Multimodal data sits in object storage. Visualization tools sit elsewhere. The gap is filled by manual exports that lose track of versions and break collaboration. DataChain closes the gap: Dataset DB records carry file pointers and rich annotation types, every modality has a type-aware viewer, and the central registry holds every version with full lineage.
DataChain is the Context Layer for Unstructured Data, and the viewer is where that layer becomes visible. Files stay in your buckets; the registry serves the typed view; teammates and agents read the same versioned source.
Type-aware viewers for every modality
Dataset DB stores file pointers, not bytes. The viewer reads bytes directly from your buckets and renders them in place: video frames, audio waveforms, 3D scans (CT, MRI, microscopy), time-series traces, bounding boxes, segmentation masks, and structured tabular data.
- View raw files inline. No export, no reformat, no second tool.
- Inspect annotations as overlays on the source media: boxes on images, masks on scans, segments on time series.
- Browse, compare, and reuse versions confidently from one registry.
One source of truth, versioned
The central Dataset DB is the registry. Every chain that runs deposits a typed dataset. Every dataset surfaces in the UI with its schema, lineage, and authorship.
- Browse by name, filter by schema, search by similarity, walk lineage backwards or forwards.
- Compare two versions side-by-side; localize what changed.
- Share a versioned dataset with a teammate as a pinned reference, not a stale CSV export.
Collaboration without exports
Teams that exchange CSVs lose track of versions and reformat files for every recipient. DataChain replaces both with a single registry of typed, versioned, lineage-traced datasets. Everyone reads from the same source. Updates propagate by appending versions, not by emailing files.
- Faster review cycles. Open a dataset version in the UI, see the chain that produced it, validate before building on it.
- Clear authorship and lineage on every record.
- Cross-team reuse. The team that produced the dataset is not a bottleneck for the team that consumes it.
Datasets become inventory. Inventory compounds.