xuzhougeng/wisp-science

Wisp-science is a desktop AI workbench aimed squarely at computational scientists who need to run real analysis pipelines, not just ask questions about data. It connects Python and R runtimes, MCP-based bioinformatics tooling, SSH remotes, WSL, and GPU runtimes under one interface, with OpenAI and Anthropic model support. What makes this worth looking at this week is the local-first architecture: everything runs on your machine or your own remote, so sensitive genomics or clinical data never touches a third-party server. Most AI coding assistants stop at Python notebooks and assume your data is public. This one is built around the constraint that it often is not. The MCP bioinformatics tool layer is the actual differentiator — if you are running BLAST jobs or wrangling VCF files, that matters. Reservation: it is clearly early-stage and the documentation looks like it is catching up to the code. If you are not already comfortable with Python and R environments, the setup will bite. -> Best for: ML researcher or bioinformatics engineer who needs LLM assistance inside a private compute environment.