| Filename | Latest commit message | Latest commit date |
|---|---|---|
Evolve protein-q-fold-bind from a Modal-hosted surface toward a sovereign one — same Boltz-2/ESMFold2 science, no cloud tax, data never leaves the room. Built with standardgalactic (Nate); powers the private pharma pilot. For Rahul and his father. |
||
| README.md | ||
proteinFolding — the sovereign protein lane
Fold structures, design binders, on hardware you own. The evolution of
protein-q-fold-bindfrom a Modal-hosted surface toward a sovereign one: same science, no cloud tax, data never leaves the room.
8b.is · the constellation · 2026.10.08
why this repo exists
protein-q-fold-bind proved the science: Boltz-2 folds, ESMFold2/ESMC
designs, Gradio + Molstar shows it — all live, all working. But it runs on
Modal (someone else's cloud). That is fine for research and wrong for the
people who need it most: regulated labs and pharma businesses, where sending a
target or a sequence off-site is a legal and trust wall.
This lane is the next step: the same folding and design, on owned silicon.
protein-q-fold-bind (today) |
proteinFolding (this lane) |
|
|---|---|---|
| compute | Modal (cloud) | owned GPU box |
| data | leaves the building | never leaves the room |
| cost | per-run cloud | capital, not rent |
| for | research | regulated labs, pharma, pilots |
what it is
- fold — run Boltz-2-class structure prediction locally (sequence → structure, Molstar view).
- bind — design binders with ESMFold2/ESMC-class models on a local box.
- own it — the box is yours; the pipeline is portable; walking away keeps everything.
the shelf it sits on
protein-q-fold-bind— the proven fold/design surface (Modal).qbit-protein-studio— the studio lane.kompress-ultra— context compression (keeps the expensive context small).knowledge-worker— private, provenance-backed memory: which target, which run, which model, why.project-zero— the sovereign inference engine.
the bridge — Nate (standardgalactic) and the clinic
This lane is built with standardgalactic
(Nate), already a member of the protein lane. His
biomechanical_DLIR
is the clinical sibling: physically-constrained deformable registration for
medical imaging — the same sovereignty question, applied to anatomy instead of
molecules. Folding and registration are two faces of one platform: model the
structure, own the compute, keep the data.
the pilot it powers
The sovereign-pilot proposal offers a private
4–6 week pharma pilot: one owned GPU box, one open protein model, one dataset
that never leaves the room. This repo is the machinery behind that promise —
the fold/design lane, running sovereign.
status & honesty
- proven: fold + design work today (
protein-q-fold-bind). - in progress: the sovereign port — folding/design models on owned silicon.
- honest: open protein models do not replace lab validation; results on any target are not guaranteed. This lane measures whether owned compute meaningfully helps real work, cheaply and privately.
the line
∅R ⟶ᴸ R₁ ⟶ᴸ R₂ ⟶ᴸ ⋯ ⟶ᴸ Rₙ ⟶ᴸ ⌂
│ μ(Rᵢ, you) ≠ 0 ∀i ∴ μ(⌂, you) ≠ 0
For Rahul, my brother — and for his father. <3