The database that proves itself

Query the generator, not the data

Most columns in a real table are a function, not data — autoincrement ids, evenly-spaced timestamps, accumulating counters. AT-1 recovers that function and answers a range predicate AND its exact selectivity from a few coefficients, reading zero stored bytes: 30,000×–60,000× faster than a scan, byte-identical to DuckDB. It isn't a faster scan — it's a different physics of query. Patent-pending; see the zero-scan docs.

Query in place

For the columns that aren't generator-produced, run predicate + projection queries directly on the compressed file — zone-map block-skip reads a fraction of the data. No restore, no decompress-the-world. Measured 13× less I/O on real lakehouse data.

Erase one record

Per-record cryptographic erasure: remove a single data subject by key destruction and emit a signed certificate. Everything else stays exactly recoverable, queryable, and verifiable. GDPR/CCPA right-to-erasure without rewriting the archive.

Verify the answer

Every record carries a SHA-256 integrity trailer; every query can return an Ed25519-signed receipt binding the answer to the exact sealed dataset. A third party can verify provenance + integrity + non-repudiation — without re-running the query or seeing the raw rows.

Generative intelligence

Recall records by structure, forecast a series beyond its stored range, rank by anomaly, and complete a record from a partial cue — served from the same artifact you store. The database doesn't just hold the data; it holds what generated it.

Provenance-sealed AI data

Assemble a training corpus that proves its own composition — % real vs synthetic, per-record lineage — sealed so any value tamper or composition lie is detected. Turnkey EU-AI-Act training-data provenance.

Connect your app

Reach it over a socket: an HTTP SQL endpoint with Python and Node drivers, a DB-API 2.0 driver and a SQLAlchemy dialect so your ORM and BI tools connect through standard plumbing, auth, concurrency, and multi-tenant isolation. JDBC/ODBC and a Postgres-wire shim follow on the same protocol.

Ask in plain English

AI-native: an MCP server lets Claude, Cursor, or any agent drive the database in natural language — “which sensors look anomalous?”, “forecast turbine_3 for 50 steps” — translated to a verified query, with the answer signed and checkable. It never guesses; it asks when unsure.

Migrate in one command

Point it at your existing data — CSV, Parquet, NDJSON, or a Postgres dump — and it lands every record in AT-1DB with a mandatory lossless round-trip check. No re-modelling, no schema ceremony; new record shapes ingest later with no migration.

Deploy anywhere

Ship it your way: Docker image, docker-compose, a Helm chart, or a Terraform module — with backup/restore, point-in-time snapshots, and Prometheus metrics built in.

Per-TB-under-management, billed monthly — calibrated to a slice of the storage we save you. On bytes alone we're at parity-to-better with the best columnar formats; the difference is everything you get on top.

We capture value on two axes — the storage we save you and the read/egress I/O we save you. The graduated per-TB-under-management rate already embeds both, so you get one clean number; high-read workloads can instead meter on measured GB served. Our 13× less I/O is value we deliver and price on, fairly.

Verifiable query receipts, per-record erasure certificates, and AI-Act corpus attestations — for regulated buyers who need to prove what happened, not just store it.