Esm

This connects you to EvolutionaryScale's protein language models for generative design and embedding tasks. ESM3 handles the generative side: you can design new protein sequences, predict structures, do inverse folding, or condition on functional properties using multimodal prompts. ESM C gives you embeddings for ML tasks. You can run small models locally with open weights (Python 3.12 required) or hit the Forge API for the bigger models like the 98B parameter ESM3. The API mirrors the local interface, which is nice. If you're working on protein engineering, variant design, or need quality protein representations without training your own model, this gets you current SOTA tools with minimal setup.