01
Method over opinion
Klyptra makes no political judgments. The scale measures language and framing — not whether a position is right. That separation is the foundation of the project.
About Klyptra
Why build a German bias tool?
English-language tools like AllSides or Ad Fontes Media have existed for years. For German sources, German discourse and the German language, a comparable tool is missing. Klyptra fills that gap — grounded in research and fully documented.
Name
Klyptra means revelation.
From the Greek kalýpto — to conceal. Klyptra turns the concept toward revelation: making visible what language hides. Not an AI buzzword, but a substantive choice.
Principles
Four guiding tenets behind every design decision.
02
Evidence, not assertions
Every score is backed by a verbatim quote from the source text. Anyone questioning a score sees the exact passage that led to the rating.
03
Models as tools, not arbiters
Language models have biases of their own. Klyptra compensates with a three-model ensemble and makes model disagreement visible.
04
Reproducibility
Code, data, prompts and model versions are openly documented. Anyone who doubts Klyptra's results should be able to recompute them.
Timeline
From idea to tool.
Q1 2026
Concept and methodology definition. The Spinde taxonomy chosen as the theoretical framework.
Q2 2026
Analysis pipeline and methodology foundation ready. Six dimensions validated against the BABE benchmark.
End of Q2 2026
Three-model ensemble with median aggregation and divergence flag in production.
Since June 2026
Closed beta launched. Research and newsrooms get access on request.
Q3 2026
Extended result features: PDF report export with AI transparency labeling.
2027
Methodology development and extended analyses.
Team & funding
Who builds it, who pays for it.
Team
Klyptra is developed as an independent solo project — from the methodology to the pipeline to the website.
Funding
The project is self-funded. Grant or research partnerships are being explored.
Contact