The technology

A compressed container that’s also a data structure.

Every block carries a tiny summary. Those summaries are what make FastCodec queryable — and, encrypted, what make it private.

The compressed domain

Answer the question without unpacking the answer.

100%
OF DATA TOUCHED

The old way — decompress the whole archive, then scan.

Keep scrolling to compare ↓

~1µs

Rank, range, threshold and quantile — answered from summaries over ~800,000 samples.

~2%

Content search and region-of-interest decode touch a single block of the data.

8–11%

The entire cost of queryability in ratio, versus an opaque whole-stream frame.

Zero-knowledge queries

Answers on encrypted data. No key required.

Encrypt the summaries and a server answers rank, range and threshold on ciphertext — while holding no key and no payload. Privacy isn’t a wrapper around the data. It’s inside the format.

Your client

Holds the key. Encrypts the summaries before anything leaves.

The server

Answers the query. Never sees the key or the data.

90%

of channels answered encrypted rank exactly, entirely on ciphertext in test. The rest return candidate blocks your client decrypts.

Two ciphers, one interface.

Order-preserving for speed, or order-revealing for minimal leakage. The privacy-performance trade-off is a swap, not a rewrite.

How it works

One structure does every job.

The stream is cut into blocks. Each block is transformed, entropy-coded, and given a tiny summary — the basis of both query and privacy.

01 · Intake

Sensor stream

Block segmentationstream → blocks
02 · Per block

Encode + summarize

Adaptive transformraw · delta-zigzag
Entropy stagezstd · order-1 range coder
Summaryzone map + histogram
03 · Surfaces

What you get

Compressed-domain queryrank · range · quantile from summaries; find / ROI from one block
Zero-knowledgeencrypted summaries; the server answers on ciphertext

Video ARCHIVE mode adds a native bijective BWT before the entropy stage — up to 665× faster than the prior implementation.

Measured, native, in-repo

The numbers, without the asterisks.

2.5GB/s

Decode throughput. Encode up to 463 MB/s at 8 threads.

665×

Faster archive encode with a native SA-IS transform. End-to-end frame 31×.

~1µs

Rank, range and quantile, answered from summaries — touching ~0% of blocks.

90%

Zero-knowledge rank answered exactly on ciphertext, with no key.

4–9%

Smaller than zlib from the native order-1 entropy stage. Dependency-free.

100%

Lossless round-trip. Bit for bit, guarded by generative property tests.

Measured on the reference host. Query and privacy behavior is structural — it depends on data shape, not magnitude.

Deploy it your way

Fits the pipeline you already run.

Native C API

fhc.h, semantically versioned — a stable link target for your integration.

Sealed evaluation kit

Air-gapped Docker, network-deny verified, deterministic report hash. Evaluate on your own data, offline.

Integration scaffolds

A GStreamer element, ROS 2 nodes and v4l2 capture to slot into an existing pipeline.

Put it to the test.

Run FastCodec against your own telemetry or video in the sealed kit — offline, deterministic, and yours to verify.

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