Biopb documentation¶
Open bioimage analysis, driven by your agent.
Biopb lets you analyze microscopy images by talking to an AI agent. You ask the agent to open an image or run an analysis, and it runs python code in a live ipython kernel: enhance contrast, remove noise, running segmentation and other algorithms, and showing results in a live napari viewer you can watch and validate.
The rationale
Traditional image analysis platforms, e.g., Fiji, rely on a curated set of plugins written by human developers. However, writing a universally applicable plugin is hard, even when the underlying algorithm is solid, because the real data in research is highly variable and continuously evolving. The biopb project is an experiment to see whether using AI to generate custom programs on-the-fly works better on scientific data.
Get started How it fits together
What you can do¶
- Open and browse large microscopy data — OME-Zarr, OME-TIFF, CZI, LIF, ND2, and more — even datasets far larger than your computer's memory.
- Do open-ended analysis in plain language. Filtering, measurements, region properties, spatial statistics — your agent writes the code and you watch results appear in napari.
- Run trained models for segmentation and restoration (Cellpose, UNiFMIR, and others) without writing boilerplate.
- Stay in control. Image results land in the viewer; numbers and tables go to the chat; you decide what to save.
Components¶
You mostly interact with your agent and the napari window. Everything else runs quietly underneath:
| Piece | What it does for you |
|---|---|
| Your AI agent | The thing you talk to. It launches biopb and writes the analysis code. |
| Napari + Data Browser | The viewer where images and results appear. Your agent drives it; you can edit by hand. |
| Dashboard | One web page at 127.0.0.1:8813 for watching sessions, data, and logs. |
| Data server | Serves your image data, lazily, so huge files just work. |
| Algorithm servers | Run the dedicated algorithms — usually on a GPU machine. |
If you are just getting started, you don't need to think about the last two — the default install wires them up for you. See How biopb fits together when you want the full picture.
Where to next¶
- Getting started — install and run your first analysis.
- Working with your agent — what to ask and how the workflow feels.
- Working with napari — What do you do in the napari window.
- Working with the dashboard — watch what your agent runs, and check on the system.
- Troubleshooting — fixes for the common snags.
These docs are for end users
Looking for protocol internals, contributor guides, or the gRPC/Arrow Flight specs? Those live in the source repositories on GitHub.