Small models. Serious work.
AI models trained for one job each, small enough to run privately: in our cloud first, then on modular hardware you own.
Built for the job, not for everything
General models are huge because they try to know everything. Ours are trained for one field each, so they can be small, fast and cheap to run.
- Depth over breadth: A model trained for one field spends its capacity where it counts.
- Efficient per token: Fewer parameters mean less memory and energy for every answer.
- Private by design: Small enough to run on hardware you control, not only in someone else's data centre.
In development: Cybersecurity (log triage, threat explanations, alert summaries). Exploring next: freight and logistics, software engineering, forecasting, and research workflows.
LilaFold Cloud
Our models behind a simple API. Pay per request or subscribe. No GPUs to rent or manage.
Modular NPU Hardware
NPU modules that connect together. Start with one base unit and add more as your workload grows. NPUs (neural processing units) are chips built specifically for AI as a lower-power, lower-cost alternative to GPUs.
Frequently Asked Questions
- What is early access?
- A first group of people who get API keys before the public launch. Leave your email and we will write when the cloud opens, with a discount on your first use.
- When does it open?
- We are an early-stage company and our first model is still in development. People on the list hear first.
- What will it cost?
- Usage-based pricing and monthly subscriptions, with a free tier for prototyping. Exact prices will be published at launch.
- Which model comes first?
- Cybersecurity: log triage, threat explanations and alert summaries. Freight and logistics, software engineering, forecasting and research workflows are the fields we are exploring next.
- Can I buy the hardware?
- Not yet. The modules are in research and the pictures on this page are concept illustrations. We plan to validate a two-module prototype first, then run pilots with early partners.