IamVERA Review: Multi-Model AI Answer Verification for Confidential Professional Work (2026)

Sponsored post. Product details checked against iamvera.ai, its evidence page, pricing, llms.txt, the IamVERA GitHub README and the sign-up console on October 8, 2026. Screenshots taken from the live sites.

Quick answer: IamVERA is a verification layer that sits on top of other companies' AI models. You ask one question. Claude drafts an answer, GPT audits the facts, Grok pushes back on the reasoning, Perplexity checks live public sources, and Claude writes the final version. Every correction, disagreement, source and cent of model cost stays visible. For confidential files, a Privacy Shield swaps out names, dates and case numbers before anything reaches an external model. It is made in the Netherlands for lawyers, notaries, compliance teams, healthcare professionals and journalists, and the first month is free.

Anyone who uses AI at work has seen it happen. You ask a sensible question, you get a fluent, confident answer, and one detail in it is wrong in a way you only notice because you happened to know. For a blog draft that is an annoyance. In a client memo, a board paper or a medical report, it is a liability.

IamVERA's homepage puts it in one line: "One AI model can be convincingly wrong. Verify its answer before you act." This review looks at how that verification works, what the company has published about its own results, how it handles confidential documents, what it costs, and who will get the most out of it.

Key takeaways

• IamVERA routes one question through Claude, GPT, Grok and Perplexity, each with a fixed job, and shows the full trail behind the final answer.
• Disagreements between models are shown to you instead of being averaged away.
• The Semantic Privacy Shield replaces sensitive values with synthetic ones before external AI processing and sends nothing if its outbound check fails.
• Vera X-Ray, a Chrome extension, moves detection and review onto your own device.
• Vera Office lets the chain summarise and edit DOCX files through a validated, limited write route.
• IamVERA publishes its own test results with raw JSON, including the one check that failed.
• Pricing runs from a free first month (10 runs) to €399 per month for teams of 25, priced per verified question rather than per token.
• The company is careful with its claims: no truth guarantee, no anonymisation claim, and professional judgement stays with you.

What is IamVERA?

IamVERA, also written as I am Vera or just Vera, is an AI answer verification console for professional work. It is built and run by The Coding Company B.V., an IT architecture and cybersecurity firm in The Hague, and was founded by Victor Angelier, who has led that company since 2008.

It is useful to be clear about what Vera is. It does not have its own language model. It coordinates existing models through OpenRouter and gives each one a role, so that no single model both writes and grades its own answer. The company's llms.txt describes the category as "multi-model AI verification, not a fixed combination of model brands," which means the specific models can change over time while the method stays the same.

One naming note so the two are not mixed up: IamVERA has nothing to do with vera.ai, the EU Horizon research project on AI-assisted verification. The company says so on its GitHub page.

How does the multi-model verification chain work?

Each model gets one job, and each job is recorded. With the full chain selected, a run looks like this:

Step	Model	Role
1	Claude	Drafts the initial answer
2	GPT	Reviews each factual claim for errors and missing support
3	Grok	Challenges assumptions, framing and reasoning
4	Perplexity	Checks claims against current public sources
5	Claude	Incorporates justified findings and writes the final answer

The open-source README adds a detail the marketing page leaves out. Claude revises after the factual review and again after the challenge, so a full run can be seven model calls. When a reviewer reports nothing to fix, the matching revision is skipped. The reviewer call itself still runs and is logged, so you can see that the check happened. In IamVERA's own factual-trap tests, 18 revision steps were skipped this way, which keeps cost and waiting time down on easy questions.

You choose the depth per session. GPT, Grok and Perplexity can each be switched off. Claude stays on, because it drafts and writes the final synthesis.

The company's favourite example is a good one. Ask when the GDPR "came into effect" and a single model will usually say 25 May 2018. In Vera's run, GPT confirms the date and adds the regulation number. Grok then points out that the GDPR was adopted on 27 April 2016, entered into force on 24 May 2016 and became applicable on 25 May 2018, and that a lawyer would care about the difference. Perplexity confirms all three dates on EUR-Lex. The final answer carries the distinction and the source.

That example shows the design idea well. Grok's job is to disagree, and the interface keeps its disagreement in view. The site calls this a feature, and for professional work it is the useful part: you see where the models differed and can decide for yourself.

What does one verified answer cost to run?

Very little at the model level. IamVERA prints an approximate model cost for its GDPR example of about $0.011 for the full chain:

Harder questions cost more because they take more calls and more web lookups. In the published test runs, single questions ranged from about $0.02 with four model calls to about $0.13 for the hardest ones, which took five to seven calls and live source searches. Those are provider costs. Your subscription is billed per verified question (a "run"), not per token.

How does IamVERA protect confidential documents?

With a gate that fails closed. Before document content goes to any external model, the Semantic Privacy Shield runs three steps inside your isolated Vera environment:

1. Detect and pseudonymise. Names, dates, locations, case numbers, medical details, roles, family links and allegations are replaced with realistic synthetic values. Relationships stay intact, so the model still understands that "person 1" is the mother of "minor child 1."
2. Verify the outbound text. A local check confirms no detected real value is left. If the check fails or the Shield is unavailable, nothing is sent, and there is no override.
3. Restore locally. When the answer comes back, the original values are put back inside your environment. The mapping between real and synthetic values never goes to the providers.

IamVERA is precise about the limits here, which a careful buyer will appreciate. The Shield is pseudonymisation, not anonymisation. Detection can miss subtle context, and a unique combination of the remaining facts can still point to a person. "Local" means inside your EU-hosted Vera environment, not necessarily on your laptop. The approved, transformed context does go to the selected providers through OpenRouter, and the site states that those providers may process it outside the EU.

Uploaded chat PDFs are processed in memory for the active run and are not kept as reusable files, although run metadata stays in your session history.

What is Vera X-Ray?

Vera X-Ray moves the privacy decision one step earlier, onto your own device. It is a Chrome extension that reads a PDF, TXT or Markdown file locally, finds sensitive values, shows you what it found and lets you decide what Vera receives.

A few design choices stand out:

• Detection starts with deterministic rules and checksums for anything with a fixed shape, such as IBANs, Dutch BSN numbers, phone numbers, postcodes, licence plates, case numbers, passwords and keys.
• For names, organisations, places, health information and special categories of personal data, you can install an on-device GLiNER2 model that runs in the browser through WebGPU or WASM. It is a one-click opt-in. Model files are checked against pinned SHA-256 digests, and document text is never sent to the model host.
• Scanned pages without a text layer go through local OCR.
• Approved values become opaque, random, document-bound tokens instead of fake identities. Every occurrence gets its own 128-bit random part, and numbering restarts for each document, so there is no identifier that links documents to each other.
• Passwords, PINs and keys are removed outright, not stored as reversible values.
• Saved mappings are encrypted, can be PIN-protected and lock after inactivity.

X-Ray works in Dutch and English and runs on Chrome; Firefox is not a production target yet. Like the Shield, it does not claim complete detection, so the human review step is part of the design. It is also not included in the free Starter plan.

What can you do with Vera Office?

Vera Office brings the chain into a browser-based DOCX editor, so you can work on a document without copying passages into a chat window and pasting answers back.

The most thoughtful part is that reading and writing take different routes:

Questions like "Which obligations rest on the contractor?" go through the full verification chain. Edits like "Rewrite this passage more formally" are turned into a limited action plan, validated, and only then applied and saved. The model never gets free-form write access to your file. Before any task, Vera saves the current version, so it never analyses a stale copy.

Every run leaves a trace you could hand to a colleague or an auditor:

Vera Office is a separate browser workspace, not a Microsoft Office add-in. Documents stay in your tenant environment because editing and version history need storage. It is included from the Essential plan up.

What does IamVERA's published evidence show?

IamVERA publishes its own behavioural test suite on an evidence page, together with the raw JSON for every run. That is uncommon in this market, and it is worth reading even if you never buy the product.

The first suite, dated 3 July 2026, has ten factual traps. The questions contain a false premise and invite the model to play along: why Einstein got the Nobel Prize for relativity, why the Great Wall is visible from the Moon, how goldfish cope with a three-second memory, and how much Everest has grown at "exactly four centimetres a year." The chain corrected the premise every time.

The trap that matters most for legal users asked for a summary of a 2019 Dutch Supreme Court ruling, "Jansen/Van der Berg", on digital evidence. That ruling does not exist. The chain said it could not trace the ruling and would not summarise it as if it knew the content. It listed the nearest real rulings with their ECLI numbers and asked for a reference so it could search properly. That is the behaviour you want from a research assistant.

The second suite has ten source-grounding and prompt-injection cases, scored by deterministic checks and by a separate language-model judge. Nine of ten passed the strict deterministic checks. The one that did not was a RAG-boundary case, where a source document contained a hidden instruction. The judge scored the answer as correct (it ignored the instruction and flagged it), while the deterministic check marked it as a fail. IamVERA shows both results instead of picking the better one.

The company's own caveats belong in any honest summary, and they are printed right on the page. These are tests of its own chain on its own harness, not a peer-reviewed benchmark. Samples are small, there is one run per case, results can change with model versions, and the judge is itself a language model. The runs were done in Dutch with English case titles. In IamVERA's words, the tests show observed behaviour, "not a guarantee of future accuracy." Read them as a transparent demo of how the chain behaves, and then try your own hard questions during the free month.

How much does IamVERA cost?

There are four plans, billed per verified question. Prices exclude VAT, you can cancel monthly, and yearly billing gives two months free.

Plan	Price	Runs	Users	What it adds
Starter	Free first month	10 in total	1	About 400,000 tokens. No Vera Office, API, reports, audit trail, support or X-Ray
Essential	€79/month or €790/year	50 per month (about €1.58 each)	1	About 2M tokens, Vera Office
Professional	€199/month or €1,990/year	300 per month (about €0.66 each)	5, extra user €19.95 adds 100 runs	About 12M tokens, DPA on request, Vera Office, 20% off extra runs
Compliance	€399/month or €3,990/year	1,000 per month (about €0.40 each)	25, extra user €24.95 adds 100 runs	About 40M tokens, 250 GPU-processed documents, DPA included, priority support, API, PDF verification reports, 40% off extra runs

A run is one question through the chain, whether it takes four model calls or seven. When you run out, verification stops with a message on screen. Nothing is charged automatically, and you can wait for the new month, buy extra runs or move up a plan.

If you saw IamVERA's launch coverage in July 2026, the early-access prices announced then (from €24.95 per month for independent professionals) have since been replaced by the grid above.

Our take on spend: use the free month to test the chain on questions from your own field. A solo lawyer or adviser who mainly needs checked answers and occasional document work fits Essential. Small firms that want shared seats and a data processing agreement will look at Professional. Compliance is for teams that need the API, the PDF verification report to hand over, and the GPU document capacity.

How do you run your first verified answer?

It takes about fifteen minutes from sign-up to your first trail.

1. Create your environment. Go to saas.iamvera.app, enter your name and email, choose an address under .iamvera.app and confirm the link IamVERA emails you. You get your own environment with its own storage and users. No credit card is needed for the free month.
2. Keep the full chain switched on. Leave Claude, GPT, Grok and Perplexity enabled for your first runs so you can see what each role adds.
3. Ask a question you already know the answer to. Pick something from your own field with a known trap: a date with several legal meanings, a ruling you know does not exist, or a statistic people usually get wrong.
4. Open the verification steps. Expand the trail under the final answer. Read what GPT confirmed, what Grok challenged, which sources Perplexity found and what Claude changed as a result.
5. Check the cost and decide your setup. Look at the tokens and cost per model call, then decide which checks each type of question deserves. For document work, Essential or higher gives you Vera Office.

Ten runs go quickly, so plan them. Two or three trap questions, a few real questions from this week's work and one test of a disagreement between models will tell you more than ten random prompts.

Who is IamVERA built for?

People whose AI output gets acted on rather than just read. The site's examples are specific:

• Lawyers and legal advisers reviewing a draft analysis for unsupported claims, or checking whether cited sources hold up before the argument goes out.
• Notaries giving a draft deed or explanatory note an extra check for inconsistent wording and missing information.
• Compliance and security teams checking policies against requirements, with a trace of the AI processing for later review.
• Board members reading a report or proposal for weak assumptions before it informs a decision.
• Doctors, occupational physicians and healthcare professionals structuring notes or reports. IamVERA is explicit that this is documentation support, not diagnosis or treatment.
• Journalists and researchers who need a claim checked against live sources, with the sources listed.
• Estate agents checking property descriptions and agreement summaries for inconsistencies before they go to a client.

It is a weaker fit if you want a fast general chatbot for casual tasks. A verified answer takes longer than a single reply, and a run budget makes more sense for questions that matter.

What should careful buyers check?

IamVERA is open about most of its limits, which makes this section short.

• No truth guarantee. The company repeats this everywhere, and you should plan your review process around it. The chain lowers the chance that an error goes unnoticed. It does not remove it.
• External providers still see transformed context. Pseudonymised text goes to the providers you select (Anthropic, OpenAI, xAI and Perplexity in the full chain) through OpenRouter. If your rules forbid any third-party processing, even of pseudonymised data, check that with your data protection officer first.
• Detection is not complete. Both the Shield and X-Ray say they can miss indirect identifiers. Keep the human review step.
• Certificates. The homepage lists ISO/IEC 27001, ISO 9001, ISO/IEC 42001 and NEN 7510 under "Independently certified." The public pages we checked did not name the certifying body or certificate numbers, and the site says it is still completing its GDPR processor documentation. If certification matters to your procurement, ask for the certificates and the DPA; IamVERA offers a DPA on Professional and includes one on Compliance.
• Evidence is self-published. The test suite is transparent and comes with raw data, but it is small and run by the vendor. Your own trial questions are the better test.
• A young, small company. IamVERA launched in 2026, and we did not find named customer case studies or independent user reviews yet. For a tool that handles client files, a call with the founder is a reasonable step before rolling it out to a team.

For background on what the regulation actually asks of AI products in Europe, our guides to the EU AI Act for SaaS founders and the 2026 SaaS compliance checklist cover the obligations in more depth.

Is IamVERA worth it?

Yes, for professionals who already use AI on work that carries consequences and want a structured, visible second opinion before they act.

The core idea is sound and well executed. One model grading its own answer will repeat its own blind spots. Giving the drafting, fact-checking, challenging and source-checking to different models, and showing every step, turns "the AI said so" into something you can inspect and defend. The privacy design is careful in the right places: a fail-closed gate, synthetic values that keep relationships intact, an on-device option through X-Ray, and a write route for documents that never hands a model free control of the file.

What made the strongest impression is the tone. IamVERA tells you what it does not do as clearly as what it does. It says it gives no truth guarantee, that its privacy layer is not anonymisation, and that providers may process data outside the EU, and it publishes the check that failed next to the ones that passed. For buyers in law, healthcare and compliance, that candour counts in its favour.

The easiest way to judge it is to put your own hardest questions through it. Start your free month on IamVERA with 10 runs and no credit card, read the published evidence first, and compare Vera Office and Vera X-Ray if your work involves confidential documents.

FAQ

What is IamVERA?

IamVERA (also called I am Vera or Vera) is an AI answer verification layer for professional work, built by The Coding Company B.V. in The Hague and founded by Victor Angelier. It is not its own language model. It sends one question through several external models with separate roles: Claude drafts, GPT audits the facts, Grok challenges the reasoning and Perplexity checks live public sources. Claude then writes the final answer, and every check, correction, disagreement, source and cost stays visible.

How does the IamVERA verification chain work?

With the full chain selected, Claude writes the first answer, GPT reviews its factual claims, Grok challenges assumptions and reasoning, and Perplexity checks current public sources. Claude revises after the reviews and writes the final synthesis. Every selected reviewer runs and is recorded even when it finds nothing, and revision steps are skipped when there is nothing to fix. GPT, Grok and Perplexity can be switched off per session; Claude stays on.

Does IamVERA guarantee that an answer is true?

No, and the company says so on its homepage, FAQ, about page and llms.txt. A verified answer means it went through independent factual audit, adversarial challenge and source checking, with the trail shown. IamVERA describes the benefit as a lower risk of undetected AI errors, with professional judgement staying with the user.

Is IamVERA safe for confidential legal or medical documents?

It is built for that use with two privacy layers. The Semantic Privacy Shield replaces detected names, dates, case numbers, medical details and similar values with synthetic ones inside the customer's EU-hosted Vera environment, checks the outbound text and sends nothing if the check fails. Vera X-Ray, a Chrome extension, does detection and human review on your own device before anything reaches Vera. IamVERA states that this is pseudonymisation, not anonymisation, and that transformed context still goes to external providers through OpenRouter, which may process it outside the EU.

How much does IamVERA cost?

Starter is free for the first month with 10 verification runs and no credit card. Essential is €79 per month (50 runs, 1 user, Vera Office). Professional is €199 per month (300 runs, 5 users, DPA on request). Compliance is €399 per month (1,000 runs, 25 users, API, PDF verification reports, priority support and a DPA). Yearly billing gives two months free, and prices exclude VAT. One question through the chain counts as one run, however many model calls it takes.

What is Vera X-Ray?

Vera X-Ray is IamVERA's client-local privacy layer, available as a Chrome extension. It reads PDF (including scanned pages through local OCR), TXT and Markdown files in Dutch and English, finds sensitive values with deterministic rules and an optional on-device GLiNER2 model, and lets you review and approve what is released. Approved values become random document-bound tokens, and the originals are restored on your device after Vera answers.

What is Vera Office?

Vera Office is a browser-based DOCX workspace inside IamVERA, included from the Essential plan. Read tasks such as summaries and risk reviews run through the verification chain. Write tasks such as replacing a term or rewriting a clause become a limited action plan that is validated before it is applied, so a model never gets free write access to the file. Each run records the document version, snapshot hash, privacy result, models used and cost.

Is IamVERA the same as the EU vera.ai project?

No. IamVERA is a commercial product of The Coding Company B.V. The EU Horizon research project vera.ai (VERification Assisted by Artificial Intelligence) is a separate, unrelated project, and IamVERA states this on its GitHub page.

Who should use IamVERA?

Professionals who act on AI output in work where a confident mistake has a cost: lawyers and legal advisers, notaries, compliance and security teams, board members, journalists, researchers, estate agents, and doctors or occupational physicians using it for documentation. It is not intended for diagnosis or treatment decisions and does not replace professional review.

If you are building an AI product yourself, SaaSCity is a free, human-reviewed startup directory with a live city map where a listing is a permanent page other founders actually see. Submitting takes about ten minutes.