AKHU is cryptographic observation infrastructure for AI systems. Every observation is SHA-256 hash-chained in real time, Merkle-verified, append-only. Three evidence layers: L0 (observations), L1 (axioms: your declared rules), L2 (contradictions: structural facts, not alerts). Ed25519-signed licenses. Self-declared compromise events. If the system goes blind, it declares its own blindness with the same cryptographic integrity as sight. Built for EU AI Act Article 12. On-premise. No SaaS.

Wispr Flow: Dictation That Works Everywhere — Stop typing. Start speaking. 4x faster.

Stop typing. Start speaking. 4x faster.

hi everyone! I'm Nehuén, founder of AKHU.

most AI traceability today depends on reconstructing what happened after the fact. logs, dashboards, audit trails built retroactively. AKHU takes a different approach: every observation is SHA-256 hash-chained in real time, append-only, with Merkle verification. If the system goes blind, it declares its own blindness with the same cryptographic integrity as sight.

three layers: L0 (raw observations), L1 (declared axioms), L2 (contradictions as deterministic structural facts). Built with EU AI Act Article 12 in mind, but it works for any regulated system.

the live demo is running right now with 10 observation channels across fintech, pharma, nuclear safety, and aerospace. Real contradictions appearing in real time. No signup, just look: https://demo.akhu.ai

want to run it yourself? download the free Docker package. No cloud, no account: https://akhu.ai/download

we publish our threat model because integrity that isn't auditable isn't integrity: https://akhu.ai/threat-model

would love feedback from anyone working on AI governance or compliance infrastructure. Let's connect on LinkedIn for updates!