AI claims live in vendor logs you have to trust. Orynq certifies inference traces through the Materios committee and anchors the receipts to Cardano — so “the model said X” becomes a proof anyone can replay, not a log line you control.
npm install @fluxpointstudios/orynq-sdk-clientPOST /anchors/process-trace → receive txHash.If a regulator or court audited your AI system in five years, what verifiable evidence could you produce to prove how a decision was made? Who governed it, and that neither the data nor the model was altered over time?
Six terms that get used interchangeably in AI audit discussions. Here's the precise meaning of each — and which one Orynq solves in code.
Each clause of the audit question maps to a specific cryptographic mechanism in the SDK. Together they make the trace replayable, attributable, and time-anchored.
Every span, event, decision, and tool call gets a hash that includes the previous event’s hash. Reorder, insert, or delete anything and the chain breaks. The full decision pathway replays byte-for-byte five years later.
Each anchor transaction is signed by a Cardano wallet — on-chain identity. Governance attestations (compliance review, release approval, role-based sign-off) live as signed events in the same Merkle bundle as the decision they govern.
A manifest hash — model checkpoint + tokenizer + system prompt + training-data fingerprint — commits with every trace. Anyone can verify the exact model and data state used at that moment. Fine-tune the model? Anchor a new manifest. Drift is cryptographically detectable.
These guarantees hold if you use the SDK with discipline — manifests must be pinned before execution, governance events must be signed by their actual signers, and the raw off-chain trace must be preserved alongside the txHash. Gaps we're actively closing live as open issues on github.com/Flux-Point-Studios/orynq-sdk.
Orynq is our cryptographic AI process tracing and anchoring SDK. It captures every span, event, decision, and tool call in an agent's execution as a rolling-hash chain, then bundles them into a Merkle tree that can be selectively disclosed.
The bundle's root hash anchors to Cardano under metadata label 2222, or to the Materios partner chain for committee-certified, high-throughput receipts with an hourly Cardano L1 anchor.
Anyone can verify any anchor using only the txHash — Blockfrost, our open verifier, or any Cardano explorer works. No API key, no token gating, no trust.
On-chain stores hashes and pointers only, never raw prompts, responses, or sensitive data. And because the trace is bundled into a Merkle tree, you can selectively disclose only the events that matter — keeping the rest of your AI's execution private.
Rolling hash chains prove event ordering across every span, decision, and tool call. Merkle trees enable selective disclosure — reveal specific events without exposing the rest of the trace.
Cardano L1 under metadata label 2222 for permanent settlement, or the Materios partner chain for high-throughput, committee-certified receipts with an hourly Cardano anchor.
Self-host with your own Cardano wallet (~0.2–0.3 ADA per anchor) or use the managed Anchor-as-a-Service with x402 / Flux payment protocols. No subscriptions required.
Provide a txHash; replay the inclusion proof with Blockfrost, our open verifier, or any Cardano explorer. No API keys, no token gating, no trust required.
Cryptographically falsifiable audit trail for AI inference. Built for regulators, security auditors, and AI governance teams that need more than vendor logs.
Drop-in MCP server lets AI agents record their own process traces. Compatible with Claude, Cursor, OpenClaw, and any MCP-aware host out of the box.
Anchor your AI agent's decisions to the Cardano blockchain in four simple steps. No blockchain expertise required.
Your AI agent runs with the Orynq SDK instrumentation, capturing decisions and actions in real-time.
SDK generates a cryptographic trace manifest containing only hashes - no sensitive data leaves your system.
Call our API with your enterprise subscription or pay-per-anchor via x402 - no wallet management needed.
Receive txHash instantly. Verify your anchor anytime via Cardanoscan or our built-in verifier.
Ready to get started? View the documentation
The three regimes that actually require AI traceability today, and the Orynq primitive that satisfies each. Cryptographic AI audit trail by design, not by procedure.
High-risk AI systems must "automatically record events ('logs') over the duration of the lifetime of the system" with traceability sufficient to "facilitate the post-market monitoring" of system behavior.
Every Orynq trace satisfies the "automatic recording" requirement. Cardano anchoring extends "duration of lifetime" past the operator’s control — the log survives company shutdown, vendor change, or adversarial deletion.
"Mechanisms are in place to inventory AI systems, deployed and decommissioned" and "AI systems are evaluated for trustworthy characteristics" — both requiring tamper-evident records of model state and inference behavior.
manifestHash anchors model + tokenizer + training-data fingerprint with every inference. The Merkle bundle gives MEASURE-2 evaluators selective access to specific decisions without exposing the full trace.
"Records of AI system decisions shall be maintained" (A.7.4) and "competence and awareness of governance roles shall be evidenced" (A.6.2) — together demanding signed governance attestations alongside decision logs.
Governance attestations live as signed events in the same Merkle bundle as the decision they govern (issue #58 on the orynq-sdk repo standardizes the schema). Auditors get decision + governance lineage in one verifiable artifact.
Experience blockchain anchoring firsthand. The demo uses Cardano preprod testnet - no real ADA required.
Independently confirm an anchor exists on Cardano mainnet. Cryptographic proof — nothing to trust about us.
Prove an Orynq anchor exists on the Cardano blockchain
Provide to verify the on-chain manifest hash matches your expected value
Orynq anchors are cryptographic proofs stored on-chain using Cardano metadata label 2222.
Verification checks the blockchain directly to confirm anchor existence and data integrity.
Ship anchored AI in your preferred language. Same SDK shape across TypeScript, Python, and curl.
LangSmith, Arize, and Datadog LLM Observability are private dashboards — useful for engineering, not defensible in an audit. Here's the gap they leave and where Orynq fills it.
Application logs DB rows, Loki, Splunk | LLM observability LangSmith, Arize, Datadog | Orynq Cryptographic anchoring | |
|---|---|---|---|
| Who can verify the record? | Whoever owns the database | Vendor + customer (login required) | Anyone with the txHash |
| Tamper-evidence | None — rows can be silently edited | Vendor-attested only | Cryptographic (rolling hash + Merkle) |
| Survives operator going dark | No — dies with the database | No — dies with the vendor | Yes — anchor lives on Cardano L1 |
| Selective disclosure | All or nothing | Whatever the dashboard shows | Per-event Merkle proofs |
| Model-state proof | None | Vendor-recorded (trust required) | manifestHash pinned per inference |
| Audit cost over 5 years | Storage + retention + legal | ~$30k+/yr vendor seat | ~0.2–0.3 ADA per anchor |
None of this means Orynq replaces LLM observability — it doesn't. Run LangSmith for your engineers and Orynq for your auditors. Same traces; two surfaces.
Pay only for what you use. No hidden fees.
No crypto wallet needed • API key authentication • Pay by card
Use your own Cardano wallet - pay only when you anchor
per anchor
At current ADA prices
Check out our managed plans above for prepaid credits and API key access, or contact us for custom high-volume pricing.
Prices may vary slightly based on Cardano network fees. USD estimate based on current ADA market rates.
Everything you need to know about Orynq anchoring
Have more questions? Check the documentation or contact us.
Interested in volume discounts or custom enterprise solutions? Fill out the form below and our team will get back to you within 24 hours.
Prefer email? Reach us directly at [email protected]