A doctor in a rural clinic. A soldier in the field. A device with no signal. We are on a mission to make AI sustainable, private, and deployable everywhere.
Vision
Not just the cloud-connected ones.
Our bet: a society of tiny, specialised models, orchestrated and local, running everywhere a human goes.
Mission
Requiring dramatically fewer resources, making it cost-effective and environment-friendly.
Products
Models, not APIs. We don't sell seats, we sell the model that does the work: an agentic platform, compressed open-weights models, and a bespoke compression service.
The Fern research, crystalized. A tiny chat-style AI you train on your own product, drop in with one URL, and run for free, for every user, forever.
Our agentic platform already in production. No-code AI agent builder, marketplace, and CLI.
Production-ready compressed versions of Qwen, Llama, and BGE models, drop-in replacements at a fraction of the running cost.
Send us your organisation's custom LLM. We return a compressed version that preserves quality while slashing compute costs.
Research
From our flagship LLM in development to advances in compression and on-device privacy, we're building the foundations of sustainable AI.
Specialised, production-grade intelligence with substantially fewer parameters. Runs fully offline on a smartphone, Apple Watch, or an embedded chip. No cloud, low latency, full privacy.
Pushing further: substantially fewer parameters, broader model coverage, provable quality bounds for safety-critical applications, and compression-native training pipelines.
On-device inference as a privacy guarantee by design, no data leaves the hardware. We're also researching differential privacy and secure aggregation for federated deployments.
Lab to live
A computer-vision customer needed to recognise food waste on-device, in kitchens with no reliable connection. We turned a cloud-scale DINOv2 vision model into a 50 MB model that runs fully offline on commodity hardware, retaining roughly 80% of its top-5 accuracy.
Before
Cloud DINOv2, GPU-bound, online only
After
50 MB, on-device, fully offline
Supported by
Work with us
Whether you're a hardware partner, an enterprise customer, a researcher, or an investor, we'd like to hear from you.
