Steward
Your data, six views, one dataset.
Steward is licensed under Apache 2.0 and the public release is in progress. The repository will be opened to the community once the first stable cut is tagged.
Overview
Steward started from a frustration that every team I have worked with shares. The same dataset gets retyped into a slide deck, copy-pasted into a spreadsheet, redrawn in a diagramming tool, and then redrawn again as a Gantt chart. Each artifact drifts from the others within a week. The dataset is real, but the views are scattered, and the labour of keeping them in sync is invisible until it breaks.
Steward treats the dataset as the canonical thing and the views as windows onto it. Add a node in the graph and the table updates, the slides update, the Gantt updates, and the Mermaid diagram updates. Pull in a CSV and Steward infers the structure, lays out the graph, and generates the first cut of every other view. The AI co-pilot operates on top of this canonical model, which means it can answer questions about your data that are grounded in its structure rather than guessed at from a prompt.
The project is built in TypeScript, runs in the browser, and is designed to be self-hosted with no required external services. It is licensed under Apache 2.0 and the public repository will open as soon as the first stable cut lands. Until then, this page is the placeholder.
Highlights
- Six views over the same canonical dataset: graph, table, slides, Gantt, Mermaid, and YAML. Edits in any view propagate to the others.
- AI co-pilot grounded in the dataset, not just the prompt. It understands relationships, hierarchy, and time the way the data does.
- Three thousand six hundred plus tests guarding the core. The dataset is the source of truth and the test surface protects it.
- Open source under Apache 2.0. Self-hosted by default. No keys required to run the base experience.