Let your agent see what you see.

Point to a chart, table, or other result in marimo, a reactive Python notebook. Lens gives your agent the selection, your note, and the code behind it, then brings the agent's work back into view for your review.

645605225-e1975ba6-3087-485e-9360-43a99adc8795.mp4

• Point and ask. Mark a point or region on a chart, table, or other output. Lens keeps the mark, note, image, and relevant notebook context together.
• Follow the work. See where the agent is working, or ask for a guided trail through the notebook's inputs and results.
• Review and continue. The agent resolves your request with a summary of what it changed and how it checked. A Lens that keeps a record lists recent resolved requests in History, with the cells each one changed and whether those changes still hold. Reopen a selection to refine it against the current output.

Get started

Use Lens with an agent connected to your notebook through marimo pair or marimo's code-mode sidebar. Lens is a separate Python package, installed in the notebook's environment.

Follow the quickstart to connect an agent, select a chart, and review your first change, or open the sample notebook in molab:

marimo shows the Lens dock once a cell that imports marimo runs. Keep a record with Lens(record=True) to get History and the code each request changed.

For agents

Read the packaged Lens briefing:

uvx --with marimo-lens agent-plugins read marimo-lens

In a connected notebook kernel, import marimo_lens._capability followed by help(marimo_lens._capability) prints the guidance installed there. marimo pair runs it from a terminal:

marimo pair execute --url <URL> --session <session_id> -c 'import marimo_lens._capability; help(marimo_lens._capability)'

Connect an agent covers connection and resource access.

Documentation

What is Lens? · Getting started · Try the demo · Connect an agent · Python API

Contributing · Report a bug · Apache 2.0

Inspired by Agentation.