Everything AT-1 does — in one place.

Byte-exact reconstruction with a SHA-256 gate and a non-inferiority guarantee — never larger than gzip/zstd/xz.

Run SQL predicates + projections straight against the compressed file — decode only the blocks a query touches.

Typed columnar codec (qcolumnar / qjson) with zone-maps and pushdown for tables, CSV, JSON and logs.

The compression-native, self-describing typed format — declarable structure compiles to near-zero bytes.

Auto-discovers columns that are functions of others (Total = A + B) and stores the rule, not the column — byte-exact.

Compresses order-free data (graph edges, posting lists, port/ID sets) by omitting the arbitrary order — losslessly.

Picks the best transform per stream and is provably never-worse than the strongest general coder.

Auditable error-bounded lossy with a signed certificate — when exactness isn't required but accountability is.

Stores the generating rule, not the output — large wins on rule-governed, synthetic and procedural data.

Store the recipe (generator + seed + params), not the trace — regenerate byte-exact at decode. ~10^5× vs xz on seeded sims/synthetic datasets, never-worse fallback.

Lossless, verifiable video container with addressable, queryable frames.

Crypto-erase one subject's record on demand with a signed certificate — GDPR / POPIA right-to-deletion.

Embedded integrity the decoder re-checks on every read; pinpoints exactly what changed.

The verified-lossless generative database: query in place, erase per-record, verify every answer, plus generative recall / forecast / anomaly — 30% smaller than Parquet, 13× less query I/O.

A drop-in proxy that strips repeated content from LLM prompts before they're sent — fewer tokens, identical answers. Up to 62% on repeated chat context, ~5.5× on PDFs.

Compressed + queryable + tamper-evident + per-subject erasable in one bundle — the flagship compliance archive.

Verified append-only event store: hash-chained frames you can append, query, aggregate, verify and tamper-pinpoint.

AI-Act record-keeping: a verifiable, queryable archive of model evidence and decisions.

A portable, tamper-evident capsule of an AI interaction — reproducible and addressable.

Label-free machine-health screening straight from the compressor's residual — CMMS alerts included.

A validated market-stress index derived continuously from compressibility.

A grounded generative knowledge base packed into a verified, addressable container.

Run any model from one self-describing, SHA-256-verified container — pack / run / serve, every weight integrity-checked on load. OpenAI-compatible.

A compressed embedding store that measures and stamps a recall@k certificate — 4× smaller hot index at 99.4% recall, or a certified-100% fp32 re-rank tier. Re-checkable with `verify`.

Pack many same-base specialists into one SHA-256-verified container, route with a signed router, and stream ONLY the routed expert — byte-identical, working set of one expert not N. Add/remove and it still verifies.

Reference-delta storage for fleets of model checkpoints — store one base + verified deltas.

Stream byte-exact inference from a compressed model — addressable, verified weights.

Certifies randomness / fairness (lotteries, PRNGs) and flags weak generators — calibrated against a null.

An isolated, verifiable workspace for sensitive data.

Verified avoided-emissions certificate over a tamper-evident audit trail.

23% smaller than Parquet-Zstd on real analytics tables, still queryable in place and byte-exact.

Tick and time-series data: smaller, queryable, erasable — real Binance-data proof.

Sparse-genotype VCF codec that beats general compressors, byte-exact and queryable.

Server/app/network telemetry retention: smaller, queryable, tamper-evident — NetFlow/Zeek proven.

Utility interval-data retention with per-meter erasure — real UCI-data proof.

Contact and record databases: xz-class size plus query-in-place and per-person erasure.

Head-to-head vs gzip, zstd, xz and Parquet on real public datasets — wins and ties, stated honestly.

The open research program: what works, what we killed, and why — radical-honesty by default.

Upload a sample, get an honest read on how much AT-1 can save on your data.

Model your storage, read/egress and compliance savings.

See the capabilities in action.

Full docs: CLI, SDKs, format, query, erasure, engines and more.

Compress and query a file right now, no install.

Value-based, transparent — 100 GB free, no card.

CLI, desktop app, SDKs and the free open decoder.

Resell AT-1 as a managed line — referral or white-label.