The architecture

Six threads. One coherent stack.
No central authority.

VTI handles trust between people, AI agents, and communities — on cryptographic rails, with human-in-the-loop governance and regulatory alignment built into the architecture, not bolted on.

01

People & Orgs

Personhood and relationship credentials, wallets, and selective / zero-knowledge disclosure.

02

AI Agents

Scoped agent identity that acts on your behalf — with mandates it cannot exceed.

03

Communities

Verifiable Trust Communities write their own membership rules and own their trust graph.

04

Cryptographic Rails

did:webvh, DIDComm v2, the Trust Spanning Protocol, post-quantum and ZK cryptography.

05

Governance

Human-in-the-loop approval, least privilege, kill-switch, and tamper-evident audit.

06

Compliance

Architecturally aligned with eIDAS 2.0, Utah SEDI, and China’s 2026 agent rules.


Agent governance · DTTE

Show the exact plan. Get it approved. Run nothing else.

Delegated Trust-Task Execution keeps a human in the loop without breaking agent workflows. Every approval is bound to an exact payload digest — single-use, no replay, two-device separation enforced.

Requester

Proposes a task and payload. Holds zero authority. Never learns who approved, or when.

Executor · VTA

Dry-runs the real handler, produces human-readable effects, and enforces the boundary. The only party that sees everything.

Approver

Receives the effects and a 6-character match code. Signs a single-use decision. Never sees the payload.

// Anti-bait-and-switch: approval carries a fingerprint of the
// exact request. Any change and the vault refuses to execute.
digest = hmac(salt, length_prefix(type_uri) + payload_bytes)

// Anti-replay: each approval is single-use, consumed at execution.
grant.status = GrantStatus::Consumed;

// Two-device separation: the device that acts is never the one
// that approves.
assert!(requester_did != approver_did);

No counterpart

Seven capabilities with no EU equivalent.

eIDAS 2.0 has no agent trust, no community trust graph, and no DTTE. VTI does.

  1. 01
    Trust Tasks

    One signed envelope for 555 agent operations, transport-independent.

  2. 02
    DTTE

    Approval bound to an exact payload digest — single-use, two-device separation.

  3. 03
    Ceremony Tasks

    Operations that carry their own authority, preventing approval regress.

  4. 04
    Contexts

    Isolated BIP-32 key hierarchies, one per life domain.

  5. 05
    Agent-native VTA

    Something that acts — not just something that presents credentials.

  6. 06
    Community trust graph

    VRCs (Verifiable Relationship Credentials), cross-community recognition, witness evidence.

  7. 07
    Policy as configuration

    “This operation needs two approvers” is one config row.


Cryptography

Modern cryptography, in the stack.

  • Post-quantum — ML-DSA / SLH-DSA, experimental support
  • Zero-knowledge — BBS+ over BLS12-381 for unlinkable proofs
  • TEE — AWS Nitro secure enclaves for key custody
cargo add affinidi-tdk
use affinidi_tdk::{VTA, CredentialBuilder, TrustTask};

let vta = VTA::new()
    .with_context("work")            // isolated key hierarchy
    .with_enclave(EnclaveMode::Nitro) // TEE for key custody
    .build().await?;

let credential = CredentialBuilder::new()
    .subject(holder_did)
    .claim("age_over_18", true)
    .selective_disclosure()          // BBS+ ZKP
    .sign(&vta).await?;

Standards & compliance

Compliance is architecture, not a checkbox.

European Union

eIDAS 2.0 / EUDI

EUDI ARF-aligned (v2.x)

  • OID4VCI + OID4VP
  • SD-JWT VC (PID format)
  • ISO mdoc / mDL 18013-5
  • SIOPv2 + DCQL
  • Selective disclosure
  • W3C VC 2.0
United States · Utah

Utah SEDI

State-Endorsed Digital Identity · aligned

  • Self-owned identity (DID + VTA)
  • DIDComm & TSP messaging
  • Open standards
  • Individual key custody
China

Agent Identity Framework

May 2026 — architecture aligned

  • Unified agent identity (DIDs)
  • Risk-tiered governance
  • Secure cross-domain interaction
  • Audit-trail requirements

VTA capability map

Eight capability planes. All live.

In plain terms: one agent that holds its own keys and credentials, signs without ever exposing them, talks to other agents, and only acts on operations a human approved. The full technical map is one click away.

Show the full VTA capability map — eight planes, for builders
Plane Capabilities Status
A · Cryptographic Core BIP-32 key derivation, Ed25519 / X25519 / P-256, ML-DSA (PQC), a signing oracle that never exports keys, secrets vault with soft-delete. Live
B · Identity Full did:webvh lifecycle, 11 DID document templates, agent naming, serverless and server-managed publication. Live
C · Credentials Issuance with EdDSA-JCS-2022, SD-JWT VC, BBS-2023, ZKP and mdoc formats; selective disclosure; human-in-the-loop approval on inbound offers. Live
D · Access Control Challenge-response auth over TSP / DIDComm / REST, Data-Integrity Trust Tasks, 5 role classes, 11 capability types, Rego policy engine. Live
E · Agents & Devices Per-context agent memory (isolated KV stores), MCP bridge over stdio, device binding, push-wake channels. Live
F · Transport Preference TSP › DIDComm v2 › REST with no silent degradation, mediator drain windows. Early TSP production deployment (with CardInfoLink). Live
G · Operations Two-phase backup / recovery, audit logging at the dispatch spine, TEE attestation (Nitro, SEV-SNP), anti-rollback anchor MAC. Live
H · Interop OID4VCI, OID4VP, DCQL, SIOPv2, SD-JWT VC, ISO mdoc 18013-5/-7, BBS-2023, TSP, DIDComm v2. Live

Cookie Preferences

We use cookies to enhance your experience. You can manage your preferences below. For more information, read our Cookie Policy.

Strictly Necessary Always Active

These cookies are essential for core website functions such as security, session integrity, and cookie preference storage. They cannot be disabled.

  • _cf_bm: Distinguishes humans from bots (Cloudflare) · 30m
  • _cfuvid: Ensures secure browsing (Cloudflare) · Session
  • __hs_initial_opt_in: Prevents HubSpot's banner · 7 days
  • _gtm_debug: GTM debug mode (testing only) · Session
Analytics

These cookies help us understand how visitors interact with the site so we can improve content and performance. All data is aggregated and anonymous.

  • _ga, _gid, _gat: Google Analytics · Session – 2 years
  • __hstc, hubspotutk, __hssrc: HubSpot visitor tracking · 13 months
  • __hs_opt_out: HubSpot opt-out preference · 6 months
Marketing & Targeting

These cookies allow us and our partners to serve personalised ads and measure campaign performance.

  • _gcl_au, _gcl_dc: Google Ads conversion tracking · 90 days
  • IDE: Google Display Network personalisation · 1 year
  • _fbp: Meta / Facebook remarketing · 90 days
  • li_gc, _li_fat_id, bcookie: LinkedIn tracking · 1–24 months
  • guest_id, personalization_id: Twitter/X analytics · 2 years