Eptesicus Laboratories
On-device AI with predictable unit economics.
We build high-performance small language models and the reliability infrastructure to deploy them on customer-controlled hardware, reducing dependence on per-call cloud APIs.
ABOUT
Our objective is to make on-device AI the default for real products.
Cloud AI is easy to start with, but expensive and restrictive to scale. On-device deployment gives teams control over cost, privacy, and availability, but only works when models are efficient and behavior is reliable.
We develop small language models optimized for strong performance per compute and pair them with reliability engineering so enterprises can ship on-device intelligence with confidence.
The Problem
Cloud-first AI creates compounding costs, vendor lock-in, and data exposure. Meanwhile, many small models still struggle with consistency and safe failure modes, which blocks serious deployment.
Our Approach
We build for shipping, not demos.
improve capability per compute in small language models (measured against public baselines and internal evals).
datasets + evaluation harnesses that catch regressions and quantify improvements.
packaging, controls, and integration patterns that work for enterprise and OEM/ISV distribution.
Vision
A world where AI runs primarily on customer-controlled hardware, with small models that are both capable and dependable. Eptesicus Laboratories will build a portfolio of on-device models and reliability tooling that can be licensed, bundled, and supported at scale, while also releasing community-facing applications that prove the on-device experience can feel top-tier.
“We’re building the foundation for on-device AI to scale commercially: efficient small models, reliability you can measure, and distribution through enterprise licensing and OEM/ISV partnerships.”
Active Directives
Currently seeking pre-seed funding to expand reliability datasets and evaluation, secure early enterprise pilots motivated to cut API spend, and deepen OEM/ISV partnership pipelines for bundling.
Lumis-1
[IN DEVELOPMENT]Our first commercial product: an on-device assistant designed to validate the business case for on-device deployment (cost control, privacy, availability) and fund continued model and reliability R&D.
- On-device
- Offline-capable
- Reliability layer