genome-writing

A verification and grounding layer for AI-driven genome writing. Foundation models and a grounded co-scientist agent propose edits; PEN-STACK checks each against validated tools, never a language model: where a write is safe and durable, which enzyme makes it, and where it goes off-target. Calibrated, provenance-tracked, pre-registered.

Transcript-unit architecture database for disease-associated large-fragment genome-writing target review.

A rule-governed, genome-writing-native biosecurity middleware that supervises agentic design AI. It inspects the plan a design agent produced, not just the final sequence, across five hazard axes and returns allow / flag-for-review / refuse with cited evidence and a signed design passport. Open-data, safe-proxy reference implementation.