Aqta.
Receipts for AI.

The trust layer for AI that acts: every decision comes with a signed receipt anyone can verify offline, so you can deploy AI in regulated finance and healthcare.

Est. MMXXVI · Dublin, IrelandReceipt of Attestation

Acta · etymology

aq·ta
/'ak·tə/ · noun
1.
From Latin acta, plural of actum: things done, the public record. Roman senates inscribed acta diurna daily, public and dated.
2.
Since 2026. Cryptographic attestation: every AI action signed at runtime, verifiable without us.

verba volant, scripta manent (spoken words fly away, written words remain)

A signed receipt for every AI decision, wherever it runs: a direct API, your cloud, or an agent runtime

Est. MMXXVI · Dublin, IrelandReceipt of Attestation

Acta · etymology

aq·ta
/'ak·tə/ · noun
1.
From Latin acta, plural of actum: things done, the public record. Roman senates inscribed acta diurna daily, public and dated.
2.
Since 2026. Cryptographic attestation: every AI action signed at runtime, verifiable without us.

verba volant, scripta manent (spoken words fly away, written words remain)

AqtaBio · Public Ledger · W19

Eight days.
Public git.

AqtaBio publishes its spillover forecasts to a public git ledger before the outcome window closes. Here is the most recent proof.

This is the pattern across Aqta. Receipts you can verify without us, signed before the moment they matter.

Orange = the five highest-risk Ebola tiles in the W19 commit. DR Congo at rank 4.

Congo Basin · 6×5 tile lattice top-5 elevated baseline

9 May 2026

AqtaBio · W19 commit

Five Congo Basin ebola tiles, listed in an append-only public git ledger. Three in Central African Republic and Republic of the Congo, DR Congo at rank 4.

commitments/2026-W19.json

17 May 2026

WHO · PHEIC declared

Public Health Emergency of International Concern for Bundibugyo Ebola in DR Congo and Uganda. Biome-correct cluster; the genus we trained on.

WHO PHEIC · 17 May 2026

Bootstrap p = 0.222; we publish rank before the outcome window. The contribution is the primitive, not a claim of predictive capability.

Bounds + AqtaCore

How the full stack works

Sensitive data is removed on-device before it reaches the model, with an on-device record of what was stripped. AqtaCore then signs the governed call, so an auditor can verify that decision receipt offline and review the redaction record alongside it.

  1. 01The input

    A sensitive document

    Personal data you cannot send to a model as is.

  2. 02On-device

    Bounds strips PII and logs what it removed

    Redacted on-device, with an unsigned record of what was removed. PII never leaves your device.

    Redaction recordBounds
    redacted_at
    "2026-06-22T09:14Z"
    items_removed
    "4 PII fields"
    residual_risk
    "low"
    on_device
    true
    content
    "never leaves device"
    on-device audit log
  3. 03The gateway

    The clean document enters your LLM via AqtaCore

    One gateway in front of every model provider.

  4. 04The receipt

    AqtaCore signs the call and what the model decided

    A receipt per call: request hash, policy applied, model, outcome. Signed with Ed25519.

    Receipt B · decisionAqtaCore
    request_hash
    "a3f9c1b8..."
    policy_applied
    ["budget_guard", "pii_redaction"]
    model
    "claude-sonnet-4-6"
    outcome
    "ALLOWED"
    Ed25519 · SIGNEDkey gUoU hIvp...
  5. 05The audit

    An auditor verifies the signed receipt offline

    Receipt B verifies against one published key. The redaction record is reviewed alongside it. No call back to us.

What's shipped

Two products ready to deploy.

AqtaBio and Testera continue in Aqta Labs.

AqtaCore

Finance · Healthcare · Public sector

Open-spec AI receipts. Verify any call offline.

Every AI request signed at the source. A receipt your regulator can read without calling us.

  • Pre-execution policy, signed receipt per call
  • Hash-chained audit log, exportable for regulatory review
  • EU AI Act Articles 12 + 14 · DORA · GDPR · MiFID II
  • Zero-knowledge verifier live on the gateway: a demo proof verifies a policy held without exposing prompt content; per-request proving in preview
boundsPRO

Legal · Compliance · HR

HIPAA-safe document redaction on-device.

On-device contextual layer for clinical-narrative PHI. Documents never leave your machine.

  • Contextual health-data detection · entity recognition · OCR · faces · 40+ regex patterns
  • Ten UI locales · GDPR / HIPAA / FADP

Early access

Proof for every AI decision.

Get into AqtaCore early, plus one sharp email a month on where verifiable AI evidence is heading, from who gets a loan today to what an autonomous system decides tomorrow. No noise.

Your email, our Privacy Policy, unsubscribe any time.