Data Formulator 0.7: AI-powered data analytics for enterprise data
Microsoft Chenglong Wang, Scott Tsukamaki, Michel Galley, Jianfeng Gao
Microsoft dropped Data Formulator 0.7, an open-source tool that lets AI agents connect to enterprise databases and build charts without SQL. It's basically a no-code analyst that remembers your whole conversation instead of forgetting everything after one chat prompt.
Based on reporting by Microsoft, Chenglong Wang, Scott Tsukamaki, Michel Galley, Jianfeng Gao — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Enterprise data analysis has a dirty secret: most of the pain isn't the analysis, it's just getting the data into a usable shape in the first place. Analysts spend hours wiring up connections to warehouses, BI tools, object stores, and random local files before they can ask a single real question. Microsoft Research's latest release of Data Formulator, now at version 0.7, is aimed squarely at that bottleneck, and it does it by treating data connectivity as a first-class citizen rather than an afterthought bolted onto a chatbot.
The headline addition is something called Data Connectors, which lets teams set up governed, reusable links across databases, warehouses, BI systems, cloud storage, and local files once, instead of re-uploading files or re-authenticating every time someone wants to poke at the numbers. That sounds mundane, but it's the kind of plumbing that platform teams actually lose sleep over, and Microsoft is betting that solving it well is what separates a toy demo from something enterprises will actually deploy.
On top of that connectivity layer sits a set of context-aware agents that do more than answer one-off prompts. These agents can see the entire workspace — connected sources, loaded tables, earlier charts, the user's stated goal — and they act through tools: inspecting data, writing and running code in a sandboxed environment, generating chart specs, and asking clarifying questions when a request is vague. That's a meaningful departure from the typical chat-with-your-data pattern, where every question starts from zero and any nuance from three prompts ago is gone.
The interface ties it together through what Microsoft calls the Data Thread, a persistent, branching chat log that keeps track of every question, intermediate result, and chart across a session, paired with an interactive canvas for directly tweaking visualizations. Users can revisit old branches, compare alternative analyses side by side, and hand off polish work — labels, colors, layout — either by dragging things around or just describing the change in plain language. It's a workspace built for the messy, iterative way real analysis actually happens, rather than the tidy single-shot query most AI analytics demos show off.
Microsoft is releasing this as open source on GitHub, positioning it less as a finished product and more as a foundation other teams can bend to their own enterprise systems. Given how much of enterprise AI adoption stalls on data integration rather than model quality, that framing feels deliberate — and probably the right one.
My take — AI-written commentary, not fact-checked reporting
Everyone's obsessed with bigger models, but the actual thing blocking enterprise AI adoption has always been boring plumbing — auth, connectors, metadata, permissions — and Microsoft quietly admitting that by building Data Connectors first is more honest than half the 'agentic analytics' hype out there. I'll take a well-engineered open-source connector layer over another flashy chatbot demo any day; the demo always works, the pilot with real Snowflake permissions is where these things go to die.
Read more about this at: Microsoft
