Inspect HDF5 datasets and export bounded CSV or NPY slices
A bounded scientific data desk for local HDF5 inspection. Select a dataset and explicit axis ranges while retaining the metadata needed to interpret the extracted values.
HDF5 Slice Desk
The real example includes a compressed 3×4×5 big-endian cube, exact 64-bit IDs, Unicode strings, compound fields, dimension scales and deliberately unsupported links/types.
Supported scope: 32 MiB input; 512 hierarchy entries, depth12; rank8; 200,000 selected elements and 16 MiB raw slice/chunk; 45-second worker timeout. Primitive numeric/string and flat fixed-width compound extraction only. NPY is primitive numeric; PNG is <=16,384 values, at most 2,048 pixels per axis and two nonsingleton axes, excluding 64-bit integers. No external/soft-link following, VDS, optional plugins, semantic MAT/NetCDF conversion or file repair.
Scientific data file
Required fields are marked with an asterisk.
Dataset slice desk
Review your output, then save it to your device.
Make the next step clear.
Your findings, assumptions and next steps will appear here after you run the check.
HDF5 Slice Desk · https://hdf.kitforma.com/ · KitForma
Real HDF5 WebAssembly reader
Multidimensional hyperslab selection
Exact numeric NPY and metadata/loss reports
Understand the evidence.
Focused guides explain the rules, the edge cases and the checks worth doing before you act.
Choose an HDF5 slice with explicit axis ranges
Extract a bounded layer while keeping its coordinates and shape.
Read the guide ↗Field guide / 02Preserve numeric types when exporting HDF5 data
Separate exact binary arrays from convenient text tables and previews.
Read the guide ↗Field guide / 03Understand HDF5 links, filters and conversion losses
Make unsupported dependencies and omitted metadata visible.
Read the guide ↗Before you start
Does the file leave my browser?
No. The self-hosted HDF5 engine runs in a local worker. It opens a temporary in-memory copy read-only and releases it when the worker ends. External links, virtual source files and soft-link targets are not fetched.
How do slice ranges work?
Enter one comma-separated range per axis. A colon selects the full axis, 2 selects index 2, 1:8 selects indices 1 through 7, and 0:10:2 uses a step of two. Bounds must be nonnegative and nonempty. A selected single index remains a length-one axis in the exported NPY shape.
Can I preserve large integer IDs?
Yes. Primitive 64-bit integers remain exact in typed NPY and become exact decimal strings in JSON/CSV. Do not let a spreadsheet automatically convert those strings to floating-point numbers. The source dtype and byte order are retained in the report.
Why is a dataset metadata-only?
It may use references, nested or variable-length compound fields, array/enum/opaque types, variable-length arrays, virtual sources, unsupported filters, oversized chunks or an empty dataspace. The reason is shown. The tool does not substitute fill values for data it cannot safely retrieve.
Is PNG an exact representation of the values?
No. PNG is a min/max-normalized grayscale visualization for small numeric slices with at most two nonsingleton axes. Nonfinite values are transparent, and the report records the range. Use NPY or JSON for quantitative work; 64-bit integer slices do not get a PNG.
Will CSV or NPY retain attributes and dimension scales?
No. Those formats do not embed the complete HDF5 object model. The hierarchy manifest and slice report retain dataset metadata, filters and bounded attribute values or explicit omission reasons. Dimension-scale references may be metadata-only. Keep the original HDF5 file.
Sources & assumptions
Formats, algorithms and supported features are documented below. Check the stated scope and inspect your output before using it.
- h5wasm — NIST ↗
Unmodified h5wasm 0.10.3 provides HDF5 C-library access. NIST is acknowledged as the source; no endorsement is claimed.
- HDF5 documentation ↗
Groups, datatypes, attributes, links, filters and datasets are distinct HDF5 concepts.
- NumPy NPY format ↗
NPY records array shape, order, dtype and binary values.
- Engine notices ↗
Full NIST/HDF5 and dependency notices are shipped with the local engine.
Reviewed 2026-10-19. Read our methodology or report an issue.