Technology

Compression, rebuilt from first principles.

How we think about the scientific raster pipeline, and how we ship the result.

Philosophy

Most codecs were designed for hardware that no longer exists.

Generic compressors treat scientific rasters like text files. Image codecs are designed for photographs and squeak by on telemetry. Even the scientific codecs that do exist were designed in an era when "compression performance" meant single-threaded scalar throughput on a 32-bit CPU.

We rebuild from the bottom up. Predictive coding driven by per-region adaptive selection. Tiled execution that fits L1 cache. SIMD as the foundation, not an optimization. Bit-exact guarantees across every backend. The result is a shared engineering substrate that products plug into — each occupying its own strategic position.

Products

Ten codecs. One suite.

RIPT
Coming Soon Patent Pending

A new codec

When you control both the writer and the reader, RIPT delivers better compression on prediction-friendly data and dramatic gains in lossy modes. New format. New decompressor. Higher ceiling.

BVC
Coming Soon Patent Pending

A new codec — volumetric

One native codec for the two formats 3D actually ships in — lidar point clouds and Gaussian splats. Lossless files smaller than the incumbents' lossy ones, decoded at GPU speed everywhere. New format. Higher ceiling.

The eight drop-ins — keep your format, swap the backend

TurboLERC Esri LERC

SIMD LERC2, wire-identical; faster encode/decode, no migration.

TurboWebP libwebp 1.6.0

Memory-safe WebP (CVE-2023-4863-class bugs gone).

TurboJXL libjxl 0.11.2

Tiled-geospatial JPEG XL: bit-exact, smaller than the cjxl -e1 fast preset, memory-safe; ~42–71× faster per-core encode than the -e7 effort that matches its size.

TurboLZW libtiff LZW

TIFF LZW; up to 10× via the parallel batch API.

TurboLZ4 liblz4 (C LZ4)

LZ4 block + frame; 2.05× decode, 1.49× encode, ≈equal size.

TurboZstd libzstd (RFC 8478)

Zstandard; faster encode L5+, smaller at L1/L9.

TurboLEPCC Esri lepcc

Point-cloud LEPCC; 1.75–1.89× encode/decode.

TurboSPZ Niantic libspz v4

3D Gaussian Splats; 4–7× encode, decode wins every mode vs libspz.

RIPT raises the ceiling for new deployments; the eight Track B drop-ins remove the floor for existing pipelines — keep your format, swap the backend. One vendor, every imagery format. See the suite overview.

Engineering Substrate

One dispatch layer.
Every modern vector ISA.

Both products share the same SIMD dispatch architecture. The codec algorithm is defined once at the algorithmic level and mapped to the widest available vector unit at runtime. Adding a new backend is a mapping exercise, not a rewrite. Output is bit-exact identical regardless of which backend ran.

Platform matrix

NEON
ARM64
128-bit
AVX2
x86_64
256-bit
WASM SIMD
browser
128-bit
In-browser decode at 60–107% of native AVX2 speed
AVX-512 Q3 2026
x86_64
512-bit
Sapphire Rapids / Genoa / Zen 4+ — benchmarks Q3 2026
SVE2 Q3 2026
ARM64
scalable
Graviton4 / Grace / Cobalt 100 — benchmarks Q3 2026
Scalar
Any
fallback

All backends produce bit-exact identical output. Auto-detection at runtime; portable scalar fallback for any CPU.

How We Ship

Compiled binaries.
Developer Hub.

Free decoders

Anyone who needs to read RIPT or LERC payloads pulls the decoder from the Developer Hub at no cost. Compiled binaries for every supported platform.

Paid encoders

Production write paths require a paid license. Per-seat, per-transaction, OEM, and enterprise tiers — see the pricing page for the full set.

No source distribution

Even paid licensees receive compiled binaries, not source. Compiled-only is part of our IP defensibility — patents pending, plus the algorithm never ships in readable form.