Data Sovereignty as a Service

They told you it's safe
because it's encrypted at rest.

At rest, in their datacentre, with their key. Encryption you don't hold is a policy, not a control.

Sovereignty by architecture,not by promise.

The Five Pillars

Residency answers one.
Architecture answers all five.

Most sovereignty conversations stop at pillar three. This is a walk through all five — each one a question a buyer eventually has to answer, and the architectural answer SkyeConnex gives it.

01 Security Who can see it?
02 Legal Who can seize it?
03 Residency Where does it really live?
04 Portability Can you leave?
05 Algorithmic What's learning from it?
Security

A breach of any one provider is a breach of nothing.

Who can see it? Conventional security is a wall you hope holds — and every operator, admin and provider behind it is another door. We don’t just lock your data, we scatter it: encrypted before it ever leaves your hands, then split into fragments no single provider can rebuild.

Per-file keys · AES-256-GCM · post-quantum ready today with ML-KEM-1024 · erasure coding that widens itself and never narrows

“How new files flow” in the SkyeConnex app: thirty days of uploads drawn as ribbons from six file types on the left, through a line of six placement gates, out to seven connexions on the right. Documents carry 136 files, data and databases 27, archives 10, GIS and geospatial 9, video 6, and email and calendar 6. Each gate shows its live state: Selection, Carbon and Space are off; Geography is amber and narrowed, with a badge reading 2 for the two connexions its policy removed; Routing is red and marked with an exclamation because GIS and geospatial has nowhere to go — that category carries a red “No home” chip and its ribbons are drawn dashed red, falling back to the whole pool; and Coding passes, in green. Five connexions sit inside the policy — AWS S3 in Montreal and Calgary and Azure Toronto in Canada, Azure Paris and Wasabi Frankfurt in the EU. The remaining two, Google Cloud Storage in Virginia and Backblaze B2 in California, are outlined amber and read “In US, outside CA, EU”: they still hold shards, so they are flagged as drifted rather than removed.
Thirty days of uploads, drawn — every ribbon encrypted before it leaves your side. A Canada-and-EU policy narrows the pool by two, and the amber 2 on Geography says which rule did it. One category has nowhere left to go, so Routing turns red and its ribbons fall back dashed. The two US connexions still holding shards are flagged, not hidden. The gates tell you what each rule removed — including when the answer is awkward.
Legal

You can’t hand over what you don’t hold.

Who can seize it? Residency doesn’t stop a subpoena — if your provider answers to a foreign government, so does your data, wherever the server happens to sit. No single operator or jurisdiction ever holds enough to rebuild it, so there is nothing complete to produce.

Signed audit trail · DGSI 100-8:2026 conformance · Protected B ceiling

The SkyeConnex Threat Simulator with the CLOUD Act preset applied, which selects both the United States and the United Kingdom. It reports the blast radius: two affected providers — Google Cloud Storage in Virginia and Backblaze B2 in California — 2,348 of 8,092 shards lost, 1,120 files still safe, 36 unrecoverable, and integrity falling from 100% to 97%.
Pick a jurisdiction and model the seizure. The CLOUD Act preset takes the US and the UK at once: two providers gone, 2,348 of 8,092 shards lost, 1,120 files rebuild from what is left. It also tells you the 36 that would not — a blast radius you can read is worth more than one you are promised.
Residency

Stop asserting residency. Start proving it.

Where does it really live? This is where sovereignty theatre lives: a maple leaf on the website while the keys and the control sit somewhere else entirely. You set the placement policy, every file obeys it — and the policy shapes the maths: the erasure coding is chosen against the provider pool your rules actually leave, so a tighter jurisdiction buys wider parity rather than thinner cover. When an auditor asks, you show them instead of reassuring them.

SkyeMap — live shard geography by file, folder and account · coding resolved per upload, after geo-policy

SkyeMap in the SkyeConnex app: a world map of where a customer’s encrypted shards sit across seven connexions — AWS S3 in Calgary and Montreal, Azure in Toronto and Paris, Google Cloud Storage in Virginia, Backblaze B2 in California and Wasabi in Frankfurt — carrying between 1,145 and 1,192 shards each, 8,092 in all. A location-confidence legend marks six as traceroute-verified and Backblaze as declared but unverified.
SkyeMap, live. Seven connexions, 8,092 shards, 12 traceroutes — and a confidence legend that separates the six locations we verified from the one a provider merely claims.
Portability

Leaving should be your decision. So should staying.

Can you leave — and can they make you? A vendor you can’t leave sets your terms. A vendor who can switch you off sets them absolutely: sanctions and court orders have killed access to fully paid services overnight. Keep your hyperscalers, demoted to interchangeable shard buckets. When one fails, is acquired, or cuts you off, your data reconstitutes elsewhere and recovery runs without us.

Escrow-backed continuity · re-point one S3 endpoint and your tooling inherits it · no one provider can revoke what no one provider holds

The “Rebuild shards from parity” dialog in the SkyeConnex app. With Backblaze B2 in California unreachable, it offers to reconstruct 642 shards across 642 files from their siblings. Automatic is selected and recommended; Azure in Quebec City is eligible; Azure in Paris is refused as read-only, taking no new files; and Google Cloud Storage in Virginia is refused because the US is outside the account’s allowed regions of Canada and the EU. The summary states that all 642 shards go to Azure Quebec City and that no file ends up with two shards on one connexion.
A provider goes dark and you keep working: 642 shards rebuild from parity onto a connexion you already own. Note the two it refuses — one read-only, one in a region your policy does not allow. Recovery obeys the placement rules; it does not quietly break them to finish.
Algorithmic

Use AI on your most sensitive data. Feed nobody else’s model.

What’s learning from it? Hand regulated data to someone else’s AI and you have disclosed it — you just called it a prompt. SkyeGXU runs training and inference against custody objects, reassembled only inside your boundary, on infrastructure you govern.

Governed distributed AI on customer-owned hardware since 2025

The Vectors tab of SkyeGXU in the SkyeConnex app: 17 text vectors and no images, all embedded with all-MiniLM-L6-v2 at 384 dimensions on the customer’s own three agents. A corpus map groups them into Contracts, Finance and Matters, accounting for 37% of the variation, beside a list of the source documents — agreements, an NDA template, a licence renewal, a data-processing addendum, a lease amendment and a matter brief — each row carrying its model, dimensions, agent and the hash of the toolchain that produced it.
Every vector records what produced it: the source asset in custody, which of your three agents embedded it, the model and dimensions, and a hash of the toolchain. Agents connect over MCP — the model comes to your data.
The compound effect
"An adversary attempting to read a single SkyeConnex file must simultaneously defeat the wrapped DEK, compel five providers across multiple jurisdictions, and complete both before the audit log surfaces the activity. There is no published successful attack against this composition."
The shift

A perimeter can only ever react.

It cannot act until something arrives at it. That is not a flaw in anyone’s product — it is what a perimeter is. The industry conceded this years ago and called it assume-breach, then answered it by reacting faster: detect sooner, respond sooner, contain sooner.

Defend the perimeter

Keep attackers out. When that fails, find out quickly and contain it. Every control in the chain works only while someone behaves — the alert fires, the team responds, the provider stays honest.

Still a promise to react.
Remove what it was guarding

Encrypted before it leaves your hands, then split into fragments no single provider can rebuild. The perimeter can be crossed. What is behind it does not reassemble.

Nothing to react to.

Cross the perimeter and you hold ciphertext fragments below the threshold that rebuilds them, with no key in the same jurisdiction. Endpoints still matter — plaintext lives where you work. Everywhere else, the breach stops being an event and becomes a non-result.

0.7×
stored volume
Sustainability

Resilience that costs less than a single copy.

Default 5+2 sharding with compression stores roughly 0.7× your original data and survives the loss of two locations — against about 3× for conventional replication. We also record the grid carbon band behind every endpoint, and say plainly where we don’t know.

See our sustainability posture →
Get in touch

Sovereignty is no longer a slide deck.

Forty-five minutes, live, on your own data. We'll show you all five pillars working — and you can decide for yourself.