SignedAI is a multi-LLM verification consensus API that uses HexaCore 7 models (3 Western + 3 Eastern + 1 Thai) to benchmark hallucination below 0.3% through ED25519 RFC 8032 cryptographic signing — with Tier S, 4, 6, and 8 consensus levels for every enterprise risk tier.
SignedAI
Multi-LLM verification consensus API — cryptographically signed responses from HexaCore 7 Models, benchmarking hallucination below 0.3% on standard workloads, with complete audit trails and ED25519 RFC 8032 signing.
Why AI Verification Matters
SignedAI is the 5th Genome in the RCT 7-Genome System. It provides verification infrastructure for all other Genomes — every DelentiaAI output, every Analysearch result, every Delta Engine compression is certifiable through SignedAI.
Individual LLMs have unique failure modes — training data cutoffs, model bias, and hallucination patterns. SignedAI uses multi-LLM consensus to cross-verify outputs across the HexaCore 7-model roster, benchmarking hallucination below 0.3% on standard workloads.
Every verified response is cryptographically signed using ED25519 (RFC 8032), creating an immutable audit trail. Essential for regulated industries that require proof of AI decision trails.
6-Stage Verification Pipeline
Every SignedAI request flows through every stage — no step is skipped.
Receives the request, validates schema, assigns session ID, and classifies task priority.
JITNA selects which models to engage from the HexaCore 7-model roster based on task type.
All selected models independently process the same request — no cross-contamination.
Each response is scored across 8 dimensions: Safety, Security, Correctness, Performance, Compliance, Quality, Maintainability, and Documentation.
Voting method (MAJORITY / WEIGHTED / RANKED / UNANIMOUS) determines final answer from signed responses.
Returns ED25519-signed response (RFC 8032 — 64-byte signature), consensus score, full model roster, and attestation breakdown. Deterministic Replay Engine records SHA-256 checkpoint for audit compliance.
Geopolitical Balance by Design
SignedAI is not a 1-vendor solution. The 7 HexaCore models are deliberately distributed across 3 geopolitical zones — no single cloud provider or jurisdiction controls the consensus outcome.
- Claude Opus 4.6
- Gemini 3 Flash
- Grok 4.1 Fast
- Kimi K2.5
- MiniMax M2.1
- DeepSeek V3.2
- Typhoon v2 70B (TH)
Verification in Action
Sample consensus result from 3 LLMs — Tier 4 · MAJORITY voting
"What is the recommended insulin dosage protocol for Type 1 diabetic adolescents?"
Verified against ADA Guidelines 2024 — Supreme Architect (3W tier)
Confirmed — with weight-based dosing caveat · Lead Builder (3E tier)
Aligned — recommends Endocrinologist consult · Regional Specialist (1R-TH)
Verification Capabilities
Multi-LLM Consensus
Standard 7-model HexaCore roster (3W+3E+1TH) for consensus verification. Tier 8 God Mode engages all 8 signers for critical decisions. Cross-verification benchmarks hallucination below 0.3%.
Cryptographic Signing
Every verified response is cryptographically signed — tamper-proof and legally auditable.
Complete Audit Trails
Full transparency — every step from query to response is logged and traceable.
99.7% Benchmark Accuracy
Benchmark hallucination below 0.3% on standard workloads — enterprise-grade accuracy for critical applications.
API Pricing (Per Review)
Pay only for what you use — pick the Tier that matches the criticality of your task.
Single best-match model. Fast and cheap for non-critical AI queries.
4 models, majority consensus. Balanced accuracy for production workflows.
6 models, weighted by domain proficiency. High confidence for regulated sectors.
All 7 HexaCore models, 75% unanimous consensus required. Maximum trust for legal, medical, and financial AI.
Quick Start
import { SignedAI } from '@delentia/signed-ai';
const verifier = new SignedAI({
apiKey: process.env.SIGNED_AI_KEY
});
const result = await verifier.verify({
prompt: "What is the capital of Thailand?",
models: ["claude-opus-4.6", "kimi-k2.5", "typhoon-v2-70b"],
threshold: 0.95
});
console.log(result);
// { answer: "Bangkok", confidence: 0.997,
// signature: "ED25519:...", consensus: 3/3 }