We’re happy to announce the release of Lumen AI 1.0, a major milestone for the project.
Lumen AI is an open-source framework for conversational data exploration that lets you go from natural-language questions to SQL, transformations, and visualizations without writing code, while keeping every step inspectable and reproducible.
Rather than treating LLMs as opaque black boxes, Lumen turns model output into structured, validatable artifacts (for example SQL queries and Vega-Lite specifications) and executes them through a typed, auditable pipeline. This makes it possible to move quickly while still retaining control, correctness, and extensibility.
What’s new in 1.0
This release focuses on stability, clarity, and laying a solid foundation for future features:
- Complete UI rewrite on
panel-material-ui, with a cleaner layout and less visual noise - New typed, context-driven API, replacing global mutable state
- Reworked execution architecture for safely running LLM-generated plans
- Improved database support via SQLAlchemy and broader LLM provider support
- Significant improvements in performance and robustness
What Lumen is today
Today, Lumen is primarily about fast, structured insights without coding. You can connect to data, ask questions in natural language, generate SQL and visualizations, and inspect exactly what was produced.
Where we’re heading next
Lumen 1.0 also establishes the foundation for what comes next:
- Persistent and shareable explorations
- Exportable, repeatable reports
- Composition of generated artifacts into data applications and dashboards
Those workflows are under active development and not yet fully productized, but the underlying architecture is now in place.
Learn more / try it out
- Blog post: https://blog.holoviz.org/posts/lumen_1.0/
- Documentation: https://lumen.holoviz.org
- GitHub: https://github.com/holoviz/lumen
As always, feedback and questions are very welcome.