Chroma (Generate Biomedicines) is an open-source generative model for programmable protein design, released by Generate Biomedicines on GitHub, where it has about 840 stars under the Apache-2.0 license, with updates in 2026. Note that this is a distinct project from the similarly named vector database product also listed on Tooloso; the two share a name but do entirely different things.
Protein space is vast and hard to navigate, so Chroma represents design problems as composable building blocks called conditioners, from which diverse all-atom protein structures are generated automatically. As a joint model of structure and sequence, it also handles tasks such as generating sequences for given backbones, packing side-chains, and scoring designs. Internally it combines diffusion modeling, equivariant graph neural networks, and conditional random fields, sampling structures with sub-quadratic complexity in the number of residues and generating large complexes in minutes on a commodity GPU.
The design philosophy is composition: a researcher specifies what a protein should look like by combining conditioners โ for example fixing a substructure, enforcing symmetry, or matching a target shape โ and the model samples structures that satisfy those constraints jointly. Because structure and sequence are modeled together, the same system can propose a backbone and then design sequences that fold into it, or score how plausible an existing design is. The underlying research was published in the journal Nature, and the codebase includes examples for sampling, design, and building custom conditioners.
Chroma is free and open source under the Apache-2.0 license. Researchers can download the code and pretrained model and run them on their own hardware at no cost.
Chroma is for computational biologists, protein engineers, and drug-discovery researchers who want a programmable generative approach to designing novel proteins with specific structural properties.