Prove it's real. Prove it hasn't changed.
A record you can quietly change is not proof. A record you can't — is. Witniumchain seals any digital record — an agreement, a decision, an AI output — to a tamper-evident chain. Verifiable by anyone, modifiable by no one, including us.
One MCP server. One API. One SDK.
We spent eight years building the answers. Almost no one was asking the questions. Now they're coming fast.
Latest episode · Friday, 16 October
Deepfakes, fines and the courts: trust is becoming a technical question
OpenAI and Meta warn US judges about deepfakes. Bank of America pays $1.2 billion. Courts face AI-generated evidence they can't verify.
- 01OpenAI and Meta warn US courts about deepfakes
- 02Bank of America fined $1.21 billion for AML lapses
- 03Deepfake video of Bhutanese woman posted without consent
One protocol.
Every integration.
Choose the surface that fits your stack. The cryptographic guarantees are identical across all three. Move between them without rewriting your evidence trail.
MCP Server
Native Model Context Protocol server. Drop into any MCP-capable platform or agent framework.
- Zero-config for Claude Desktop, Cursor, any MCP host
- Same cryptographic proofs as the API and SDK
- Tool descriptions tuned for LLM tool-use
REST API
OpenAPI-described, language-agnostic, every framework. Witness an event with a single HTTP request.
- Idempotency keys on every write
- Webhooks on every state change
- SDKs auto-generated from OpenAPI
SDKs
Native libraries. TypeScript shipping today; Python, Go, Java in the pipeline.
- npm install @witnium-tech/witniumchain
- Five auth modes, one client
- Strongly typed end-to-end
Three lines from import to first witness.
No keys to generate in advance. No infrastructure to provision. The SDK handles PKCE login, refresh, signing, and witness allowance reservation. You handle your domain.
See the full quickstartimport { WitniumchainClient } from '@witnium-tech/witniumchain';
const wc = new WitniumchainClient({
baseUrl: 'https://auth.witniumchain.com',
chainBaseUrl: 'https://api.witniumchain.com',
accessToken: oauthToken,
});
// Open a witness intent against your contract
const intent = await wc.proposeWitnessV5(myContract, {
dataId: sha256(aiOutput),
context: { model: 'claude-sonnet-4.7', actor: userId },
requiredSigners: [myDelegatedKey],
});
// Once signatures collected, seal it on-chain
const sealed = await wc.finalizeWitnessV5(myContract, intent.intentId);
// → sealed.witnessId, verifiable foreverBuilt for what's coming
through 2028.
The Act phases in across four dates. Each one expands the market of organisations that need provable AI infrastructure. Witniumchain is the primitive built for all four.
Article 50 transparency obligations apply: people must be told when they interact with AI, and AI-generated content must be marked. Obligations for general-purpose AI models have applied since August 2025; from now the Commission can fine their providers.
Generative AI systems placed on the market before August 2026 must now also mark their output in a machine-readable way. Marks can be stripped. A sealed record can't.
Annex III high-risk obligations apply, including Article 12: automatic logging across the lifetime of every high-risk AI system. Under Articles 19 and 26(6), providers and deployers keep the logs under their control for at least six months. The Act requires the logs. Tamper-evidence is what makes them hold up.
Annex I obligations apply to AI in regulated products such as medical devices, toys and radio equipment.
We can't see what you witness.
Article 12 requires high-risk AI systems to log events automatically across their lifetime. It says what to record, not how to protect it. A log its operator can quietly edit is weak evidence in front of an auditor or a court. Witniumchain makes every change detectable, and by design Witnium cannot read your data, hold your keys, or alter your records.
Owner keypairs are generated in your browser using Ed25519. The private key never reaches our servers. Witness content is hashed client-side; what reaches Witnium is the SHA-256 fingerprint, never the underlying data. Signatures are produced locally and submitted as opaque bytes.
This isn't a privacy policy commitment. It's a structural property of the SDK and the protocol. If a regulator, a court, or a competitor asks Witnium to produce a customer's content, we can't — there is no version of us with that capability.
A log the vendor can modify is not evidence. A log the vendor cannot see, cannot touch, and cannot alter is. That difference is what makes a Witniumchain record audit-grade.
What we do see: operation metadata (timestamps, public keys, fingerprints, witness counts), required for credit accounting and chain integrity. What we cannot see: any content, any private key, any signing material. The chain is verifiable independently of us at witniumchain.com.
The same chain solves five other problems
your team already has.
Forensic readiness
Reconstruct what an AI decision was based on, after the fact. Court-grade, third-party-verifiable.
Internal governance
Board-level visibility into what your autonomous agents actually did.
Customer trust
Every output ships with a verifiable evidence chain. A differentiator that scales.
Multi-party workflows
Co-signed witnesses, attestation chains, evidence binders for any process where more than one party needs to agree.
Incident reconstruction
When something goes wrong, the chain tells you exactly which output, when, on which inputs, under which model version.
Built for the teams
regulators actually visit.
Legal AI platforms
For platforms serving EU law firms and in-house counsel. Most legal AI is not high-risk under the AI Act, but lawyers live on evidence. Every output can carry a verifiable chain: what the AI produced, when, and on which sources. AI intended for use by courts, or in a similar way in alternative dispute resolution, is high-risk under Annex III. Your customers gain defensible work product; you gain a moat.
Enterprise AI agents
For platforms running autonomous agents on customer data. When an agent makes a decision that materially affects an employee, candidate, or customer, the chain proves what the agent saw, what it did, and why.
High-stakes vertical AI
For healthcare, financial services, public sector, and any AI domain where reconstruction matters. Witniumchain is the audit primitive that makes your AI defensible — in front of a regulator, an auditor, or a court.
We didn't build this for the AI Act.
The AI Act caught up to what we built.
Witnium Technologies was founded in Stockholm in 2016. We've been building Digital Witness — the cryptographic primitive that powers Witniumchain — for nearly a decade, in production with European law firms handling real legal evidence.
When the EU AI Act was drafted, our technology was already running. We didn't pivot to compliance. Compliance pivoted to us.
We are a Swedish company, in the EU jurisdiction the Act actually applies to. Our chain runs on infrastructure we control. Our cryptographic guarantees are not theoretical.
Get access now.
Run your first witness today.
Free tier includes 2,000 witnesses per month. No credit card required.