Three short clips show the application in action: from the boot-up screen to transcripts, TTS playback, and AI chat with exam questions.
The clips show 'Attention Is All You Need' (Vaswani et al. 2017, arXiv:1706.03762). Excerpts are shown under § 42f UrhG (right of quotation) for educational purposes. Full attribution: /en/credits.
PDFEnhancer is an AI-powered tool that simplifies even the most complex scripts and textbooks so much that you can listen to them on a treadmill at the gym or out on a jog. The application is meant for self-learners, for students preparing for an exam, and for instructors who want to encourage their students towards more independent and barrier-free study. Long scripts and dense slide decks on their own are often not enough to really understand a topic, and not everyone wants to absorb knowledge only as text. PDFEnhancer therefore provides a 24/7 AI companion built directly into the PDF that answers content questions and explains connections. The features include per-slide generated transcripts at different difficulty levels, structured like a personal lecture. There are also automatically generated summaries and exam questions on the material, a built-in AI chat with the full context of all slides, and a text-to-speech playback of transcripts and summaries with automatically running subtitles, so that the content can be consumed hands-free, for example during a workout. Another feature is the optional recording, with which instructors can record their own lecture voice and provide it to their students as a complementary slide transcript. The audio and podcast output also serves accessibility, for example for students with visual or mobility impairments who cannot regularly attend in-person classes. In future versions, downloadable podcasts on slide content are planned, along with annotations directly on the slides, a cloud synchronisation, and a dashboard for the management of documents and accounts. The feature set for mobile devices is also planned to be expanded over time.
PDFEnhancer is offered in three tiers. The Free tier (BYO, i.e. bring your own API keys) runs entirely on the user's own device: no signup, no cloud storage, and all features available as long as the user supplies their own OpenAI and Anthropic API keys. The Pro tier (currently work in progress) is the managed version: API keys included, cloud sync between devices, and a real account, backed by a maintenance and support agreement. The Academia tier is a one-time licence for universities and other educational institutions, with an optional maintenance subscription on top for updates and a guaranteed response window. The Free tier sits at the bottom on purpose. Anyone can try every feature of the application without giving us any money or personal data, and without trusting our servers to hold their lecture material. We only ask for money when we are demonstrably taking work off the user's plate, in the Pro tier that handles keys and sync, or in the Academia tier where an institution wants a managed deployment. The Pro tier is marked as work in progress because we will not build cloud sync infrastructure until we see real demand for it; selling features that do not exist yet would be dishonest. The Academia tier exists for a specific reason. It has become almost inevitable that students upload their lecturers' slides, scripts, and resources to public AI platforms; once that happens, the content is out of the lecturer's control, with no traceability, no enforceable copyright protection, and no clear path back from a foreign AI provider. A self-hosted PDFEnhancer changes that picture: a university can deploy the application on its own infrastructure and swap OpenAI and Anthropic for an in-house AI provider of its choice. With that setup, the institution keeps full transparency over how slides and study material are processed, and it can apply a strict policy on general-purpose AI use while still giving students one sanctioned tool that is fully under the university's control.
The application is planned to be released under AGPL-3.0 after launch. AGPL was a deliberate choice over more permissive open-source licences like MIT or BSD, because copyleft closes the SaaS loophole that the permissive licences leave open: a competitor cannot take this code, close-source it, and resell it as a hosted service without giving their modifications back to the community. The hardest engineering work in PDFEnhancer (the Whisper, GPT, and Claude orchestration, the audio pipeline, the slide-tracker, the per-slide transcript merging logic across multiple sessions) is exactly the kind of thing a competitor would skim if the licence allowed it. AGPL prevents that. There is also a fairness argument. Modern AI is built on open source. The language models that helped co-write this codebase were trained on the work of the global open-source community: on top of libraries like PyTorch and NumPy, and on top of countless freely licensed repositories of human-written code. Since PDFEnhancer is an AI-co-written piece of software, it would feel wrong to keep the result closed. The fair outcome is to give it back: free to use, free to read, free to contribute to. In practice, this means anyone can read the source code, audit how user data is actually handled, audit how the AI is prompted and constrained, run the application on their own infrastructure, fork it, or contribute fixes. Universities and other educational institutions can pair this with the Academia tier story described above for a fully self-hosted and audited deployment that they control end to end.
PDFEnhancer's privacy posture comes down to three commitments. First, no content surveillance: we do not read the PDFs that are loaded, the transcripts the application generates, the summaries, the chat history, the exam questions, or any audio recordings. None of that material reaches our servers, and we have no business reason to look at it. Second, no tracking on this website: zero cookies, zero analytics tools, zero session recording; Cloudflare collects standard server logs (IP address, timestamp, URL) for availability and security of the website only. Third, all data is local by default: in the Free tier, everything stays in the browser's IndexedDB or local Electron storage, and in the Pro tier (work in progress) only the content that has been explicitly released will be synchronised, and even then only into Cloudflare R2 and D1 buckets whose use has been authorised in advance. The complete legal grounding for all of this lives on the privacy policy page: legal bases under the GDPR, the processor agreements with Cloudflare, OpenAI, and Anthropic, retention periods, and the data subject rights including the right to access, correction, deletion, and complaint to the Austrian data protection authority. The summary above is meant as a quick read; the binding version is the full text at /en/legal/privacy.
Three distinct trademark forms used to identify the application: a word component, a graphic mark, and a brand colour.
Word mark
Nice classes: 9, 42, 41
The brand name 'PDFEnhancer' in clean sans-serif type, without graphic or colour elements. The protected element is the word component only.
Filing plan: Planned: filing with the Austrian Patentamt and EUIPO after market launch.
Figurative mark
Nice classes: 9, 42, 41
Stylised document with a folded corner and three horizontal text lines, accompanied by a four-pointed sparkle in the bottom right corner. Used as a graphic identifier independent of the word.
Filing plan: Planned: filing with the Austrian Patentamt and EUIPO after market launch.
Colour mark
Nice classes: 9, 42, 41
The amber tone with the hexadecimal value #F59E0B, used consistently in the figurative mark (sparkle) and in the website's visual hierarchy (Pro tier highlight, 'Most popular' badge). The protected element is the colour shade in isolation.
Filing plan: Planned: filing with the Austrian Patentamt and EUIPO after market launch.
Availability checked on 17 May 2026: 0 hits in EUIPO eSearch, TMview, and see-ip (Austrian Patentamt).