LLMs · Physics operators · Detectors

Foundation models, trained on your world.

We adapt a proven base model to your domain — and ship the checkpoint together with the data it was trained on.

Request a custom build Models on Hugging Face

30,000+ downloads and growing

Language

84%

MedQA accuracy — Thinking-LQ-1.0

Physics

1.8%

Wake error — Flow-1.0 on Poseidon-B

Vision

Custom detectors & segmenters

weld 0.97

defect 0.91

Open weights

30K+

Hugging Face downloads

How it works

Adapt a proven base — never train from scratch

The same four steps behind every checkpoint we've shipped, whether the domain is medical text, fluid dynamics, or a factory camera feed.

Start from a proven base

A frontier open checkpoint — a reasoning LLM, a physics operator like Poseidon, or a YOLO backbone — brings the general capability, so your data doesn't have to.

e.g. DeepSeek-R1-Distill-Qwen-32B

Adapt it to your domain

We fine-tune on your corpus, geometry, or labels — QLoRA and GPTQ for language, operator learning for physics — without erasing what the base already knows.

your data stays on your infrastructure

Evaluate in the open

Every build is benchmarked against frontier models on public tasks and unseen regimes — accuracy, speed, and cost per query on your hardware.

84% MedQA — within 4 points of GPT-4o

Ship the whole build

You receive the checkpoint, the training data, and the eval report — compressed to run self-hosted on a single datacenter GPU, no third-party APIs.

~20GB — self-hosted on a single L40 GPU

Backed by the programs we build on

Compute credits, cloud infrastructure, and industry partnerships that keep our training runs and open checkpoints moving.

CHAPERONE AI

Foundation models, trained on your world.

Pick the line that sounds like your job

Language

Physics

Vision

Adapt a proven base — never train from scratch

Start from a proven base

Adapt it to your domain

Evaluate in the open

Ship the whole build

Platforms powered by our models

DeepResearch

PropensityAI

NumericalAI

AI 3D Generator

Simvera

Backed by the programs we build on

Different corpus, geometry, or label set?