The repository accompanies work on a whole-brain in vivo CRISPR Perturb-seq atlas and assigns independent AI agents to individual perturbations.
The agents combine cell-type-resolved expression data with literature databases, then produce narrative and structured reports. The project says it can synthesize those reports to recover known biology, predict held-out responses, identify links between perturbations, and prioritize follow-up hypotheses. The repository includes the runtime, corpus tooling, notebooks, scripts, figures, and supplementary-analysis materials.
A hosted corpus offers read-only Findings, Claims, evidence, and release-specific analysis tools through public endpoints, including MCP connections for compatible assistants. It does not expose raw dataset files or support recomputing dataset statistics. Local use defaults to the Codex CLI; the full atlas download is listed as 10.8 GB, with about 35 GB free disk space required during preparation.
The repository does not establish that the reported biological predictions are independently validated, that results generalize beyond its supplied atlas, or how its approach compares with alternative interpretation workflows.