Applied AI, local and cloud, for real-world services
AI Academy Lab builds a hybrid AI platform. Models run on your own hardware when data must stay in-house and costs must stay flat, and move to cloud GPUs when demand grows. Your applications see a single API either way.
Our open-source inference engine, PureDrive, is public on GitHub and Hugging Face.
Rhombicuboctahedron, after Leonardo da Vinci’s plates for Luca Pacioli, De divina proportione, 1509.
Most AI pilots never reach daily service
Organizations want AI inside their customer service, documents and operations. Three things stop them.
-
The data cannot leave
Patient files, contracts, case records and internal knowledge are covered by GDPR, professional secrecy or contracts. Sending them to a public model API is often not allowed.
-
Cloud costs grow with usage
Per-token pricing is easy to start with and hard to budget once a service handles thousands of requests a day.
-
Pilots are built as one-offs
A demo wired to one model and one vendor has to be rebuilt to move it on-premise, change model or connect real systems.
One platform, two places to run
The platform decides where each request runs, under rules you set: on local hardware for sensitive or high-volume work, in the cloud for peaks and the largest models.
Local inference is open source today. The other modules are in development.
- a
Local inference
Open models on hardware you control, starting with Apple silicon workstations and our PureDrive engine.
- b
Cloud orchestration
Overflow and large-model workloads sent to NVIDIA GPUs on the cloud you choose.
- c
Data connectors
Documents, email, ticketing, CRM and databases, read with their existing permissions.
- d
Console
Usage, cost, quality checks and audit logs in one place.
- e
API
One OpenAI-compatible endpoint, wherever the model runs.
How it works
- I
Connect your data
Point the connectors at the systems the service already uses. Access follows the permissions those systems already enforce.
- II
Set where it runs
Write simple policies: personal data stays on local hardware, public content can use the cloud, cost ceilings per service.
- III
Ship and measure
Applications call one API. The console shows what ran where, what it cost and how answers scored against your checks.
Built for services people rely on
The same platform adapts to very different work. Six sectors we design for:
-
Customer service
Answers grounded in your own knowledge base, across chat, email and phone.
-
Healthcare and professional firms
Documents and notes processed without leaving the practice.
-
Retail
Product data, stock and support handled consistently across channels.
-
Manufacturing
Manuals, maintenance logs and quality reports, on the factory network.
-
Public administration
Citizen requests and records handled on sovereign infrastructure.
-
Education and training
Course material turned into tutoring and assessment support.
Proven where it matters: inference
Running large models locally only works if it is fast enough. PureDrive is our open-source engine for exact, fast local inference on Apple silicon.
| Context | PureDrive | First lossless baseline |
|---|---|---|
| 2K | 110.3 85.8 | |
| 4K | 107.3 85.1 | |
| 16K | 102.6 70.8 | |
| 32K | 95.8 62.8 | |
| 40K | 90.2 60.4 | |
| 64K | 78.3 50.6 | |
| 128K | 62.8 34.6 |
tokens per second, by context length
Bring AI into a real service
Early access will open to a small number of organizations. Tell us about the service you want to improve.