Technology

How AIA actually works

Four pillars: multi-agent orchestration, a verifiable knowledge graph, a cryptographic audit layer, and a post-quantum agent-to-agent protocol.

Start with the runtime
Multi-agent runtime Verifiable knowledge Audit layer Post-quantum A2A
The whole system in one diagram

One request end to end

A request is routed, reasoned over, coordinated, and attested by a zero-knowledge proof — without data leaving the perimeter.

ENTERPRISE PERIMETER · YOUR NETWORK REQUEST user · system · agent t = 0 ms ROUTE L1 · ORCHESTRATION Trust-weighted routing · specialist agents REASON L2 · KNOWLEDGE DKG · graph retrieval COORDINATE L3 · A2A PROTOCOL PQ-encrypted state-sync ATTEST L4 · PQC + ZK SP1 · ZK proof L1 ANCHOR Ethereum ZKP INPUT · PLAINTEXT OUTPUT · ENCRYPTED + ATTESTED PUBLIC · VERIFIABLE
Five stages inside the perimeter. One proof leaves.
Pillar 01 · Orchestration

Specialists routed by outcome

Aegis splits each request into atomic subtasks and routes each to the best-suited specialist agent.

Routing ranks agents by a trust score that updates from outcomes; failed subtasks re-route automatically.

Pillar 02 · Knowledge

Provenance by construction

A Neo4j graph spans your documents, combining semantic similarity with provenance-bearing multi-hop traversal.

Answers trace back to the exact nodes that justified them. No "the model said so".

Pillar 03 · Audit layer

The audit trail as an artifact

An agent's work — task runs, model and tool calls, knowledge writes, payments — is designed to become a signed, hash-chained, Merkle-batched attestation record.

An auditor verifies a disclosure bundle offline, without the underlying data; anchoring is optional — a public chain, your own contract, or none. The record schema maps to OpenTelemetry GenAI, so other agent stacks attest the same way.

Pillar 04 · Post-quantum A2A

A forward-secret agent channel

A forward-secret, post-quantum agent-to-agent protocol with NIST Level 5 ML-KEM-1024 — built in, not bolted onto TLS.

Agents run under self-sovereign did:aia identities, with keys in a three-tier HSM-backed hierarchy.

Cryptography

Post-quantum at the perimeter
zero-knowledge across it

AIA's wire format was post-quantum from day one, its primitives NIST-standardised.

CRYPTOGRAPHIC FOUNDATIONS · PRIMITIVES INSIDE · PROOFS OUT YOUR PERIMETER · DATA STAYS IN ML-KEM-1024 FIPS 203 · key encapsulation ML-DSA-87 FIPS 204 · digital signatures SLH-DSA-256s FIPS 205 · stateless hash-based ChaCha20-Poly1305 AEAD · agent-to-agent transport SP1 PROVER PLONK · FRI · Goldilocks field no trusted setup recursive batch aggregation DID-BOUND IDENTITY did:aia:pqc:* · per-agent Double Ratchet · forward secrecy ZKP settles REGULATOR · L1 Ethereum and EVM chains STABLECOIN RAIL MiCA · atomic settlement PUBLIC CHAIN anchor of last verification zero raw data crosses the perimeter · only signed and zero-knowledge proofs
The primitive stack stays inside. Only the zero-knowledge proof and the settlement attestation leave.
Post-quantum primitive set

Transport and signatures rest on four NIST-standardised primitives, keyed by ML-KEM-1024 (FIPS 203).

Hybrid mode pairs post-quantum with classical primitives, so a flaw in either keeps the channel secure.

Zero-knowledge layer

SP1 — a STARK prover wrapped to a succinct Groth16 proof for cheap L1 verification. Leaf proofs aggregate recursively into one batch proof on an Ethereum-compatible chain.

Circuit family and L1 verifier contract in the threat model below.

Why this matters now

Traffic captured today is stored to decrypt once a quantum computer arrives. With AIA, migration is configuration, not redesign.

Threat T1 · harvest-now, decrypt-later · see threat model
Post-quantum migration clock

The clock started in 2022

US NSM-10 and Germany's BSI TR-02102 mandate post-quantum migration toward 2035. AIA was already there in 2024.

May 2022
NSM-10 · policy issued
Federal civilian agencies must inventory and migrate vulnerable cryptography to NIST-standardised PQC.
Federal contracting · OMB oversight
Aug 2024
FIPS 203 · 204 · 205
NIST publishes three post-quantum standards — the reference primitives for federal systems.
Reference standard · CNSA 2.0 alignment
Jan 2026
BSI TR-02102 · 2026-01 revision
German federal crypto guidance updates its PQC recommendations for regulated industries.
Procurement criterion · BSI-aligned tenders
2 Dec 2027
EU AI Act · Annex III high-risk obligations
Art. 12 logging, human oversight, and conformity assessment apply under Reg. (EU) 2024/1689, as re-timed by the Digital Omnibus.
Up to €35M or 7% · penalty regime of the Act
31 Dec 2030
BSI TR-02102 · quantum-safe deadline
BSI and the EU PQC roadmap require quantum-safe cryptography for high-protection systems by end of 2030.
Procurement criterion · high-protection systems
By 2035
Federal PQC · target horizon
US federal systems on PQC; RSA and ECDSA become audit findings.
Latent risk on today's traffic

Dates from NIST CSRC, BSI TR-02102, and EUR-Lex 2024/1689.

Threat model

What we defend against

Four threats a regulated deployment has to answer — and the mechanism that answers each.

T1
Harvest-now, decrypt-later
All wire traffic uses ML-KEM-1024, hybrid-paired with X25519 so a flaw in either leaves the channel secure.
T2
Compelled disclosure
Plaintext, keys, and model state stay inside your perimeter — the vendor has nothing to surrender.
T3
Agent compromise & exfiltration
A2A messages are authenticated under ML-DSA-87 with a forward-secret session layer — no replay, no forgery.
T4
Unverifiable AI claims
State transitions are designed to emit zero-knowledge proofs — verifiable without revealing what they ran on.
Open standards

Aligned with the bodies
that wrote the rules

We don't write our own cryptography — we implement the standards bodies that already have.

  • NIST FIPS 203 — ML-KEM (key encapsulation)
  • NIST FIPS 204 — ML-DSA (digital signatures)
  • NIST FIPS 205 — SLH-DSA (stateless hash signatures)
  • NIST SP 800-208 — stateful hash-based signature schemes
  • BSI TR-02102 — German federal PQC migration guidance
  • W3C DID Core 1.0 — decentralised identifiers
  • W3C Verifiable Credentials 2.0 — attested claims
  • IETF RFC 8439 — ChaCha20-Poly1305 AEAD
  • IETF RFC 7748 — X25519 (hybrid mode)
  • IETF RFC 9116 — security.txt
Interoperability

The standards your stack
already speaks

At every boundary, AIA speaks the standard your architects already use. We differentiate on what we prove and where it runs — never on proprietary wire formats.

BoundaryStandard
Agent & tool accessFull MCP server — AIA's agents, knowledge, and tools usable from any MCP client.
Your existing agentsAttestation records map to OpenTelemetry GenAI conventions — designed to attest agent runs from LangGraph, Microsoft Agent Framework, or any OTel-instrumented stack, not only AIA's own.
Paymentsx402 rail with any MiCA-regulated, ERC-20 stablecoin — EURC, USDC, and others; AP2-compatible by design.
AnchoringEAS on EVM chains, a plain self-hosted contract, or no anchor at all — your choice per tier.
Identity & disclosureW3C DID + Verifiable Credentials-compatible disclosure bundles.
CryptographyNIST FIPS 203/204/205, BSI TR-02102-listed primitives, audited liboqs.
DeploymentKubernetes, Helm, Terraform; OpenAPI-documented REST — any cloud or air-gapped, no lock-in.

Put the runtime
on your stack

A thirty-minute technical session — your workload against the runtime, and the path to production.

Book a call Write to us

The source is Apache 2.0 and AGPL 3.0 — public when we exit the design-partner phase, under NDA on request until then.