Why pbigen — in a world full of AI & agentic BI¶

AI dashboard generation is crowded in 2026: Power BI Copilot & Agent Skills, agentic BI platforms like ThoughtSpot, Tableau Pulse, Sigma, Domo, Tellius, and general LLMs (ChatGPT, Claude) that can sketch charts. They're capable. So where does a small open-source library fit?

pbigen is the open, local, file-first option. It turns a table into portable Power BI files you own — generated on your machine, for free, deterministically if you want, with no data leaving your environment. It doesn't compete with a chat surface; it produces version-controllable artifacts that drop into the Power BI workflow your organization already runs.

The detailed comparison¶

	pbigen	Power BI Copilot / Agent Skills	Agentic BI platforms (ThoughtSpot · Sigma · Tableau Pulse · Domo · Tellius)	General LLM (ChatGPT / Claude)	Build by hand
Output is native Power BI you own (PBIP files)	✅	⚠️ built in the service	❌ their own BI surface	❌ snippets only	✅
Version-controlled, CI-friendly text (PBIR + TMDL)	✅	⚠️ not the generation flow	❌	❌	⚠️ only if you enable PBIR
Runs locally / in CI, no paid cloud capacity	✅	❌ needs Fabric capacity (F2+)	❌ SaaS subscription	⚠️ API/subscription	✅ Desktop
Works with no LLM / no API key (deterministic)	✅	❌ requires their AI	❌	❌	✅
Reproducible output (same inputs → same result)	✅	❌	❌	❌	⚠️ human-dependent
One interface across warehouses + lakehouse (Iceberg/Delta) + semantic layer, multi-cloud	✅	⚠️ Fabric / OneLake-centric	⚠️ varies by vendor	❌	⚠️ manual per connector
Metadata-only — no row data leaves your environment to design	✅	⚠️ cloud service	⚠️ SaaS	❌ you paste data	✅
Match a house style — theme + logo + nav, or clone a `.pbix` shell	✅	⚠️ manual	❌	❌	⚠️ manual
Open source (MIT), self-hostable, extensible	✅	❌	❌	❌	n/a
Cost	Free	Paid (Fabric capacity)	Paid (per-seat SaaS)	Usage-based	Free (Desktop)

⚠️ = partial or conditional; reflects each tool's common default in 2026, not every edge case.

What each alternative is great at — and where pbigen wins¶

Power BI Copilot / Agent Skills¶

Microsoft's in-product AI is powerful and native — it drafts pages, writes DAX, and (with Agent Skills) does end-to-end agentic authoring. But it lives inside Fabric, needs paid capacity, works in the service (not as version-controlled files you generate in CI), and is non-reproducible. pbigen wins when you want free, local, deterministic, git-tracked Power BI that any teammate can regenerate identically — and it happily coexists (generate the baseline with pbigen, refine with Copilot).

Agentic BI platforms (ThoughtSpot, Sigma, Tableau Pulse, Domo, Tellius)¶

Excellent conversational analytics — but they build dashboards in their own surface and bill per seat. If your organization standardizes on Power BI, their output doesn't live there. pbigen produces Power BI, on your terms, for free.

General LLMs (ChatGPT, Claude)¶

Great for a one-off chart or a DAX snippet, and you can paste data in. But they don't emit a complete, connected, refreshable Power BI project, they're not reproducible, and pasting rows means your data leaves your environment. pbigen sends only metadata to an optional LLM — and needs none at all by default.

Building by hand¶

The gold standard for a bespoke, pixel-perfect report — and the right tool for exactly cloning one existing report (Desktop's Save as .pbip). But it's slow and inconsistent at scale. pbigen does the mechanical 90% (connect, model, measure, chart-select, lay out, theme) in seconds, leaving you the 10% that needs judgement.

The moat, honestly¶

Correct PBIP/PBIR/TMDL emission is hard. pbigen's output validates against Microsoft's published schemas, every file, every run — so projects open in Desktop without repair.
A real design brain, not a prompt. Column-role classification, cardinality-driven chart/filter choices, geo/id exclusion, legend guards, safe DAX — reproducible and free, LLM optional.
Breadth behind one contract. 16 sources across every major cloud + lakehouse + a semantic layer, all through the same introspect → cardinality → connect interface.
Enterprise-safe by construction. Read-only, metadata-only, no capacity to buy, no egress.

When not to use pbigen (we'd rather be honest)¶

You need an exact clone of one existing report on the same data → use Power BI Desktop's File → Save as → .pbip (or pbi-tools). pbigen reuses a shell, it doesn't clone a specific report.
You want a live natural-language Q&A chat over your data → that's Copilot / ThoughtSpot territory, not a file generator.
You're not on Power BI at all → pbigen emits Power BI; a Looker/Tableau shop wants a different tool.

Everywhere else — "I have a table (or 200) and I want clean, on-brand, governed Power BI I can regenerate and version" — pbigen is built exactly for that.

Ready? Start with the recipes →