What the Statisticians Workbench is, who it's for, and the core concepts behind it
The Statisticians Workbench is a self-hosted web application that helps researchers perform statistical analysis on their data. It brings together database management, AI assistance, and data visualisation in one place — so you don't need to be a database expert, a statistician, or a software developer to get answers from your data.
The project originated from a specific problem: applying AI capabilities to analyse patient data held in an MS Access database, with the constraints of zero change to the existing application and complete data privacy. From those constraints grew a solution built around locally-operated AI working on sanitised copies of live data — and from there, the broader Workbench platform.
There is a historical parallel. Around 1300, the invention of reading glasses solved a biological limit — presbyopia, the hardening of the eye's lens around age 40, which made fine-detail work impossible. Before spectacles, a skilled artisan's most productive years were cut short. Reading glasses turned a natural decline into continued contribution.
In a related way, the explosion of specialist disciplines means experienced professionals can become less capable due to advances outside their area. AI used carefully acts like "glasses for the mind" — experienced workers remain productive despite needing skills they cannot have time to master. The Workbench is built on this principle: AI as a guide, not a replacement.
The defining architectural constraint of the Workbench is that no data leaves the user's machine. Every component operates on local data. The AI engine (Ollama) can run locally. Database connections are direct. File processing happens on your hardware.
This local-first design is not optional — it rules out cloud-dependent alternatives. It is essential for healthcare data, government services, and any situation where data privacy is paramount. Cloud AI providers are supported as an opt-in choice for users who prefer them, but the local path is always available.
The Workbench is available in four editions. The base edition is open source. The other three are binary builds targeting less technical users who want an easy install:
Recipes are the heart of the Workbench. A recipe is a reusable, repeatable analysis workflow — a pipeline of stages (filter, join, aggregate, pivot, impute, and more) that transforms raw data into results. Recipes produce datasets that can be charted, exported, or opened in Jupyter notebooks.
AI assistance is designed as a guide, not a code generator. The AI sees your actual schema and data, suggests which stages to use and how to configure them, and records every suggestion alongside your decisions for a full audit trail. You can use cloud AI providers or run a local LLM.
Multi-database support means you connect to PostgreSQL, MySQL/MariaDB, SQL Server, and MS Access — all from a single interface. A universal SQL execution engine compiles operations for each backend, with automatic fallback to pandas when a backend lacks native support.