Skip to content

Guides

Ordered by how often you will need them. The first two cover most real use.

Your first folio Create a folio, add a mixed bag of objects, read them back by name, understand what landed on disk. Ten minutes. Start here.

Everyday patterns The moves that recur: one path per analysis, lineage, folio metadata, bulk reads, batched writes, tidying up, handing over a path.

Tables: big, external, and lazy The only item type where size forces a decision. Parquet and pandas by default; Polars lazy scans when a table is too big to load; reference_table() when the data is not yours.

Models joblib vs. skops, what each actually protects you from, and why a custom transformer must live in an importable module.

Sharing a folio Cloud paths, read-only opens, multiple readers, sync vs. fork, and what to send someone who will never install the package.

Snapshots Naming a state you may need to return to — how it works, what it does not cover, and when not to bother.

Reading a folio without DataFolio The on-disk format as a documented, public surface: the catalog schema, how to resolve payloads, and the concurrency rules.

What DataFolio is not Strengths, limits, sharp edges, and what to reach for instead. Worth reading before you build on it.

Migrating to 2.0 Method mapping and behavior changes from 1.x.

If you are in a hurry