Claude Science: an AI research agent that runs on your own lab infrastructure

Anthropic has launched Claude Science, a public beta app—not a new model—that wraps its existing Claude models in tooling aimed at computational life-science work. The app installs wherever a researcher’s data lives (laptop, lab Linux box, HPC login node, or cloud VM) and orchestrates real compute: it builds environments, writes batch scripts, and submits jobs over SSH to a Slurm cluster or through a Modal account, keeping variables and loaded models in memory across an analysis for fast iteration. It ships pre-configured for genomics, single-cell, proteomics, structural biology, and cheminformatics, can query 60+ scientific databases, and connects natively to open models via NVIDIA’s BioNeMo Agent Toolkit, including Evo 2, Boltz-2, and OpenFold3.

The central pitch is reproducibility and provenance. Every figure, table, or notebook carries the exact code, environment, plain-language description, and conversation that produced it, so results can be re-run or defended months later. A background reviewer flags untraceable numbers, mismatched figures, and bad citations before findings surface. Researchers can view proteins, structures, genomic tracks, and PDFs natively, annotate figures to request edits, and save pipelines as reusable skills or wire in lab tools through connectors.

On data handling, raw datasets and compute stay local while prompt and response content is processed by Anthropic under standard retention. It’s available in beta on macOS and Linux for Pro, Max, Team, and Enterprise plans, with SSO, SCIM, and usage analytics for enterprise admins, plus a discounted Team tier for verified academic and nonprofit labs.