Get EU AI Act ready —
prepare your Annex VI self-assessment with multi-model evidence material.
Quorum is an advisory toolkit that UK and EU AI providers use to prepare the internal documentation referenced by Articles 9, 12, 13 and 14 of the EU AI Act. Every consensus run produces a tamper-evident PDF evidence record, a disagreement matrix across 14+ frontier and open-source LLMs, and a SHA-256 hash-chained traceability log. Advisory only — not a conformity assessment under Regulation (EU) 2024/1689.
Why EU AI Act readiness is a 2026 problem you can't outsource
Article 16 enforcement begins 2 August 2026. For most Annex III high-risk AI systems, the conformity assessment is yours — internal, under Annex VI. Quorum produces the evidence material that fills the gap between “we ship AI” and “our documentation file is review-ready.”
Annex VI self-assessment is yours
For most Annex III systems (recruitment, credit scoring, education, public services), the conformity assessment is internal — the provider is on the hook to assemble the documentation file. A Notified Body only enters the picture for biometrics and certain safety components (Annex VII). Quorum is the toolkit that produces the per-query evidence record you put in that file.
Articles 12, 13, 14 demand evidence — not assertions
Article 12 requires automatic record-keeping with traceability. Article 13 requires transparency for deployers. Article 14 requires human oversight that catches automation bias. Quorum's SHA-256 hash-chained PDF records and 14+ LLM disagreement matrix give your reviewer something concrete to point at for each of those requirements — instead of paragraphs of prose.
The penalty side is real
Article 99 sets tiered non-compliance penalties: up to €35M or 7% turnover for prohibited practices (Art. 5); up to €15M or 3% turnover for provider-obligation violations (Art. 16); €7.5M or 1% for supplying incorrect information to authorities. SMEs receive reduced penalties under Art. 99(6). For most Annex III providers the relevant tier is the Art. 16 one — and under-documenting Articles 12-14 is the expensive failure mode.
How Quorum builds your readiness evidence
Three stages, async fan-out across 14+ LLMs, full tamper-evident traceability log per query. Drop-in for any AI pipeline you already operate.
Intelligent Router
Prompt analyzed.
Parallel Execution
Isolated processing across 14+ LLMs.
Semantic Consensus
Cosine similarity on embeddings — paraphrases count.
# pip install quorum-aifrom quorum import consensus# Fan out to every configured provider, score agreementresult = await consensus("Draft the SEC filing.")print(result.answer)print(f"confidence: {result.confidence:.0%}")print(f"cost: ${result.total_cost_usd:.6f}")# Or self-host: docker run -p 8080:8080 quorum-ai# Or BYOK CLI: quorum ask "..." --all
What the evidence looks like
Every consensus call produces a tamper-evident PDF record and a disagreement matrix you can drop straight into your Annex VI technical documentation file.
One prompt → 14+ frontier and open-source LLMs in parallel → semantic agreement scoring → tamper-evident PDF evidence record per query. Format designed for inclusion in your Annex VI documentation file.
Disagreement-matrix dashboard surfaces where your 14+ LLMs converge and where they don't — the exact “automation-bias risk” signal Article 14 asks human supervisors to monitor. Shipped weekly.
What ships today
FSL-1.1-Apache-2.0 source-available core, BYOK any backend. Honest scorecard — source on GitHub means you can verify every claim against the repo before you commit a budget line.
Vectorized Semantic Consensus
We don't just check for matching words. Quorum compares vector embeddings of each model's output using Cosine Similarity to ensure deep, semantic agreement before returning a binding answer to your users.
HSP Gate (Patent-Pending)
Patent-pending Hybrid Sovereign Protocol (PCT/US26/11908). For high-stakes evolution actions (promoting a checkpoint, A/B-test push, federated update), the gate routes the decision through an async human-approval webhook (HSP_GATE_WEBHOOK env var) before the function executes. Optional, off by default — fully Apache 2.0 without it.
EU AI Act Readiness Toolkit
Every query generates a SHA-256 hash-chained PDF record via reportlab — every model that ran, its weight, raw response, cost, latency. Designed as internal evidence material to help providers prepare documentation referenced by Articles 12 (record-keeping) and 13 (transparency) of EU AI Act enforcement (starts 2026-08-02). Advisory toolkit only — not a conformity assessment under Regulation (EU) 2024/1689. Final conformity assessment remains the responsibility of the AI system provider (internal, Annex VI) or a designated Notified Body (external, Annex VII).
Pricing — built for UK and EU AI providers
Advisory toolkit pricing. Annual contracts available. For individual developers and self-hosted use, see Quorum for Developers.
For teams preparing internal documentation for EU AI Act Annex VI self-assessment. Advisory toolkit — not a conformity assessment.
- 50,000 hosted queries/month across 14+ LLMs
- Per-query tamper-evident PDF evidence record (Art. 12/13 material)
- Disagreement-matrix dashboard (Art. 14 oversight material)
- Readiness gap-analysis report (advisory)
- 90-day traceability log retention · Priority support · SLA
Advisory toolkit only. Not a conformity assessment under Regulation (EU) 2024/1689. Sovereign Chain Ltd is not a Notified Body under Article 31.
Contact for Pro+ ReadinessOn-prem or single-tenant VPC for regulated workloads (fintech, healthtech, public-sector AI). Annex VI readiness toolkit + sovereign data handling.
- Unlimited requests, custom SLA
- On-Premise / VPC deployment in your jurisdiction
- Dedicated BYOK key management · SSO
- Procurement-friendly contract template (Sovereign Chain Ltd UK)
- Indefinite traceability log retention
Advisory deployment only. Final conformity assessment remains with the AI system provider or a designated Notified Body.
Contact EnterpriseThe Economics of Consensus
Absolute transparency. Infinite scalability. Zero markup.
BYOK — Your Provider Keys
Zero token markup. Quorum never proxies your API keys; the orchestration fee covers the engine only.
MoE Router (Functional)
Per-query-class routing picks the top 2–4 of 14+ configured providers. Heuristic Top-K threshold today; ML-trained router target v0.5.
Distillation Loop (Roadmap)
Cheap models learn from expensive consensus. Skeleton in v0.1.x; functional target v1.0.
{
"billing_mode": "BYOK",
"orchestration_fee": "£149/mo",
"providers": [
"$ANTHROPIC_API_KEY",
"$OPENAI_API_KEY",
"$GEMINI_API_KEY",
"$NVIDIA_API_KEY // 6 OSS models free",
"$REPLICATE_API_TOKEN",
"ollama://localhost // self-hosted Llama"
],
"router": "moe_top_k",
"observed_cost_per_query": "£0.008",
"hsp_gate": true // fail-closed by default
}