FastCodec

Compressed. Encrypted. Still answering.

The one lossless format that moves your sensor data, stores it small, and answers questions on it in place — without decompressing it, and without handing anyone the key.

2.5GB/s
Decode throughput, measured natively
~1µs
Per rank, range or quantile query
90%
Of channels answered exactly on ciphertext
100%
Lossless, bit for bit, every round-trip
The hidden tax on data

You shouldn’t have to unpack data to ask it a question.

Most stacks re-encode at every hop, decompress whole archives to answer one query, and decrypt before they can compute. You pay for each step in time, in compute, and in exposure.

The old way

Decompress. Scan. Repeat.

  • A new format at every hop.Edge, transport and archive each re-encode the same bytes.
  • Whole archives unpacked for one answer.A single percentile means decompressing everything, then scanning it.
  • Privacy bolted on from outside.To compute on encrypted data, someone has to hold the key.
With FastCodec

Compress once. Ask anything.

  • The same bytes, end to end.Compress at the source; everything downstream reads one container.
  • Answers from a tiny index.Queries read per-block summaries and touch a sliver of the data.
  • Privacy inside the format.Servers answer on ciphertext, holding no key and no payload.
One format, three jobs

Edge to cloud to query. The same bytes throughout.

01 · Transmit

Move less. Move faster.

Built for austere, low size-weight-and-power edge devices — and scales to hundreds of thousands of concurrent encoders. Lossless from the first byte.

2.5GB/s decode
02 · Store

Small, and never opaque.

zstd-class lossless ratios, with query and privacy built on top. Store more per dollar — and keep every byte usable without unpacking it.

100%lossless
03 · Compute

Answers, not archives.

Rank, range, threshold, quantile, search and region decode run against a tiny index instead of the whole payload.

~1µs per query
Compressed query

Answer without unpacking.

Rank, range and quantile read a tiny per-block summary — about a microsecond over hundreds of thousands of samples.

Explore compressed query
Zero-knowledge

Answers on encrypted data.

Encrypt the summaries and a server answers on ciphertext — holding no key and no payload.

Explore zero-knowledge
90%exact on ciphertext

Share of channels answered exactly, in test.

Where it earns its keep

Built for fleets that never stop recording.

Telemetry and video you can act on in place — on one engine.

CAN telemetry

Questions that used to be decompress-the-archive jobs become summary reads, answered in microseconds.

›every frame where node 0x18 faulted ›p95 wheel current across a 10-year archive

Raw camera video

Per-tile residual summaries drive motion-gating and region-of-interest decode, so you pull back only the pixels that matter.

›decode only the tiles where something moved
Performance

Measured, not marketed.

Native, in-repo benchmarks on the reference host. No asterisks.

2.5GB/s

Decode throughput. Encode up to 463 MB/s.

665×

Faster archive encode than the prior implementation.

8–11%

The entire ratio cost of making data queryable.

575+

Engine, certification and sealed-kit tests.

Your evaluation

Proof on your data, not ours.

Three steps from first conversation to results you can verify yourself.

  1. Tell us about your data.

    Telemetry, video, or both — fleet size, retention, the questions you wish you could ask.

  2. Run the sealed kit.

    Air-gapped, network-deny verified, deterministic. You run it on your own data, inside your own walls.

  3. Review the results together.

    Compression, query and zero-knowledge, end to end — with a report hash you can reproduce byte for byte.

Questions, answered

What engineering teams ask first.

Is it really lossless?

Yes — bit for bit. Every round-trip is verified, and the guarantee is guarded by generative property tests. All the bytes, in the proper order, every time.

How can you query data without decompressing it?

Every block in a FastCodec container carries a tiny summary — a zone map and a histogram. Rank, range, threshold and quantile queries are answered from those summaries alone, in about a microsecond over hundreds of thousands of samples. Content search and region-of-interest decode open a single block, roughly 2% of the data.

What does the server see in zero-knowledge mode?

Ciphertext. Your client holds the key and encrypts the summaries; the server answers rank, range and threshold queries without ever holding the key or the payload. In test, 90% of channels were answered exactly on ciphertext. For the rest, the server returns candidate blocks that your client decrypts.

What does queryability cost in compression?

Between 8% and 11% in ratio versus an opaque whole-stream frame. That is the entire cost — in exchange, the archive stops being something you unpack and becomes something you ask.

Does our data leave our network during the evaluation?

No. The evaluation runs in a sealed, air-gapped Docker kit with network-deny verified and a deterministic report hash. You run it on your own data, offline, and we walk through the results with you.

How does it fit into our stack?

Through a native, semantically versioned C API (fhc.h) — a stable link target — plus integration scaffolds for GStreamer, ROS 2 and v4l2 capture, so it slots into the pipeline you already run.

See it on your own data.

Bring a slice of telemetry or video. We’ll walk you through compression, query and zero-knowledge results, end to end.

Evaluated offline in a sealed, air-gapped kit. Your data never leaves your control.