Skip to content

Getting started

This page gets you from nothing to your first agent-driven analysis. For a friendly, click-through installer tailored to your operating system (including Windows), use the interactive install guide. The steps below are the short version.

1. Install biopb

For the impatient, here's the command:

curl -fsSL https://biopb.org/install.sh | bash

irm https://biopb.org/install.ps1 | iex

For the rest, go to the interactive install guide and follow either route.

What does curl … | bash (or irm … | iex) do?

It downloads the setup script and runs it. You can read exactly what it does first — it's plain text at biopb.org/install.sh (and install.ps1 for Windows).

The installer also registers biopb with any agent it finds — Claude Code, Claude Desktop, Cursor, or opencode — so there's nothing to wire up by hand. To check or change that later:

biopb agents list           # which agents are registered?
biopb agents register --all # register biopb with every agent found

2. Open your agent and ask

That's it. Open your AI agent and ask it, in plain language, to do something:

"Connect to biopb and report status"

Your agent launches biopb on its own whenever it needs it, opens a napari viewer, runs the work, and shows you the result. You watch it happen and can adjust the layers by hand.

See Working with your agent for more examples of what to ask.

3. Open the dashboard (optional)

Everything biopb serves in a browser lives at one address. Run:

biopb dashboard

That brings up http://127.0.0.1:8813, where you can watch what your agent is running, browse your data, and check on the data server. You don't need it to get work done — but it's the best window into what's happening underneath. See Working with the dashboard.

4. What LLM should you use?

If the agent workflow is new to you, you should first check what LLM you are using. In opencode (and many other agent tools), you type:

/model

to see current and other available models.

Use a free model or not

You may see models being labeled free. Great! Try all of them to get a feeling of the workflow. But know that they may not be smart enough to do a complicated task. A few smart models that we have tested and are happy with:

Having trouble?

Head to Troubleshooting, or reach out via Help & community.