Regulated entities; banks, trust authorities, sovereign wealth funds; now face three simultaneous pressures: AI-driven decisions that must be auditable under the EU AI Act, cryptographic foundations that must survive the post-quantum transition, and compliance infrastructure that cannot depend on intermediaries whose interests diverge from yours. Vauban is built to answer all three: private by default, post-quantum proof systems, no intermediary in the settlement path.
The architecture is chain-agnostic by invariant; deployable across public blockchains, permissioned networks, or conventional web infrastructure; with no vendor lock-in by design. Current proof system: ZK-STARK, hash-based, post-quantum-secure, no trusted setup, Starknet-anchored.
Verifiable trust, composable across roots and chains. The post-quantum claim algebra behind sovereign AI memory, proof of humanity, and auditable agents.
Why Vauban
Proof of humanity, institutional DeFi settlement, and verifiable AI memory run on Starknet mainnet today. The compliance, sovereignty, and AI-accountability requirements that regulated entities will face in 2030 are already knowable; Vauban is built backwards from those requirements. We verify. We prove. We show our work at cryptographic precision.
Most production zero-knowledge systems today rely on elliptic-curve assumptions that a cryptographically-relevant quantum computer breaks. Vauban's proof stack uses ZK-STARKs: hash-based, post-quantum-secure by construction, with no trusted setup and no ceremony risk. Compliance attestations issued today remain cryptographically verifiable in 2035, regardless of advances in quantum hardware. Your audit trail does not deprecate.
When an AI agent makes a recommendation, executes a transaction, or produces a compliance assessment under EU AI Act Article 12, regulators may require an audit trail that cannot be retroactively altered. Brain Protocol anchors every AI decision chain with a Poseidon-hashed Merkle root, published on-chain. Every knowledge entry consulted, every reasoning step, every output is replayable and independently verifiable; not by policy, by cryptographic proof.
EUDI Wallet and eIDAS 2.0 require identity assurance at Substantial and High levels; without centralising biometric data under a single custodian. Glacis Protocol delivers verifiable personhood via STARK proofs of ICAO CSCA passport data: the verifier learns the minimum required by compliance (age range, uniqueness) and nothing more. Personal data never touches the chain. V2 trustless live on Starknet mainnet; passports from 134 countries supported (ICAO CSCA registries): Schengen core plus UK, Turkey, Ukraine, the Balkans and micro-states, plus USA, India, China, Brazil, Japan, Australia, South Korea, Malaysia, Russia, and more. Designed for regulated onboarding, sovereignty-grade KYC, and EU Digital Identity Wallet integration.
Institutional participation in decentralised finance has been structurally blocked by one unsolved problem: compliance requires knowing your counterparty, but DeFi mechanics assume pseudonymous participants. Bastion resolves this through verified transaction claims; STARK-backed attestations satisfying AML and MiCA requirements without routing assets through a custodian. Vault management and solver execution run directly on Starknet mainnet, with every action cryptographically receipted.
Venture capital introduces a structural conflict in sovereignty infrastructure: the pressure to grow user numbers quickly degrades the security posture and institutional credibility that make a system worth trusting. Vauban is bootstrapped by design, and governance decisions are logged publicly.
A proprietary standard for cryptographic claims is a vendor dependency, not a trust anchor. The Vauban x402 STARK receipt format is documented in an IETF Internet-Draft (draft-vauban-x402-consolidated-00, Independent Submission stream); a companion draft for the Vauban Claim Algebra specification is in preparation, following the precedent of RFC 9449 (DPoP) and the W3C Verifiable Credentials framework. Every published SDK tarball is supply-chain-signed via Sigstore, with verification bundles recorded in the public Rekor transparency log. You can verify what you depend on.
Four interdependent layers, each independently sovereign. Each can be audited, operated, or replaced without disrupting the others. The stack is a consequence of the claim-algebraic design; not an integration: each layer produces and consumes cryptographically verifiable claims in a shared grammar.
Brain Protocol; verifiable cognitive infrastructure, open agent standards, cryptographic provenance on every decision
AI-augmented decisions cannot be trusted in regulated contexts unless every step of the reasoning chain is provable. Brain Protocol provides institutional-grade memory for AI systems: every knowledge retrieval, every decision, every pattern consulted is Poseidon-hashed, Merkle-anchored on-chain, and replayable. Designed for environments where hallucination is not merely a quality problem; it is a record-keeping liability under EU AI Act Article 12.
Command Center; 5 AI agents, every action signed + audit-logged + on-chain anchored
Autonomous agents that act without cryptographic accountability are undeployable in regulated environments. Command Center answers three questions that conventional agent observability tools do not pose: what did each action cost in outcome terms, not token terms? Who approved it? Can you prove it? Every agent action produces a RunProofCertificate; a signed, on-chain-anchored attestation of the full decision-execution chain, independently verifiable by auditors and regulators.
Vauban Claim Algebra (VPSF); composable claim grammar above any settlement layer, Starknet-anchored today
Cryptographic compliance claims from different systems; identity attestations, financial transaction proofs, AI decision receipts; cannot be composed without a shared grammar. The Vauban Claim Algebra provides that grammar: chain-agnostic by invariant, with no blockchain-specific types in the core specification. Claims compose algebraically across products and settlement layers. Current reference implementation anchored on Starknet via ZK-STARK proofs.
Bastion (DeFi mainnet), Glacis (post-quantum PoH; passports from 134 countries), Vauban Finance (wealth tech), and more
The claim algebra is not theoretical. Bastion runs institutional DeFi vault management on Starknet mainnet today; verified transaction claims, ZK-anchored settlement, STARK proof of solver execution. Glacis delivers ICAO-rooted identity attestations for EUDI Wallet integration; V2 trustless live; passports from 134 countries supported (ICAO CSCA registries). Vauban Finance produces AI Act-compliant audit trails for wealth management recommendations. These are running systems, not proofs of concept.
Vauban Agent SDK
The open-source toolkit for institutional AI agent development. Every action signed. Every memory cryptographically anchored. Every decision audit-replayable.
Agent frameworks built for developer productivity treat audit as an afterthought; observability dashboards, token counts, latency traces. None of this satisfies a regulator asking 'prove to me that this AI recommendation was grounded in verified data and approved by an accountable human.'
The Vauban Agent SDK makes cryptographic accountability the default, not the exception. Every run produces a RunProofCertificate: a signed, Poseidon-hashed decision chain anchored on-chain, independently verifiable without access to Vauban infrastructure. HITL (Human-in-the-Loop) approvals are themselves cryptographically signed, so the approval chain is as auditable as the action chain. Brain Protocol integration means every knowledge entry consulted by an agent carries its own provenance; no hallucination can silently enter the decision chain without leaving a cryptographic trace.
Supply chain: every published SDK tarball is Sigstore-signed with a Rekor transparency log entry. You can verify the build provenance, the SBOM, and the OIDC identity of the signing workflow before deploying to a regulated environment.
Key capabilities
Each Vauban product is an independently sovereign layer; its own cryptographic boundary, its own audit trail, its own deployment footprint. For institutional adopters, this means you can integrate a single capability (AI memory provenance, identity attestation, DeFi compliance) without committing to the full stack, then extend as requirements evolve. Every product produces and consumes claims in the shared Vauban Claim Algebra grammar.
Platform infrastructure
Verifiable Cognitive Infrastructure for AI agents
Institutional memory for AI systems. Every knowledge retrieval, decision chain, and contradiction resolution is Poseidon-hashed, Merkle-anchored, and audit-replayable. Open agent standards. Designed for environments where hallucination is an EU AI Act liability.
Verify: MerkleAnchor contract ↗
Post-quantum Proof of Humanity; passports from 134 countries
ICAO CSCA-rooted identity attestations backed by STARK proofs. V2 trustless live on Starknet mainnet. Passports from 134 countries with registered CSCA: Schengen core plus UK, Turkey, Ukraine, the Balkans and micro-states; plus USA, India, China, Japan, Brazil, Australia, South Korea, Russia, Malaysia, and major emerging markets. Designed for eIDAS 2.0 assurance levels and EU Digital Identity Wallet integration. Zero personal data on-chain; only nullifier and age-range attestation.
Verify: Verifier V2 contract ↗ CA registry (134 countries) ↗
Wealth tech for institutional and high-net-worth clients
Institutional portfolio analytics and AI-assisted wealth optimisation with EU AI Act Article 12 audit trail on every recommendation. Via accredited intermediaries; no retail self-service.
Autonomous agent orchestration with cryptographic accountability
Outcome-level accountability for AI agent operations; not just telemetry. Every run produces a signed RunProofCertificate: cost, approval chain, decision trace, and on-chain anchor. Built for CTOs and Heads of AI in regulated environments.
DeFi products
Institutional DeFi vault · verified settlement, ZK-anchored
The reference production deployment of Vauban Claim Algebra for verified financial transaction claims. Solver-based CoW execution on Starknet mainnet. STARK proof of solver execution and non-discriminatory matching; publicly verifiable, every batch. Institutional vault management and DeFi compliance without custodial intermediaries.
Cryptographic proof layer for regulated AI and agent payments
Post-quantum STARK proof of payment conditions for x402-native flows. Designed for EU AI Act Article 12 logging obligations and MiCA record-keeping requirements. IETF Internet-Draft draft-vauban-x402-consolidated-00 submitted (Independent Submission stream, May 2026); 13 organizations engaged in the Linux Foundation x402 working-group discussion; Zero-Trust Receipt Verifier demo live (JCS · SHA-256 · felt252 · on-chain anchor); reproducible interop fixtures across Rust, Python, Node.js, Go, Java.
Verify: Rekor entry (zkpay v0.3.0) ↗
The Vauban Proof Stack Framework (VPSF) defines a composable claim algebra; a chain-agnostic formal grammar for cryptographic claims that operates above any settlement layer. The core specification contains no blockchain-specific types: chain adapters implement the grammar for each target environment. Today's reference implementation anchors on Starknet via ZK-STARK proofs; hash-based, post-quantum-secure, no trusted setup. The same grammar already expresses six classes of claim: personhood, agentic authority, knowledge and decision provenance, process and computation integrity, financial attestation, and state and artifact provenance. One grammar, six classes, composable across trust roots.
The architecture preserves institutional sovereignty at the specification level: no vendor lock-in by invariant, not just by policy. Claims produced today against Starknet are consumable by future adapters targeting other settlement environments. An IETF Internet-Draft to standardise the Vauban Claim Algebra is in preparation, following the precedent of RFC 9449 (DPoP) and W3C Verifiable Credentials. Starknet's STRK20 delivers value-privacy on-chain ; it hides who paid whom and how much ("follow the money"). Vauban operates on a different axis ; it proves attributes about a subject without revealing identity, and composes those proofs across domains ("prove the actor"). The two are complementary: an anonymous on-chain action can be gated on a Vauban claim that the actor is a unique human or an eligible party.
VPSF is chain-agnostic by invariant. Starknet is the current reference implementation (ZK-STARK, no trusted setup). Multi-chain adapters are in research; first non-Starknet testnet target Q3 2027.