Using AI to advance science in every domain and to reduce its cost.
Blankline is an AI research lab. We treat AI as a research instrument: a way to do more science, in more fields, with less energy than the previous decade of methods required. The product of that work is the research itself, and the tools we ship so others can use it.
Four positions that shape every project.
Across every domain.
We do not pick a single field and call it the future. AI is now useful wherever measurement, modelling, or search shows up. We work in physics, biology, astrophysics, AI safety, and the domains where those overlap.
Energy as a constraint, not a footnote.
Frontier-scale compute is now a meaningful share of global electricity. Treating that cost as an externality is the wrong frame. We measure it, publish it, and design our models and our workloads to reduce it. The Joule Index is part of that public accounting.
The runtime is the product.
A model is one ingredient. The product is the system around it: which weights you run, where inference happens, what each tool call is allowed to do, and who approves it. Dropstone is our reference implementation of that posture for coding agents.
Honest about what we cannot prove.
No party can prove a foundation model is free of embedded behaviours. The runtime is what makes model origin a non-issue for execution safety. We say this out loud rather than imply that the problem is solved.
Research, instruments, products.
Each thing we ship serves the same mission from a different angle. The research is where the claims come from. The instruments are how we measure them in public. The products are how we make the result useful to someone who has work to do today.
What we hold ourselves to.
These are the rules that govern how the work above is done. They apply to every product, every paper, every model release.
Read the governance frameworkOpen research, openly published.
Methodology, evaluation scripts, and benchmark numbers are public. The Joule Index is the most visible example; the same standard applies to every paper we ship.
External oversight on consequential decisions.
Major model deployments, training-data decisions, and tier promotions pass through the Governance & Advisory Council. Publications pass through the Research Integrity Council. Both are public.
A single security model across products.
Every Dropstone tier — Fast, Pro, Heavy — and every future product runs the same approval-gated runtime on the same US-hosted, zero-retention inference. There is one security model, not one per release.
Work on this with us.
Read what we publish. Use what we ship. If the mission lines up with yours, the open roles are where to start.