Add lumen to AI and Agents > Data Layer

[ghostiee-11]

Adds lumen to AI and Agents > Data Layer.

Quality requirements

• Python-first: yes, pure Python built on Panel and DuckDB
• Active: commits today; releases in January, March and June 2026
• Stable: 1.2.1, classified Development Status :: 5 - Production/Stable
• Documented: https://lumen.holoviz.org with tutorials, configuration guides and API reference
• Established: repository created May 2020

Acceptance criteria: Hidden Gem

Lumen is at roughly 300 stars, inside the 100-500 band described for this category.

The gap it fills. Data Layer currently covers documents and memory. llama-index is a data framework for LLM applications over documents, mem0 is agent memory, instructor and outlines handle structured output. Nothing here covers conversational analysis of tabular and scientific data, which is a distinct problem: schema inspection, SQL generation, transformation pipelines and rendering.

How it differs from text-to-SQL tools. Those return a query string. Lumen returns a declarative specification, so a generated analysis can be version-controlled, reopened in a notebook, or composed into a dashboard. It also reads xarray and Zarr, so it works on multidimensional scientific data rather than only rows and columns.

Real-world usage. BSD-3-Clause, NumFOCUS-affiliated HoloViz project alongside Panel, HoloViews and Datashader. On PyPI as lumen and on conda-forge.

• Repo: https://github.com/holoviz/lumen
• Docs: https://lumen.holoviz.org