Veritas Chat.
What 5 AIs confirm is hallucination-resistant.
Ask your question simultaneously to 5 leading AI models. A leader model filters the consensus and hands over only what the majority agrees on. Disagreements are shown transparently — single hallucinations don't pass the filter.
Four pillars, one promise
Veritas Chat is not a model comparison — it is a stand-alone protection layer: anti-hallucination through consensus.
Anti-Hallucination Protection
Single AIs hallucinate — invented facts, fake sources, made-up numbers. When 5 independent models agree, the risk drops drastically.
Majority Decision, Not Single Opinion
You get the answer 3 out of 5 models agree on — disagreements are shown transparently, not hidden.
Full Transparency
Every one of the 5 raw answers stays expandable. You can verify the consensus against the individual statements at any time — before compliance, audit or board meetings.
Freely Configurable Setup
You pick 5 slots + 1 leader from 100+ EU models. Standard defaults set by superadmin, permanently customisable per user.
AIs invent facts. Veritas filters them out.
Studies show: even the best language models hallucinate in 5-15% of all factual statements — fake sources, wrong dates, made-up numbers. For compliance, research and strategy questions, that's a risk nobody wants to bear.
But when 5 independent models — trained by different providers, with different datasets and architectures — make the same statement, the probability that they all produce the same hallucination simultaneously drops dramatically.
- Statistical protection layer on top of normal chat
- Majority decision instead of blind trust in a single AI
- Disagreements become visible, not hidden — you decide
- Audit trail over all 5 raw answers — traceability for DPO, board, auditors
How Veritas Chat works
Five steps from click to consensus answer — all in a single chat surface.
- 1
Ask the question
You write your question like in a normal chat. One single input, one click.
- 2
5 models start staggered
One per second, 5 independent AI models start in parallel — Claude, GPT, Gemini, Mistral, DeepSeek or any other from your setup. Tokens stream live and visible.
- 3
Leader analyses the answers
A 6th model (your leader) receives all 5 raw answers and looks for substantive consensus — even when wording differs.
- 4
Consensus answer streams back
You see only what at least 3 models agree on. Disagreements appear separately under "Differing views" — transparent, neutral, no judgement.
- 5
Sub-answers expandable any time
One click opens the individual answers of all 5 models — for audit, compliance review, or simply to see how the consensus emerged.
Who benefits from Veritas Chat?
Anywhere a wrong AI answer would be costly — compliance, strategy, research, engineering.
Legal & Compliance
GDPR questions, contract clauses, AI Act interpretation — answers backed by consensus instead of a single opinion. Sub-answers are directly audit-ready.
Executives & Strategy
Strategic decisions, market analyses, risk assessments — if AI should act as your sparring partner, then better as the consensus of 5 than the opinion of one.
Research & Education
Subject definitions, historical facts, technical explanations — the consensus of 5 independent models replaces tedious source-checking for routine questions.
Engineering & Architecture
Tech decisions, best practices, tool comparisons — when four models recommend Postgres and one MongoDB, that's a strong signal. Veritas shows both.
Benefits at a glance
The 5th layer of the MultiLayer Data Shield
Veritas joins the protection architecture: privacy, content, manipulation, audit — and now anti-hallucination. A protection layer that doesn't filter the input but quality-controls the output.
View all 5 layers →Frequently asked questions on Veritas Chat
Answers on models, costs, data protection and configuration.
How many models actually run per Veritas request?+
Six models per request: five parallel sub-models that answer your question independently, plus one leader model that synthesises the consensus. Budget consumption is therefore about six times a normal chat request — shown as a percentage in the usage line under every consensus answer.
What happens when the models disagree?+
The leader looks for statements where at least 3 out of 5 models agree in substance — wording may differ, content counts. These statements form the consensus answer. Disagreements are listed separately under "Differing views", neutral and without judgement. If fewer than 3 usable answers exist, it is shown transparently ("Consensus based on N models").
Can I pick the 5 models myself?+
Yes. The superadmin sets 5 default models that are pre-filled for new users. Each user can then pick a different model per slot from the full active chat catalogue (100+ EU models) and make one of them the leader. The slot configuration is persistently stored per user.
How long does a Veritas request typically take?+
Between 20 and 90 seconds — depending on the chosen models, question complexity and provider load. Fast models (Mistral, GPT-Mini, Gemini Flash) deliver in 5-10 s, reasoning models (Claude Opus, GPT-Pro) take longer. You see staggered live progress for all 5 slots.
Are my Veritas requests used for AI training?+
No. All integrated providers have contractually committed that neither prompts nor outputs are used for model training. Veritas follows the same protection as the normal HOVIGuard chat — EU-hosted throughout, DPA in the contract.
How long are Veritas conversations stored?+
Seven days. Veritas messages are automatically deleted from the database after 7 days (cron job, idempotent). Audit-log entries are retained longer (see retention obligations). This retention is intentionally shorter than for the normal chat because Veritas requests generate 6 model calls and storage load is correspondingly higher.
Are the 5 raw answers visible to me?+
Yes. Below every consensus answer there is a "Details on the analysis" button. One click opens an expandable panel with all 5 individual answers, each with model brand, status (OK/timeout), latency and token consumption. For audit purposes, sub-answers are additionally persisted in the message metadata.
Can my admin disable Veritas for my team?+
Yes, in three stages: the platform superadmin can switch Veritas off globally via kill-switch. The company admin can disable Veritas for the entire tenant (tenant default) or per individual user. By default, Veritas is active for all invited users.
Which models make sense as leader?+
A strong reasoning model with a large context window. Claude Sonnet/Opus, GPT-Pro and Gemini Pro work very well in practice because they recognise semantic equivalence even with differing wording. Fast mini models suit the 5 sub-slots better than the leader role.
What does Veritas Chat cost extra?+
Veritas is included in the Pro and Enterprise plan at no additional cost — you use your existing fair-use budget. Since a Veritas request generates 6 model calls, it consumes about six times a normal chat call. If your daily budget gets tight, you see this as a percentage under each request and can continue via overage activation or wait.
Try Veritas Chat yourself
7 days free — all features unlocked, including Veritas Chat.
Start Free Trial