Meta
Hyperscale AI infrastructure
We provided Meta with a customized MarsRover to help its engineers optimize internal AI workloads. Read the story.
Company
We started with the hardest layer: seeing and improving what a GPU does inside a kernel. On top of it we now run an OpenAI-compatible model API, a tracing service, and applications of our own.
What we do
Token Factory serves open text and video models through one OpenAI-compatible API, paid for by the token.
Token FactoryMarsRover and MarsTrace show what a GPU is really doing, and agents use them to make kernels faster, turn after turn.
TechnologyAI-native products and solutions for industries, enterprises and tasks, starting with Paper Review Studio.
SolutionsWho we work with
Our customers include Meta and 4Paradigm, among others, and we collaborate with AI Singapore.
Meta
We provided Meta with a customized MarsRover to help its engineers optimize internal AI workloads. Read the story.
4Paradigm
Enterprise AI has to perform on the hardware a business already owns. Getting that across mixed accelerators is a planning problem, not a hand-tuning one.
AI Singapore
AI Singapore develops the SEA-LION family for the region's languages. Qwen-SEA-LION v4.5 is live on our API today.
Work with us
We work with teams that run production AI systems, where performance on the hardware is critical.
A technical look at your workloads and the hardware they run on.
Collaboration with hardware, platform and ecosystem teams.
Work with us directly as a design partner.
Join us
Full-time roles in Singapore or the United States, working on LLM serving and GPU kernel optimization. Visa sponsorship and relocation assistance are available for eligible candidates.
Sign up for the API, or talk to the team about your own kernels and hardware.