Vértice Criativo
Verifiable AI · Self-hosted stack

Verifiable AI infrastructure

Don't trust. Recompute.

A stack for AI whose output the client reproduces and verifies independently — without trusting the operator and without calling its servers. Every action leaves a receipt anyone can re-derive from public data. This page covers what it is, how a client adopts it, the components, and indicative pricing.

Product & services briefRev. Sep 2026verticecriativo.pt

The product

Proof that travels with the output

Most "trustworthy AI" asks you to trust a dashboard. Vértice inverts it: every agent action emits a recomputable receipt — anchored on-chain, re-derivable by a third party from public data alone. The verifier never re-runs the model and never contacts us. Verification is separated from the machine that produced the result — which is what makes AI at scale auditable by people who will never own the infrastructure.

It is in production today, not a proposal: agents emitting receipts at recomputableagents.com, a public no-auth verify endpoint at gateway.verticecriativo.pt, and even the capability list itself is recomputable — hand our registry any entry and it re-derives it live (don't trust the list, recompute it). The open primitives are going through the Ethereum standards process. The same discipline recently caught real build errors in a third party's model release.

core
Layer 1

Technical proof

The recomputable receipt. Self-verifiable by anyone, no trust and no vendor call. This is the spine — always present.

Layer 2 · optional

Legal proof

An eIDAS qualified timestamp / seal composed over the receipt — proof a regulator, auditor or court accepts, which cryptography alone doesn't confer.

Layer 3 · optional

Post-quantum

Non-custodial dual-family (ML-DSA + SLH-DSA) key binding for agents and wallets (ERC-8373). Live today — 29 agents PQ-bound, mainnet-anchored. The key never leaves the client.

Application model

How a client adopts it

Engagements ladder up from a bounded proof-of-concept to a full regulated deployment. It runs on infrastructure the client owns (self-hosted) or hosted by Vértice — keys stay non-custodial either way. Each step is a real deliverable, not a demo.

01
Pilot / PoC
Recompute-receipt + on-chain anchoring bolted onto one existing workload. Proof it verifies, end to end, on the client's real data.
02
Full implementation
The whole stack tailored to the client's agents and identities — gateway, SIWE auth, recomputable receipts, anchoring, the public verify surface, admin.
03
Enterprise / regulated
Full stack plus the post-quantum binding, the eIDAS / legal layer, and compliance & audit alignment (EU AI Act) — with ongoing support.
04
Ongoing
The install is the moat, so it doesn't ship flat: a support retainer keeps proofs current and post-quantum-ready, with optional per-verified-output or per-seat licensing.

Product list

The components

Sold as a tailored implementation or as individual seams that drop onto an existing stack. The verifiable-AI core is below; the wider studio also delivers products (clock-in.pt), automation (n8n / MCP) and technical AV services.

Recompute Kit

Core

Recomputable receipts + on-chain anchoring. The primitive everything else builds on — re-derive any result from public data.

Production · open primitive

Attested AI agents

Core

Agents whose every action carries checkable provenance — EIP-712 signature, per-agent PQ companion signature, on-chain anchor.

Live · recomputableagents.com

PQ key binding

Security

Non-custodial post-quantum identity binding (ML-DSA + SLH-DSA) for agents and any wallet, independently recomputable via the ERC-8373 enforcer.

Live · mainnet KYA-L4 · 29 agents

Verification surfaces

Product

The public proof pages a client or their auditor uses to recompute a claim themselves — the trust made visible.

Live · /verify · /review · /quantum

Primitives registry

Product

The capability index that verifies itself — lists what's built, and re-derives any entry live (on-chain, endpoint or conformance). Drift fails a check; nothing on faith.

Live · trustless-ai/primitives

Commercial MCPs

Product

Specialized capability servers carried by an agent, gated on-chain and billable pay-per-use (A2A micro-payments).

Available

Identity gateway

Infra

CCIP-Read ENS names, subnames, records and IPFS — the identity substrate the receipts and bindings resolve against.

Live · gateway.verticecriativo.pt

Independent review gate

Assurance

A signed, Bitcoin-anchored review verdict on a release — recomputable against a published key by anyone, in partnership with invinoveritas.

Live · /review

Indicative pricing

What an engagement runs

Estimate — a scoping call sets the real number
TierScopeIndicative
01 Pilot / PoCRecompute + anchoring on one workload, on the client's existing stack.€15k–40k
02 Full implementationThe whole stack tailored to the client's agents & identities.€50k–150k
03 Enterprise / regulatedFull stack + PQ binding + eIDAS legal layer + compliance alignment + support.€150k–400k+

Day rate

€800–1,500

Time & materials, or to sanity-check a fixed price. A full build is 2–4 months.

Support retainer

€2k–8k / mo

Keeps proofs current and post-quantum-ready after the install.

Licensing

per output / seat

Optional recurring layer for hosted verification — worth more over time than a flat install.

  • →Scope — one workload versus the full stack moves the number most.
  • →Buyer — a regulated enterprise buys the outcome (court-admissible AI), priced accordingly.
  • →One-off vs recurring — the install is the moat; it carries a retainer, not a flat fee.
  • →Bespoke vs productized — fixed tiers make it repeatable and quotable in a sentence.

This is a genuinely rare skill set — verifiable compute, post-quantum crypto, on-chain anchoring and live standards work in one stack. Price the outcome, not the hours.