biopb is an open protocol and toolchain for AI-assisted bioimage analysis. Stream microscopy images far larger than memory, and let an AI agent run real analysis in a live napari session — results in the viewer, answers in chat. Watch a demo.
$ curl -fsSL https://biopb.org/install.sh | bash
One command. Sets up the data server and wires the agent into Claude Code, OpenCode, or Cursor.
Drive a live napari viewer through an AI agent. One install command registers biopb with Claude Code, Cursor or your favorite AI assistant — then ask for the analysis you actually want. Intermediate results rendered in the viewer; counts, measurements, and explanations come back in chat. You control the iteration.
Point biopb at your files and it reads them — TIFF, OME-Zarr, and more — and serves every dataset as a uniform, lazily-streamed array. Allow your agent to browse data catalog and work on multi-dimensional images that are far larger than your machine's memory.
Tool for your agent on problems they cannot solve out-of-the-box. Run as standalone servers your agent can call. Use the hosted ones that ship preconfigured, or create your own on your metal. Adding a new algorithm is wrapping a model, not building a server.
biopb is an open protocol with client libraries for Python, Java, and JavaScript. Bring your own data sources, algorithm servers, and agent tools for your lab's own problems — nothing is locked to a vendor or a single application.