Scope

Run the Dayhoff reference app locally to generate novel protein sequences, score variants, scaffold motifs, and (for some variants) condition on homologous sequences.
Three ways to run the reference app, in increasing order of capability and setup cost:
1. Path A — Cached demo: frontend only (npm run dev), no GPU, no backend. Returns precomputed real Dayhoff outputs for four bundled prompts (Cas9, insulin, DNA polymerase, SARS-CoV-2 spike) across all four model variants. Best for evaluating the UX or demoing on a laptop with no GPU access.
2. Path B — Proxy mode: Flask proxy on :8000 + frontend on :5173, calling a dayhoff-multi Azure ML endpoint. Two terminals, no local model load. Live arbitrary-prompt generation.
3. Path C — Fully local: adds a self-hosted score server on :5001 (and optionally a fold server) running on a CUDA-capable NVIDIA GPU. Three to four terminals.
Hosted models are also published to Hugging Face and Azure AI Foundry — use the proxy in backend/app.py to call a dayhoff-multi AML endpoint, or use the transformers library directly against the Dayhoff model collection.
Four model variants (3b-GR-HM-c, 3b-GR-HM, 3b-UR90, 170m-UR50-BRn) with different strengths — see docs/about-dayhoff.md for the picker.
Bundled biosecurity guardrail: backend/sequence_screening.py runs a 3-layer Select Agent toxin screen on inputs and outputs. This is a guardrail, not a guarantee — see docs/responsible-use.md.

msresearch-dayhoff