Live H100, H200 and B200 ready to launch
The launchpad for private GPUs.
Pick a model and a GPU, and Sigilnet launches it in seconds. Before any of your data goes in, your browser checks a report signed by the hardware itself, so you know exactly what you're running on.
- GPU allocated1× H100 80GB · eu-central 0.8s
- Confidential VM bootedIntel TDX · GPU secure mode on 6.1s
- Hardware report signedCPU and GPU, fresh for this launch 6.4s
- Verified by your browserImage v0.4.2 · all 8 checks match 7.2s
- Endpoint liveOpenAI-compatible API ready 7.9s
Launch to live7.9 seconds
- Intel TDX
- AMD SEV-SNP
- NVIDIA Confidential Computing
The fleet
Pick a GPU. Launch it.
Every GPU runs in confidential mode. Pay by the hour, stop any time.
- NVIDIAMost launched
H100
80GB memory · confidential mode
$4.90 / hour
Launch H100 - NVIDIA
H200
141GB memory · confidential mode
$6.20 / hour
Launch H200 - NVIDIA
B200
192GB memory · confidential mode
$8.40 / hour
Launch B200
Seconds to launch
No quotas to request and no tickets. Pick a GPU and it's running before your coffee cools.
Verified, every launch
Each GPU proves what hardware and software it runs. You check it yourself, in your browser.
OpenAI-compatible
Point your existing client at your endpoint. Same requests, private hardware.
How a launch works
From click to live. Checked by you.
The check runs on your side, so we can't fake it. Our servers only pass messages along.
Pick your GPU
Choose a model, a GPU and a region. Open models are built in, or bring your own, uploaded encrypted.
H100, H200 or B200, billed by the hour.
We launch it
Your endpoint boots in a confidential VM with GPU secure mode on, running our public software.
Live in seconds, not after a sales call.
Your browser verifies it
The CPU and GPU sign a report of exactly what is running. Your browser checks it before anything else happens.
Signed with Intel, AMD and NVIDIA keys, not ours.
Send your key and go
Your browser locks your key so only this GPU can open it. Then call it like the OpenAI API.
We pass the key along, but we can't read it.
What it proves
Every launch comes with proof.
Clear answers for your security team.
A passed check proves
- Your GPU is real, secure hardware with secure mode on.
- It runs our public software, unchanged.
- The report is new and made for your check.
- Only that GPU can open your key.
It doesn't prove
- That our software has no bugs. It's small and public, so anyone can review it.
- That we can't switch it off. We still run the machines.
- Protection from every hardware attack. Some advanced attacks are out of scope.
Your next GPU is one click away.
Launch one now and watch your browser verify it. One demo GPU runs tampered software, so you can watch the check fail.