Constitutional System Governance · Patent Pending

Architecture that
cannot be violated.

Vestral enforces your software's constitutional rules on every commit — before the code merges, before the damage is done.

csg — constitutional check · Layer 6 of 8
$ git commit -m "add wallet endpoint"
↺ Running Vestral constitutional check…
✓ Layer 0 dependency 112 edges 0 findings
✓ Layer 3 trust-boundary 67 edges 0 findings
✗ Layer 6 governance 2 findings
 
MISSING role:public →[prohibits]→ op:wallet-operations
Unauthenticated callers can reach the coin ledger.
Fix: add authMiddleware to POST /api/wallet
 
✗ Commit blocked. 2 constitutional findings must be resolved.
The problem

AI writes code faster than
humans can govern it.

Architecture decisions get made in Slack. Enforced nowhere. Dead within weeks. AI coding agents have no memory of what the system is constitutionally permitted to do.

📉

Architecture rules decay

ADRs are documents, not enforcement. Every sprint introduces drift that no existing tool catches until production breaks or an audit fails.

🤖

AI agents start blind

Each coding session begins with zero awareness of your governance doctrine. The agent generates — your constitution is invisible to it.

🔍

Audits are manual and expensive

EU AI Act, SOC 2, ISO 27001 require provable architectural lineage. Most teams have spreadsheets. Auditors have billable hours.

No existing tool can answer: "Is this commit constitutionally legal?" — before it merges.
How it works

From questionnaire to
enforced constitution in minutes.

Answer questions about your system. Vestral synthesises the full 8-layer constitutional model, wires it into your CI pipeline, and enforces it on every commit.

Step 01

Describe your system

Fill in the setup wizard — domain, roles, workflow states, economic model, threat model, integrations. Upload specs or regulations if you have them.

~10 minutes
Step 02

Receive your package

Vestral generates an 8-layer doctrine model, pre-configured analysis server, project knowledge base, and governance context file for your AI coding agent.

AI-synthesised · ~90 seconds
Step 03

Run the analysis server

Start node serve.mjs. Vestral extracts your code's live topology and reconciles it against your constitutional doctrine. Findings appear in the browser.

Runs locally · zero cloud dependency
Step 04

Enforce on every commit

A pre-commit hook and CI gate block the merge of code to the repository when there are findings pending fix.

GitHub · GitLab · Bitbucket · Azure DevOps
Constitutional architecture

8 semantic layers. Every angle covered.

Vestral reconciles observed code topology against your doctrine. Below: two live layer graphs from an actual project.

Layer 6 — Governance

Which roles are authorized, prohibited, or required to perform each operation
role:public role:authenticated role:admin role:orig-notary role:system op:registration op:case-mgmt op:wallet-ops op:delivery-mgmt op:pdf-certification op:case-lifecycle authorizes prohibits authorizes authorizes authorizes constitutionalizes
authorizes
prohibits
constitutionalizes

Layer 3 — Trust Boundary

Which routes require which middleware guards before reaching controllers
GET /relay/:token POST /api/cases POST /api/wallet POST /api/admin/** POST /api/events authMiddleware walletGate requireAdmin relayController casesController walletController adminController eventsController rate-limited · no-auth
public route
authMiddleware
walletGate
requireAdmin
Live output

What Vestral shows after analysis.

Layer 0 · Dependency graph
Dependency Graph LIVE
0 · Dependency
3 · Trust
4 · Econ
6 · Governance
⟳ Recalculate
L0 Dependency
L1 Causality
L2 Lifecycle
L3 Trust Boundary
L4 Econ State
L5 Econ Auth
L6 Governance
L7 Adversarial
Constitutional Graph
Every node is a module, route, or service in your codebase. Every edge is a relationship — dependency, causality, trust gate, economic flow — mapped across 8 constitutional layers. The sphere makes the full topology visible at a glance.
Where text rules require a reader who already knows where to look, a graph makes the shape of your system immediately readable — who calls whom, which boundaries exist, which paths are guarded. Violations in the findings panel on the right trace directly back to edges and nodes visible here.
The graph also acts as persistent memory for the AI coding agent. Rather than re-deriving context from source files on every change, the agent reads the live graph state — knowing exactly which nodes are clean, which are flagged, and which constitutional rules apply before writing a single line.
14Compliant
2Partial
1Unmonitored
1Excess
0Missing
Remediation report
Open findings
PARTIAL cases/ → wallet/
Dependency exists in code. Doctrine mandates walletGate middleware on this path — not wired up.
EXCESS admin/ module
Module present in codebase but absent from Layer 0 doctrine. Vestral cannot classify any dependency into or out of it.
UNMONITORED delivery/ → notify/
notifications/ not declared in doctrine. Outbound calls from delivery/ are outside constitutional scope.
System architecture

Vestral controls every path to the repository.

spec ingestion doctrine YAML const. context state sync pre-code context drops files pre-commit hook ✓ authorized non-compliant findings post-reject CI / pre-commit divergence? Vestral Constitutional Enforcement Agent 8-layer graph · doctrine diff autonomous gates Spec & Docs OpenAPI · arch · requirements Questionnaire Setup wizard · AI synthesis ~10 min · generates doctrine 8 × doctrine.yaml L0–L7 constitutional rules Wiki Brain AI context filesystem/ staged changes hook triggers AI Coding Agents Claude / Copilot / Cursor / Codex, etc. generates code changes Repository Only Vestral-authorized commits reach here git push · merge · deploy spec ingestion doctrine / code input authorized path blocked · findings returned
Setup wizard

Your constitutional model
in 10 minutes.

Answer 7 questions about your system. Vestral reads your answers — and any documents you upload — and generates the full constitutional package using AI.

No code reading required at this stage. You describe the intent; Vestral enforces it.

  • Project identity & paths
  • Supporting documents
  • Purpose & domain
  • Roles & access
  • Workflow & state
  • Economic model
  • Threat model
What you receive
✓ 8 doctrine YAML files
✓ Pre-configured serve.mjs
✓ AI context file (CLAUDE.md / Cursor / Copilot)
✓ CI gate workflow
✓ 8-page knowledge base
✓ invariants.md
app.vestral.io/setup
Step 3 of 7 — Purpose & Domain
What does this system do?
Describe the system's core function. This becomes the foundation of your constitutional doctrine.
A platform for licensed notaries to certify and deliver legal documents across jurisdictions, with cryptographic integrity guarantees under D.Leg. 1049…
Legal · Notarial · FinTech
Peru D.Leg. 1049 · GDPR · SOC 2
Never certify without notarial signature
Never release payout before delivery…
← Back
Progress auto-saved
Next →
Pricing

One plan for every team size.

All plans include the full 8-layer engine, AI fix generation, CI gate, and knowledge base. No per-seat fees.

Solo
$499/yr
1 project · 1 developer
  • 8-layer constitutional engine
  • AI fix generation
  • CI gate (all platforms)
  • Wiki knowledge base
  • 14-day free trial
Start free trial
Popular
Studio
$1,999/yr
Up to 5 projects · unlimited devs
  • Everything in Solo
  • 5 simultaneous projects
  • Multi-agent governance
  • Audit export (PDF / JSON)
  • Priority support
Start free trial
Enterprise Gov
$4,999/yr
Unlimited projects · compliance
  • Everything in Studio
  • EU AI Act reports
  • SOC 2 / ISO 27001 exports
  • SSO / SAML
  • SLA + dedicated support
Start free trial
Enterprise
Contact us
Custom · on-premise · white-label
  • Everything in Gov
  • On-premise deployment
  • Custom doctrine layers
  • White-label option
  • Volume pricing
Get in touch
FAQ

Common questions.

Everything you need to know before you generate your first constitutional package.

What exactly is Vestral? How is it different from a linter or static analysis tool?
A linter checks code style and known anti-patterns against a fixed rule set. Static analysis finds bugs and type errors. Vestral is neither. It extracts an 8-layer semantic graph from your repository — dependency, causality, lifecycle, trust-boundary, economic-state, economic-authorization, governance, and adversarial propagation — and compares the observed topology against a user-authored constitutional doctrine: what the system is architecturally permitted to do. Violations are classified as MISSING, EXCESS, or MISMATCH and blocked at the pre-commit hook before they reach production. No existing linter or static analysis tool makes the distinction between "what the code does" and "what the code is constitutionally allowed to do."
Do I need to rewrite my codebase? How disruptive is the integration?
No rewrites. Vestral reads your existing repository as-is. The setup wizard generates a constitutional package — doctrine YAMLs, a pre-configured server, an AI context file, and a CI gate — and drops it alongside your code. You point it at your repo root, run node serve.mjs, and the 8-layer graph is computed from your existing dependency tree, routes, middleware, and event handlers. The only code you touch is adding a pre-commit hook (one line) and a CI step (one job). Existing code is read, never modified.
What languages and frameworks are supported?
The current extraction engine covers Node.js / TypeScript (Express, Fastify, NestJS) natively — function-level call graph, route-to-controller mapping, and middleware chain extraction are all fully implemented. Python (Django, FastAPI) and Go extractors are in active development. The doctrine layer and Vestral server are language-agnostic: any language with a call-graph extractor can feed the 8-layer model. Enterprise customers can request custom extractor development for Java/Spring, Ruby on Rails, or other stacks.
Is my source code sent to the cloud? What data leaves my machine?
The Vestral analysis engine runs entirely on your machine. The server (serve.mjs), all extractors, and the doctrine evaluation pipeline execute locally. No source code, graph data, or diff results are ever transmitted to Vestral's servers during normal operation. The only outbound calls are: (1) license key validation at activation (sends only your license key and a machine fingerprint), and (2) the AI fix generation feature, which sends a short excerpt of the specific violating code to the Anthropic API over TLS — and only when you explicitly click "Generate fix." Both can be disabled. Enterprises may deploy in fully air-gapped environments.
How does Vestral work with AI coding assistants like Claude Code, Copilot, or Cursor?
The setup wizard generates an AI context file tailored to your assistant of choice — CLAUDE.md for Claude Code, .cursor/rules/csg.mdc for Cursor, .github/copilot-instructions.md for GitHub Copilot, and equivalents for Windsurf and Aider. This file instructs the AI to read diff.json before writing any code, confirm findings before proceeding, and treat invariant violations as hard blockers. The AI assistant effectively internalises your constitutional doctrine and enforces it before suggesting or writing changes — not just after a commit. Every generated context file also includes the Vestral consultation protocol: the agent must print [Vestral] Reading <artifact> — <reason> on screen before any read from the package, giving you visible, auditable evidence in every session that the constitution was in the loop.
What is the EU AI Act connection? Why does it matter right now?
The EU AI Act (fully applicable from August 2026) requires high-risk AI systems to maintain technical documentation demonstrating that the system's architecture is consistent with its declared design. Article 9 mandates a risk management system; Article 11 mandates technical documentation including "the design specifications of the system." Vestral produces provable architectural lineage — a machine-readable diff between declared doctrine and observed topology at every commit. This is exactly the kind of evidence Article 11 compliance requires. For organisations using AI to write code (which, by mid-2026, is most of them), Vestral answers the compliance question the Act cannot: "Is this AI-generated commit constitutionally legal?"
How does the trial work? What happens after it ends?
Every organisation gets a free 3-day trial, one trial per organisation per 60 days. During the trial you get the full product — all 8 layers, AI fix generation, CI gate, and knowledge base — with no feature restrictions. When the trial expires, the server enters read-only mode: you can still view your existing diff results and graph but cannot run new analyses until you activate a license. Activating is one command: enter your license key at localhost:8082/install. Licenses are annual; they are tied to the organisation, not the developer seat count.
Can Vestral detect security vulnerabilities?
Indirectly, yes — and often earlier than vulnerability scanners. Vestral's trust-boundary layer enforces that every externally reachable route passes through the declared authentication and authorisation middleware chain. If a developer adds a new route that bypasses authMiddleware, Vestral classifies it as a trust-boundary EXCESS violation and blocks the commit. The adversarial propagation layer models known attack vectors — privilege escalation, replay, IDOR, injection — and verifies that the declared mitigations remain in the graph. It is not a CVE scanner; it is a constitutional guarantee that your security architecture cannot structurally regress.
What does the patent protect? What does "15 claims pending" mean in practice?
The patent application covers three claim families: (1) the 8-layer constitutional graph system as a data structure and extraction method, (2) the governance-constrained reachability analysis — the specific algorithm that reconciles observed topology against a constitutional doctrine to produce typed violation findings, and (3) the AI governance gate — the method of injecting constitutional context into an AI coding session to enforce doctrine at suggestion time, not just at commit time. "15 claims pending" means 15 individual claim scopes have been filed. The application is currently under examination; pending status provides provisional protection from the filing date and puts competitors on constructive notice that this territory is claimed.
How long does the initial setup take for an existing codebase?
For most codebases: under 30 minutes end-to-end. The setup wizard takes roughly 10 minutes to complete (or 2 minutes if you paste an existing spec). AI synthesis — three passes through the constitutional model — takes another 3–5 minutes. Pointing the generated server at your repository and running the first analysis takes 2–10 minutes depending on codebase size. Adding the pre-commit hook and CI step takes under 5 minutes. The generated CLAUDE.md or Cursor rules file is ready to use immediately. There is no training period and no model fine-tuning — Vestral reads your code structurally, not semantically.
Patent pending · 15 claims filed

First implementation of
governance-constrained reachability.

Vestral separates normative topology — what your code is constitutionally permitted to do — from descriptive topology — what it actually does. No existing tool makes this distinction.

8
Constitutional layers
15
Patent claims
0
Direct competitors
$12B
Developer tooling TAM
⚖️  Patent pending · Method + System + CRM claim families · 2026
Investor Relations

A new category in developer tooling.
No direct competitors. Patent pending.

Vestral is the first tool that enforces what software is constitutionally permitted to do — not just what it currently does. We operate at the intersection of governance, AI coding, and enterprise compliance, in a space no existing tool addresses.

$12B+
Developer tooling TAM (2025)
15
Patent claims filed across 3 families
0
Direct competitors in this category
Aug 2026
EU AI Act full enforcement deadline
🏗️

The space Vestral occupies

Developer tooling has three layers: syntax (linters), behaviour (static analysis), and now — for the first time — constitutional governance. Vestral owns this third layer: the enforcement of what a system is architecturally allowed to do, expressed as a machine-readable doctrine and verified at every commit.

🏢

The enterprise problem

At scale, architecture decays silently. Rules live in PDFs, wikis, and Notion pages that no tool reads. Every pull request is a trust decision made by a human who may not know the full doctrine. Vestral replaces that trust gap with a gate: no commit that violates declared architecture reaches production, regardless of who wrote it — human or AI agent.

🤖

The AI coding shift

By 2026, the majority of production code is written by AI agents — Claude Code, Copilot, Cursor, Codex. These agents are fast, capable, and constitutionally blind. They have no awareness of trust boundaries, economic-state rules, or governance invariants. Every enterprise adopting AI coding now needs a constitutional layer. Vestral is that layer.

📋

Regulatory tailwind

The EU AI Act (fully applicable August 2026) requires high-risk AI systems to maintain technical documentation proving architectural consistency with declared design. Vestral produces that evidence automatically at every commit — a machine-readable audit trail no compliance team can generate manually. For enterprises subject to the Act, Vestral moves from useful to necessary.

⚖️  Patent protection — 15 claims pending, 3 families

The patent application covers three claim families: (1) the 8-layer constitutional graph as a data structure and extraction method — the core architecture no competitor can replicate without infringing; (2) governance-constrained reachability analysis — the specific algorithm that reconciles observed code topology against a constitutional doctrine to produce typed violation findings; and (3) the AI governance gate — the method of injecting constitutional context into an AI coding session to enforce doctrine at suggestion time, not just at commit time. The application is under examination. Pending status provides provisional protection from the filing date and puts any would-be competitor on constructive notice. No comparable system exists in prior art.

Get in touch

We're open to conversations with seed and pre-seed investors, strategic partners, and enterprise design partners. Deck and data room available on request.