern on the Azure/AWS “sovereign cloud”, but this doesn't apply here. The CLOUD Act reaches data held by US entities regardless of where the servers sit, which is exactly why those offerings are problematic. We’re outside that jurisdiction by corporate structure, not just geography.
I must be missing something, but is the API post-paid only? I couldn't see where to add pre-paid credit to the account, like OpenAI & Anthropic offer. I'm not really looking to be on another subscription and I would be accessing via third-party harnesses anyway. I'd also want to opt-out of training, which I understandably can't do on the free plan.
Is your business plan essentially to run mid tier models on hardware in the UK?
I do see the value in this as some enterprises need local data residency, the UK energy grid realistically can't handle new multi GW xAI-style data centres, and many applications don't need frontier models (but do need more than small local ones).
So the pricing is 12.50/month for unlimited chat, or 60p per million tokens output/10p per million input? For use with a coding assistant it would be the latter?
Hi Ben - how are you positioning yourself vs LocAI? I had a few chats with them and they have a fairly similar pitch.
We’re closely partnered with the LocAI model lab, we’re looking forward to running their models on the platform in the next few months!
good prices, but I don't see info on the token cache hits prices (in/out), are they available?.
Congrats on the launch. More options for consumers in this space the better!
why use this over openrouter?
I'd expect for workflows where there is value in knowing that the data is processed in the UK. From a contractual/data protection standpoint, that could be very useful, depending on the use case.
[dead]
While I'm British, based in the UK, seeing prices in £ really throws me
Token prices should be in $ as that's how our brains work
Why'd you want your expenses be billed in a foreign currency if you can help it?
Converting once, at the time you pick a provider, is trivial compared to having to continuously think about it as you're paying them.
Likewise, strange right. I live in Japan now, but even living here, I just expect all online provider pricing to be in USD.
UK sovereign data? Land of arrests for posts on social media? Member of five eyes, "you spy on our citizens and we'll spy on yours and call it intelligence sharing"? Land of the infamous Online Safety act? That UK? Why would anyone want their data in the UK?
Having data processed in the UK is not for privacy reasons, but legal reasons when working in Govt. fields such as legal/healthcare/security. For privacy most companies choose juristications with better privacy laws such as switzerland.
Because they reside in the UK? Why would anyone outside the US want their data in the US, the organiser and leader of the Five Eyes?
There is a long list of countries other than the US and the UK. I will go to bat for the US on this one though and say that one might want their data in the US because of the first amendment. Even for people that reside in the UK, what is the selling point of having data with them in the same country?
For all its flaws and limitations in practice, you have far greater data privacy in the UK than the US. Largely because it seems you basically have none as a private citizen in the US.
First amendment has nothing to do with it.
Did you know people live in the UK?
Just to pick on one thing, do you think incitement to murder people by setting fire to hotels, acts that subsequently happened, should just be shrugged at?
What I think about the particular case you're talking about is irrelevant. I guess we'll see if allowing the government to police speech is still such a great idea when Reform and all future governments are in power.