augmntd labs.
Independent projects and research from augmntd. Each one an experiment in how AI can work for people.
Right now, the research here focuses on two things: measuring models in the environment they'll actually live in, and making local inference fast enough to use all day.
Companion Bench
A benchmark we're developing inside Cube. It runs models natively in the environment people actually use them, with real files and real context, across a range of everyday tasks. Less about puzzle scores, more about whether a model is actually useful.
Local inference
Getting more out of open-weight models running locally. We're working the pipeline end to end, tuning for the hardware we're building, so a local model can keep up with your thinking on the machine it runs on.
Experiments.
Small, fully-featured apps we build quickly to try a new model, a new interaction, or a new idea in a real product.
Learn more Auglie · 2025
Virtual try-on for fashion. See how new pieces look on you, style them with what you already own, and build outfits that actually work.
Learn more Thoughts · 2024
A journal that creates long-term context about yourself. Building on insights from Mind, Thoughts captures your memories and reveals the themes in your life.
Mind · 2023
Knowledge collection and discovery with proactive assistance. Our first exploration of agentic chat patterns and long-term agent memory.
More to come.
We're always experimenting. Follow our journey as we explore new ideas.
