agnosphere Fact Checker · Munich

The most dangerous error is the convincing one.

Upload the content of your choice. The agnosphere Fact Checker checks every claim against the source it cites, the web and the document’s own numbers. You get the errors, with the proof, on the annotated PDF.

For presentations, reports, strategy papers, memos and every document that needs to be right.

Start your check in 1 minute · 5,000 words free · no card

Stored in the EU · never used to train a model · data processing agreement available · Security and data handling

See a checked report
What happens to a report, in about a minute.
What the film says

By the time a report goes out, everyone has read it. No one has checked every single claim. We do. So you drop it into the agnosphere Fact Checker and press start. It pulls every factual claim out of the document and goes looking for the evidence behind each one: in the sources it cites, the open web, and your own internal data if you wish. The claims that don't hold up come back flagged. Red for a contradiction, amber for a discrepancy. Open the contradictions and there they are, each one with its proof. Click any one of them and the finding opens. At the top, what the report claimed. Below it, what the source actually says. And a correction you can paste straight in. Underneath is the verification trail: every source it opened and every search it ran. And one click takes you to the exact line in the original document. So you know exactly what to change. Or exactly what to challenge.

Check a report free

Start your check in 1 minute · 5,000 words free · no card

Our productAvailable now

agnosphere Fact Checker

One sentence, three claims,
one convincing error.

The agnosphere Fact Checker is fact-checking software for professional reports: it splits a report into its individual claims and checks each one against the source the report cites and against the open web. Below is the result for one sentence from the sample report. The margin notes show which claim is the hallucination and what the sentence should have said.

Sample report · Page 3

“AI-server power density increased elevenfold between 2020 and 20251 and is projected to rise another fourfold by 20272; one advanced rack could then consume as much electricity over a year as 65 households3.[6]”
  1. Verified

    The cited IEA source and web search agree.

  2. Verified

    Matches the projection in the same source.

  3. Wrong

    The source says peak power, not electricity used over a year.

    Suggested wording“one advanced rack could then have peak power demand equivalent to that of 65 households.”

That was one sentence. The full review covers all 206 claims in the report.

See the full annotated report(opens in a new tab)

Demonstration report. The source report contains intentionally introduced factual and chart errors. Do not rely on it for factual, investment, or other professional decisions. Created with generative AI. No client material was used.

Try it on your work

What would it catch in your report?

Start your check in 1 minute · 5,000 words free · no card

Check a report freeCannot upload here?

How we work

What happens to your report.

How we handle confidential reports, from the first request to the returned document.

The four rules

  1. No document moves automatically.

    Nothing is checked before you press start, and nothing is shared. You upload inside your own workspace, or over a private route we arrange with you.

  2. Make data use explicit.

    For each use case, we agree on access rules, data flows, retention, and model providers.

  3. Keep evidence and open questions visible.

    When decisions depend on sources, users can inspect the evidence and see what remains unresolved.

  4. Keep responsibility with people.

    Your team decides what can be automated, what must be reviewed, and who signs off.

Two founders run every engagement, from first call to checked report

Urs Steurer

Urs Steurer

Urs worked with enterprise customers at Google and in strategy at SAP. He makes sure technical decisions fit the workflow, its users, and the business case.

Jakob Schlegel

Jakob Schlegel

Jakob ran automation and data programs at Google and led analytics teams at adidas. At agnosphere he builds the AI systems and decides when they are good enough to go live.

Professional and academic experience at

Beyond the Fact Checker

Custom systems, built by the team behind the Fact Checker.

Research pipelines that check their own sources. Document workflows that keep the evidence chain visible. Review steps fitted to one team’s rules. Each of these systems is built for a single team, by the two founders on this page, under the same rule: tested on real cases before anything goes live.

or email Urs directly

Discussing your requirements

Someone will find the wrong number. Better it’s us.

In a no-obligation 15-minute call, we develop a shared understanding of the problem. We then determine whether the Fact Checker, a custom system, or a simpler solution is the best next step.

15 minutes · free and no obligation · no preparation needed

Prefer not to book a call?

Our email address is in the legal notice.

A founder reads every note and replies personally.