AI that runs
where you decide it runs.
Most AI tools answer you by sending your files and your half-formed thoughts to someone else’s servers. The problem is not the server. It is that nobody told you. We build two products that put the work somewhere you chose on purpose.
Armore Edge
A desktop app that runs a real AI model on hardware you already own. Point it at your files; the reasoning happens on your machine. The model is third-party open weights we package — we did not train it.
Not a one-machine product. Join your devices into a workspace and they work as one private system: encrypted messaging, offline messages that deliver when a device returns, and answers grounded on a document sitting on another machine — your question crosses as an embedding, and the passage that answers it comes back. Host a session for a colleague, or push a repo between your own machines.
Local after setup
Once the model is installed, asking works offline. No account, no server call for an answer.
A mesh with visible edges
Your devices link into one encrypted mesh. When a direct link is not possible a relay carries the traffic: it cannot read anything, and it can see which devices are talking and when.
Builds you can check
94 files on the public mirror — the 47 payload files among them each carrying their own Sigstore attestation of how and where it was built, 1:1 — binding that payload’s digest to the repository, workflow and commit, logged publicly. 22 of those payloads are CycloneDX bills of materials, so you can see what is inside before you install it. A signature over a digest — not an OS-trusted signature.
Coming soon
Hosted — reach your own device from a browser, without setting up a network.
Pipeline
Compute sharing, mesh file transfer and group-membership propagation are built but not yet witnessed on two machines. Code-signed Windows and Linux builds, and end-to-end auto-update, are intended and not started.
Armore Sovereign
The enterprise line: the model runs on hardware you control, inside your own perimeter, for work that cannot go to a cloud model at all. Every deployment is a custom implementation — we scope it with you rather than sell you a tier.
Legal is the first use case, and it is coming soon. We are also building for healthcare, space technology and defence — areas we take on under contract, shaped by what the engagement requires. We hold no certification and claim no compliance regime — we are doing SOC 2 readiness work for a cloud tier, with no auditor engaged and no report, and any assessment in your industry remains yours.
Sovereign is coming soon, and every deployment starts as a conversation about what the work actually requires.
Placement is the difference
Not two tiers of one thing. Edge runs on the device in front of you and you can download it today. Sovereign runs inside your perimeter and is built to your engagement.
| Product | Where your work runs | Operated by | Status today |
|---|---|---|---|
| Armore Edge | On your own device | You | v1.0.10 published · free to download |
| Armore Sovereign | Inside your perimeter | You | Coming soon · custom implementation, scoped with you |
How we work
No photos, no titles, no headcount. A small team builds all of this, and how the work is done tells you more than who does it.
Limitations ship with the feature
A limitation is written at the same time as the thing it limits, so the bound on a number ships in the same sentence as the number.
Numbers carry their method
Every figure here names how it was counted and against what control. Anything we could not re-derive is marked unverified, not rounded into a claim.
Checks must be able to fail
Automated checks ship with their own negative tests, so a rule that stops firing breaks the build. A check nobody has watched fail is not evidence.
Neither one is asking for your money
Edge is free to download today, Sovereign is in build, and there is nothing to buy here. If your work turns on a constraint — a contract, a statute, a privilege posture — tell us now, while both can still be shaped by it.