Power BI practitioners
Analysts and BI developers who want to go from idea to a well-built project faster, without being an expert in every layer of the stack.
AtelierBID runs entirely inside your Claude Code session. Describe what you want — a model, a measure, a report, a theme — and it writes genuine Power BI Project files you open in Desktop. No API key. No backend. No per-call cost.
Producing a production-quality Power BI solution means crossing disciplines most practitioners only partly own: a clean star schema, safe relationships, correct and performant DAX, an accessible on-brand report, and a model Copilot can actually answer from. AtelierBID covers the whole offline build lifecycle, grounded so it binds to your real schema instead of inventing one.
Analysts and BI developers who want to go from idea to a well-built project faster, without being an expert in every layer of the stack.
Teams producing repeatable, consistent client deliverables across many projects, with a shared design system and best-practice library.
Upstream data engineers building pipelines or warehouses, or Power Platform (Canvas / Dataverse / Flow) builders — different stacks, not covered.
Run these inside a Claude Code session. Every run writes checkpoints to
output/<run-id>/, so an interrupted run resumes where it stopped.
/AtelierBID:design-report
8-agent pipeline → a self-contained HTML report-layout mockup with chart sketches and narrative titles, convertible straight into valid PBIR.
/AtelierBID:build-theme
Schema-valid Power BI theme.json, AI-assisted from a brand description,
with palette, font-colour and a design-quality critique.
/AtelierBID:model-star-schema
4-step Kimball modeller → SQL DDL, a Mermaid ERD, DAX measure stubs and a validation report scored against nine Kimball rules.
/AtelierBID:prototype-report
2-agent plan→build pipeline producing a clickable dashboard with real charts, KPI cards, dark mode and a responsive grid.
/AtelierBID:generate-background
A sanitised, on-brand SVG report-canvas backdrop — zoned, restrained, 80%+ negative space, brand colours only.
/AtelierBID:powerbi-design-system
Reference layer — the SCAN framework, five audience archetypes, six layout patterns and semantic colour roles. Auto-loads as context; creates nothing.
/AtelierBID:author-dax
One measure or a whole batch — from plain English, a spec file, or explicit DAX. Linted, validated, with a per-measure scoreboard so nothing fails silently.
/AtelierBID:optimise-dax
Ranks anti-patterns worst-first in plain English, then rewrites only what you accept — dry-run by default, with a before/after equivalence artefact.
/AtelierBID:update-metadata
Format strings, display folders, sort-by and hidden keys, plus the descriptions and synonyms Copilot and Q&A need to answer reliably.
/AtelierBID:document-model
A regenerable data dictionary straight from the TMDL, with undocumented elements flagged rather than invented. Read-only, so it is safe to run anywhere.
One PBIP container is written into at every step — never a dead-end preview. At any point it opens in Power BI Desktop. A cross-cutting spine (schema-grounding contract, best-practice rule library, offline/connected lifecycle with a drift guard) underpins every phase.
Scaffold a folder, conventions and an openable PBIP shell; seed the best-practice library and the schema-grounding spine.
Capture KPIs and grain, then author dialect-aware SQL and folding-aware Power Query/M — reviewable text, never executed.
Star-schema TMDL: fact and dimension tables, safe relationships, deliberate settings, partitions, incremental refresh, a ranked optimisation pass.
Idiomatic DAX measures, a consistent time-intelligence family, result-preserving rewrites, and EVALUATE validation queries.
Tokens → accessible theme → zoned background → a mockup IR that converts into valid PBIR, plus a shareable interactive prototype.
Metadata enrichment so Copilot and Q&A answer reliably, plus an offline Fabric data-agent config and provisioning runbook.
A deterministic, regenerable data dictionary straight from the TMDL, with gaps flagged rather than silently filled.
Every step reads and writes Power BI Project text — PBIP, TMDL, PBIR,
theme.json. No binary .pbix authoring, ever.
Generation binds to the schema-grounding contract. Anything off-schema surfaces as a proposed addition — never a silent fabrication.
Usable with zero auth, network or cost. Live read and validate are opt-in; publish and refresh always stay a deliberate manual step.
The 32 CLIs use no clock, no randomness and no network. Same input, same bytes out — which is what makes the conformance harness meaningful.
claude plugin marketplace add https://github.com/DATA-AI-XYZ/Atelier
claude plugin install AtelierBID@AtelierBID-Marketplace
Requires Node.js on your PATH — the deterministic scripts run via
node. The scripts are dependency-free (Node stdlib only), so there is no
npm install step.