Run a sovereign AI assistant on your own machine.
SIOS installs like an app and runs offline: it codes, builds, and drives the studios on hardware you own. Probe your machine to see what it can run, then apply for a key matched to it. No cloud, no data leaving the device, every action audited.
How the beta works
The probe runs in your browser and reads your graphics adapter and coarse system hints. It never leaves the page until you apply. It tells you which Poros backend your machine uses and roughly which model tier it can run.
The beta is invitation-controlled, so we can support each tester properly and keep the release posture tight. Tell us who you are and what you want to run.
If your machine is a fit, we issue an access key tied to your hardware fingerprint. It entitles exactly the Mickai models your machine can run, and it will not verify on any other machine.
Your key resolves to the correct download for your hardware. The installer verifies every model file against its published checksum, then runs the assistant offline, on your device, with every action sealed to the audit record.
The probe runs entirely in your browser. Nothing leaves this page until you submit the application below. It reads your graphics adapter and coarse system hints to pre-qualify your machine. The exact model match is confirmed at install by the native probe that ships with SIOS.
What you are running
Poros keeps model weights in system RAM and caches the hot parts on the GPU, so system memory, not video memory, sets the ceiling. A modest GPU with plenty of RAM runs far larger models than its VRAM alone would allow.
SIOS runs on hardware you own, fully offline. There is no cloud call, no telemetry, and no model call leaving the device. Any outbound path is off unless you enable it for a named purpose.
Every privileged action is sealed to a tamper-evident, hash-chained audit record before it runs, and the record is signed with a post-quantum scheme (FIPS 204, ML-DSA-65), so a record made today stays verifiable for the long term.