Home

Your report's design. Your data's story. Generated.

Give pbigen any .pbix you like and a table. It reuses that report's design, reasons like a senior BI consultant — business context, objectives, KPI tree, research, storyboard, self-critique — and writes a ready-to-open Power BI project. Free, local, open source.

Get started Template packs The AI pipeline

any .pbix as the design · 5-stage AI story pipeline · 16 data sources · MIT

Install¶

pip install pbigen                 # core (dependency-light)
pip install "pbigen[bigquery]"     # + one source driver: bigquery, snowflake, lakehouse, …
pip install "pbigen[llm]"          # + the AI design pipeline (any LLM, hosted or local)

60 seconds, no cloud account¶

pip install "pbigen[lakehouse]" pyarrow
python -c "import pyarrow as pa,pyarrow.parquet as pq,random,datetime as d; r=random.Random(1); \
pq.write_table(pa.table({'order_date':[d.date(2024,1,1)+d.timedelta(days=r.randint(0,540)) for _ in range(500)],\
'region':[r.choice(['North','South','East','West']) for _ in range(500)],\
'status':[r.choice(['New','Shipped','Returned']) for _ in range(500)],\
'revenue':[round(r.uniform(50,5000),2) for _ in range(500)]}),'orders.parquet')"

pbigen generate --source parquet --set uri=orders.parquet \
  --objective "Sales overview by region and status over time" --theme midnight --out out

Open out/Orders/Orders.pbip in Power BI Desktop (enable the PBIR preview once — see Recipes → Open in Desktop).

What you can do¶

Connect anything BigQuery/BigLake/Omni, Snowflake, Redshift, Synapse/Fabric, Databricks, ClickHouse, Postgres, Athena, a Parquet/Iceberg/Delta lake on GCS/S3/ADLS, or a Cube semantic layer. Sources & auth →
Thinks like a BI consultant Business context → objectives & KPI tree → research (live web search) → storyboard → critique. Or a reproducible deterministic storyboard with no key at all. The AI pipeline →
Wears your design pbigen template build report.pbix captures canvas, background, sidebar, logo, title font, theme and per-visual formatting; every new dashboard lands inside it. Template packs →
A story, not a chart dump A question per page, a subtitle per visual, KPI cards with green/red period deltas, ranked bars, monthly trends, an About page with KPI definitions and a DESIGN.md with the reasoning.
Controls that matter Import vs DirectQuery, row_limit sampling for huge tables, and a clean Python API + CLI. Recipes →
Safe & portable Read-only, metadata-only, standards-based PBIP output that validates against Microsoft's schemas and opens in Desktop. Version-control it like code.
Test everything A self-serve verification harness — offline check, per-source, schema validation, and the full matrix. Testing →

Why not just use Copilot or an agentic BI tool?¶

Most AI dashboard tools are cloud services that build inside their own surface and cost per seat or per capacity. pbigen is a small open-source library that produces portable Power BI files you own, on your machine, for free — deterministically if you want, with no data leaving your environment. See the full breakdown: Why pbigen →