Tools · Local AI
FARADAY
FARADAY explores running AI models and embeddings on hardware you control. The public proof it is working toward is an egress measurement.
Research. The on-page visual is an illustration, not a live measurement.
What it is
FARADAY works on running model and embedding workloads on hardware you control. Private tests and an evidence viewer have been reviewed. The intended public proof is an egress measurement with a stated configuration and scope.
Why it exists
Environments where the data cannot leave, and any team that has restricted public AI and needs a sanctioned place to send people instead.
FARADAY and JCR are separate tools that answer different questions: where a model runs, and what an agent was allowed to do.
Intended flow, not results
How it works
At a high level only. Nothing here is a result.
- Step 1
Model and embeddings on your hardware
Run model and embedding workloads on hardware you control.
- Step 2
Measure what leaves
The intended public evidence is an egress measurement.
- Step 3
Publish the measurement
With a stated configuration and scope.
Illustration, not a product view
Inside FARADAY
Research · illustration, not a measurement
Where the model runs
- Illustration
1 · Typical public AI service
The prompt leaves your hardware and crosses to a model that runs somewhere else.
- Illustration
2 · FARADAY’s design goal
Model and embeddings run on hardware you control, so the prompt’s round trip stays inside.
Egress: to be measured and published
A drawing of the idea, not a measurement or a deployment. No egress figure has been published. The intended public proof is an egress measurement with a stated configuration and scope.
What you can check
A dash means not measured, not zero.
- Public egress measurement
- —Nothing public to check yet
- Public deployment
- —Nothing public to check yet
- Benchmark
- —Nothing public to check yet
What’s real, and what isn’t claimed
Real today
- Nothing public to check yet.
Not claimed
- The private test results are not public product acceptance.
- No production deployment, customer, benchmark or date.
- No zero-egress guarantee or physical air gap is claimed.
Questions
Who is it for?
Environments where the data cannot leave, and any team that has restricted public AI and needs a sanctioned place to send people instead.
What will the proof look like?
The intended proof is an egress measurement with a stated configuration and scope.
When can I use it?
Not yet. When there is something to show, it appears on this page first.
Other tools
JCR
Checkpoint and receipts for agent actions
Explore JCR PrototypePROBE
Adversarial AI testing, step by step
Explore PROBE BetaRADAR
AI platform changes, explained, with corrections
Explore RADARDiscuss the evidence you need.
The same discipline behind these tools, applied to your systems.