Provable AI Infrastructure

AI decisions
auditors can
verify.

Your AI makes decisions. Zorynex makes them provable.

✓ v1.0.0 tagged ✓ 593 tests passing ✓ Independently verifiable
Open source on GitHub  •  30k+ lines  •  593 tests passing
<10ms
Added latency
4
Verification checks
0
Server trust required
verify_package.py — Proof Verification Report
$ python verify/verify_package.py proof_loan_9284.json

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
ZORYNEX
PROOF VERIFICATION REPORT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Instance loan-9284
Fingerprint fea8858c0d9014...
Mode OFFLINE
────────────────────────────────
✓ VERDICT: PROOF IS VALID EVIDENCE
────────────────────────────────
CRYPTOGRAPHIC VERIFICATION (4 checks)
────────────────────────────────
Package structure valid
Package untampered
Chain valid
Original signer verified

$

AI regulation is moving faster than audit infrastructure  —  Most companies can explain AI decisions  —  Very few can prove them.

Live verification

See it work.
Real proof. 30 seconds.

A real cryptographic proof artifact — open the verifier and watch all 4 checks pass. No account. No API key. No server required.

Open live verifier →

Sample proof auto-loads — just verify it.

How it works
1  Click Open Live Verifier below
2  Sample proof loads automatically
3  Watch all 4 cryptographic checks pass
4  Download PDF audit report
  Package structure valid
  Package untampered
  Chain valid
  Original signer verified
FINAL VERDICT: ✓ VALID EVIDENCE
The problem

A regulator walks in and asks:
“Prove exactly what this AI decided — with evidence I can verify.”

Not describe it. Not reconstruct it later. Prove it.
Most teams cannot.

Without Zorynex
✗  2–6 weeks reconstructing evidence
✗  Legal exposure during audit gap
✗  Manual engineering work
✗  Logs that can be altered
With Zorynex
  Proof ready in seconds
  Independent verification
  Cryptographic evidence
  Tamper-evident by design
Where it breaks today
Loan rejected
No verifiable reason trail. Auditor asks which model made the decision. Your team cannot prove it.
Insurance claim denied
No immutable audit record. Regulator demands the exact policy version in effect at decision time.
AI agent executes action
No independent verification layer. Something went wrong downstream. Evidence reconstruction takes weeks.
Regulatory examination
Examiner asks for proof. Your team sends logs, dashboards, screenshots, and manual explanations. None of it is verifiable evidence.
When an auditor asks “prove what happened”
  Logs that can be edited after the fact
  Dashboards that are not evidence
  Model outputs without provenance
  Screenshots and manual reconstructions
  “Give us two weeks to pull this together”
With Zorynex
  Cryptographic proof — signed at execution
  Hash-chained ledger — tamper-evident by construction
  Exportable artifact — one file, forever
  Independently verifiable — no trust required
  Evidence ready in seconds, not weeks
Architecture

Sits after the decision.
Changes nothing.

Keep your existing AI stack. Add independent verification in one afternoon.

< 10ms overhead at p95 2,100+ decisions/sec sustained Full benchmarks →
Your existing stack Provable AI adds this
AI system Model + input data Decision Model output + policy applied Provable AI layer Deterministic + signed · sits after decision · zero changes Cryptographic ledger entry prev_hash + curr_hash + signature Storage / logs Your existing infrastructure Proof export proof.json Browser · CLI · API python verify/verify_package.py proof.json
Your existing stack
Provable AI layer

Want to test this in your environment?

Book 30-min technical walkthrough →
Core features

8 features.
One verifiable system.

01

Deterministic Decision Protocols

Workflow specs compile to a deterministic protocol hash. Same spec → same hash across all environments. Proves execution consistency.

02

Governance Enforcement

Controls which models, agents, and policies are approved to execute. Unauthorized versions are blocked at runtime automatically.

03

Cryptographic Execution Ledger

Each decision is Ed25519 signed and SHA-256 hash-chained. Append-only enforced at the database trigger level — even a compromised application layer cannot modify ledger history.

04

Signed Proof Artifact Export

Every decision exports a self-contained package: GET /proof/export/{id}?inline=true. Hand it to any auditor. No server access ever needed.

05

Three-Path Verification

CLI (verify/verify_package.py), browser drag-and-drop (/verify-ui on your deployment), or API (POST /verify-package). Same 4 cryptographic checks. Zero server access.

06

Replay-based Tamper Detection

Reconstruct any execution history and replay it exactly. If the output differs from the recorded proof — tamper detected. Mathematically impossible to falsify.

07

Environment Drift Detection

Compare system roots across dev, staging, and production. Detects governance drift before deployment — not after a failed audit.

08

SIEM & KMS Integration

4 SIEM transports (webhook, syslog, Splunk, Datadog). AWS KMS signing with auto-failback. Private key never leaves your infrastructure.

How it works

From decision
to proof.

Every AI decision follows this chain — automatically, at the moment it fires.

STEP 01

Record Decision

POST /decision — 4 fields minimum. Governance auto-resolves.

STEP 02

Governance Enforced

Only approved models, agents, and policies can write decisions.

STEP 03

Ledger Written

Ed25519 signed. SHA-256 hash chain. Append-only.

STEP 04

Package Exported

GET /proof/export/{id}?inline=true — self-contained. Verifiable forever.

STEP 05

Independently Verified

Browser, CLI, or API. 4 checks. Zero server access. Zero trust required.

Verification

Three paths.
One proof.

Auditor verification — no server needed
$ pip install pynacl            # one-time
$ python verify_package.py proof_loan9284.json
FINAL VERDICT:  ✓  VALID EVIDENCE
No Zorynex account  •  No API key  •  No network at verify time  •  Works offline permanently

Every path runs the same 4 cryptographic checks. Zero server access. Zero trust in Zorynex required.

PATH 01

Browser — for compliance teams

On your deployment: open /verify-ui. Drag the proof package in. Four checks run in the browser. Download a branded PDF audit report. No terminal. No API key. No account.

PATH 02

CLI — for engineers

One-time setup: pip install pynacl. Then run python verify_package.py proof.json — no server, no account, no network. Returns FINAL VERDICT: VALID EVIDENCE or exact failure reason.

PATH 03

API — for integrations

POST /verify-package returns structured JSON with per-check results, failure reason, instance metadata, and signing key. Plug into your CI/CD or compliance tooling.

verified true
instance_id loan-9284
final_state approved
proof_count 2
checks[0] ✓ Package structure valid
checks[1] ✓ Package untampered
checks[2] ✓ Chain valid (2 decisions, seq 1→2)
checks[3] ✓ Original signer verified (env-a428fc62...)
chain_length 2
fingerprint b7ee4d91b9fcde28...
failure_reason null
Who it's for

Built for the people
regulators call first.

Chief Risk Officer / CRO

Head of Risk

When regulators demand reproducible proof of AI decisions, show them cryptographic evidence — not logs that could have changed. Designed for teams operating under SR 11-7, EU AI Act, and CFPB oversight requirements.

  • Independent verification without server access
  • Deterministic replay of any past decision
  • Regulatory evidence package in seconds
Model Risk / AI Governance

Model Risk Manager

Stop reconstructing validation evidence manually. Every model decision auto-generates its own proof artifact. Environment drift detected before deployment.

  • Governance enforcement at every execution
  • Environment drift detection across dev/prod
  • Immutable model version audit trail
CTO / VP Engineering

AI Engineering Lead

FastAPI server alongside your existing AI stack. Python & TypeScript SDKs. Docker and Helm. 593 tests. No rearchitecting required.

  • Python SDK, TypeScript SDK, Postman collection (38 requests)
  • Docker: docker compose up — server running in seconds
  • Kubernetes: helm install zorynex ./helm/zorynex
  • Swagger UI, browser verify-ui, compliance dashboard — all included
When proof matters

Built for this.
Not everything.

Built for
  • Financial decisions — credit, lending, underwriting
  • Insurance claims processing and denial workflows
  • Healthcare recommendations with regulatory exposure
  • Government and public-sector AI automation
  • Autonomous agents executing consequential actions
  • Any AI decision that can trigger compliance, legal, or financial risk
Not the right fit if
  • Chatbot and conversational AI products
  • Marketing personalisation and recommendation engines
  • Internal productivity tools with no regulatory exposure
  • Low-stakes AI workflows where audit trails are optional
If your AI makes a decision that could be questioned by a regulator, contested in a dispute, or examined in an audit — that is the decision Zorynex is built for.
Enterprise trust layer

Built to pass
security review.

Every question a bank's InfoSec team, DPO, or compliance officer will ask — answered before the pilot starts.

Multi-tenant isolation

Every proof, chain, and export is scoped by tenant_id and enforced at the database query layer. No cross-tenant access — even by accident, even by a privileged user.

Append-only ledger

UPDATE and DELETE are physically blocked by database triggers. Even a compromised application layer cannot alter ledger history. The block happens at the database, not the application.

PII never stored

Raw customer inputs are SHA-256 hashed before anything is written. Only the hash enters the proof ledger. The original value is discarded immediately. Your customers’ data stays yours.

System root hash

One hash proves your entire ledger collection is intact. GET /system/root recomputes it on demand. If any record anywhere has been altered, the hash changes. No external auditor required.

Deployment

Runs in your
infrastructure.

No Zorynex cloud. No data leaves your environment. Deploy anywhere your team already operates.

Self-hosted
Private VPC
On-premise
AWS / GCP / Azure
Air-gapped verify
Docker: docker compose up  •  Kubernetes: helm install zorynex ./helm/zorynex  •  SQLite (dev) or PostgreSQL (prod)
FAQ

Everything you need
to know.

How long does integration take?

+
A developer can have proof artifacts generating in an afternoon. Provable AI sits after your existing AI decisions — nothing changes in your pipeline.
Step 1 pip install + uvicorn server.main:app → 5 min
Step 2 Point your AI output to the API endpoint → 1–2 hrs
Step 3 Test proof export + verify → 30 min
Total: half a day for one developer

Does it change anything in my existing AI pipeline?

+
Nothing changes. Provable AI is a layer that sits after the decision. Your model makes the decision exactly as it does today. Provable AI records and signs what already happened — it doesn't touch your inference pipeline, your model weights, or your existing infrastructure.

Does the verifier need access to our servers?

+
No server access required — ever. The proof artifact is a self-contained signed JSON file. Auditors run python verify_package.py proof.json entirely offline — one file, no server, no account. The proof either verifies cryptographically or it doesn't.

What regulatory frameworks is this designed for?

+
Zorynex helps teams operating under frameworks such as SR 11-7 (model risk management), the EU AI Act, and CFPB oversight requirements — by creating verifiable audit evidence for every AI decision. It does not replace legal, compliance, or regulatory counsel.

Can our compliance team verify without running code?

+
Yes — browser drag-and-drop. On your Zorynex deployment, open /verify-ui. Drop the proof package file. Four green checks appear in seconds. Download a branded PDF verification report. No terminal. No API key. No Zorynex account. Works completely offline. This is the path we recommend for compliance teams, CROs, and regulators.
90-day pilot

Enterprise
pilots open.

We're running a 90-day pilot with fintech teams this quarter. Full deployment. Defined outcomes. Direct access to founding team throughout.

  • Full Provable AI infrastructure deployment (35-endpoint FastAPI server)
  • Integration with your existing AI stack — half a day, one developer
  • Audit-ready proof artifacts generating from day one
  • Browser verify-ui + CLI verifier for your compliance team (pip install pynacl — one-time)
  • Branded PDF verification report for regulatory submissions
  • Environment drift detection, KMS signing, SIEM integration
  • SR 11-7, EU AI Act, CFPB documentation pack delivered at day 90
  • Direct access to founding team throughout pilot

Try live verifier → Book a demo →

✓ Received. Hanif will reply within 24 hours.