AI transcription, processed exclusively in Europe.

Klangskript transcribes and translates your audio and video data with state-of-the-art AI — no US cloud, no permanent storage, no compromises on data protection. Built for those who know exactly where their data needs to be.

Get in touch
Processing exclusively in German and European data centres · volatile memory only · 100 % European infrastructure

Klangskript

Sicher. Europäisch. Verständlich.

Zero Data Retention Your data is never stored
100% Europe No US cloud, no CLOUD Act
§ 203 StGB compliant Suitable for professional secrecy obligations
End-to-end encrypted WireGuard tunnels between all components
GDPR Art. 28 Data processing contractually guaranteed
Own content only No third-party platform downloads

Who Klangskript is built for

Companies

International team meetings, internal discussions and minutes — without US cloud services.

Law firms

Client meetings, witness statements and hearing recordings under attorney confidentiality.

Medical practices

Communication with non-German-speaking patients: transcription and translation of the conversation.

In development — reservations accepted.

Individuals

Lectures, interviews and your own recordings — for students, creators and freelancers.

Our architecture is uncompromisingly European — and technically verifiable.

Klangskript was built from the ground up as a sovereign alternative. Three principles define our infrastructure.

100% European infrastructure

Not a single byte leaves European soil.

Our orchestrator server runs with a German provider. The AI compute nodes are booted exclusively in German and European data centres — with providers operated entirely within Europe. Neither our infrastructure nor our subcontractors are subject to US law or the CLOUD Act.

Zero Data Retention through ephemeral processing

Your audio data never touches a hard drive.

A dedicated GPU instance is spun up for each session. AI model processing takes place exclusively in the volatile memory (RAM) of this instance. After the session ends, the memory is actively overwritten and the instance is irrevocably destroyed. No copy, no log file, no training dataset exists.

End-to-end encrypted transport

Military-grade encryption — across the entire chain.

Data transfer between your device, our orchestrator and the AI compute nodes uses dynamically generated WireGuard VPN tunnels. Each session uses its own keys. No third party — not even us at transport level — can view the transmission.

What we do — and what we never do.

At Klangskript, privacy isn't a setting. It's the architecture. Two sides of the same decision.

What we do

  • Processing exclusively in European data centres
  • § 203-suitable — contractually assured in the DPA
  • Data processing under Art. 28 GDPR
  • End-to-end encrypted between all components
  • Zero data retention — volatile memory only

What we never do

  • No US cloud, no CLOUD Act access
  • No US AI services
  • No tracking, no analytics
  • No selling or sharing your data
  • No AI training on your data

Verifiably tracker-free.

No trackers, no cookies, no analytics — and you don't have to take our word for it. A daily automated scan shows it, and an independent third party re-checks it on demand.

See the transparency page

Meetings in a foreign language? No more comprehension gaps.

Anyone who regularly participates in English-language meetings without being a native speaker knows the situation: an unfamiliar technical term comes up, a moment of uncertainty arises — and while you are still thinking about the meaning, the speaker is already three sentences ahead. The gap grows, the context is lost.

Available today: Post-meeting transcription and translation

Internal meetings are already routinely recorded in many companies. Klangskript enables you to upload these recordings afterwards and have them fully processed:

  • Transcription of the entire meeting using faster-whisper large-v3 (90+ languages)
  • Speaker recognition: Automatic attribution of who spoke when
  • Translation into your own language — either hallucination-free (basic) or context-aware (premium)
  • Result: A complete, searchable protocol in your own language

Coming soon: Live translation during the meeting

For situations where comprehension gaps cannot wait until after the meeting, we are developing live translation via the browser: microphone access on your device, real-time streaming to Klangskript, translation with under two seconds latency — read along on screen while the meeting is happening.

Availability will be announced separately.
Many employees are already resorting to their own solutions: consumer translation services on personal devices, headphones with built-in simultaneous translation, transcription apps on smartphones. From the company's perspective this means: confidential conversation content flows uncontrolled to US-based services, without a data processing agreement, without the IT department's knowledge, without deletion guarantees. Klangskript offers a controlled alternative.

Trade secrets, professional secrecy and the DPA — why architecture alone isn't enough.

Confidentiality cannot be delegated away. Whoever hands trade secrets or client data to a service provider — for transcription, translation or processing — remains responsible for ensuring that provider is bound by the same confidentiality. For companies this follows from the German Trade Secrets Act (GeschGehG) and Art. 28 GDPR; for those bound by professional secrecy, from § 203 StGB and § 43e BRAO. This is precisely the obligation US providers cannot give, because their parent companies are subject to the US CLOUD Act.

Companies and their trade secrets

Strategy discussions, personnel decisions, development projects, M&A considerations: under § 2 No. 1 GeschGehG, a trade secret is only legally protected as long as the company takes reasonable confidentiality measures. Having recordings of internal meetings processed by uncontrolled US services jeopardises exactly this protective requirement — and, without a data processing agreement, additionally violates Art. 28 GDPR. Klangskript processes within an openly documented European chain, with a DPA and without CLOUD Act exposure.

Lawyers, notaries and tax advisors

Attorney confidentiality under § 43a (2) BRAO and § 2 BORA is at the heart of the client relationship. § 43e BRAO governs the conditions under which firms may engage service providers: the provider must be obligated to confidentiality in text form. Klangskript meets this requirement across the entire subcontractor chain — flanked by § 53 StPO, which additionally protects client communications from seizure.

Why US cloud and US AI models are ruled out

Even when a US provider operates European servers, its parent company is subject to the US CLOUD Act and can be compelled to hand over data — regardless of storage location. An effective commitment declaration under § 203 StGB or § 43e BRAO therefore cannot be concluded. The same applies to generative AI services from US providers to which content is sent via API. For those bound by professional secrecy, both routes are effectively closed.

How Klangskript meets the requirements

  • Processing exclusively in European data centres — no US corporation in the chain, no CLOUD Act access
  • Zero data retention: processing only in volatile memory, actively overwritten after the session — no disk persistence, no training dataset
  • Commitment declaration under § 203 StGB and § 43e BRAO across the entire sub-processor chain, fully documented
  • Data processing agreement under Art. 28 GDPR, prepared and ready to sign

We provide the DPA template, commitment declaration and architecture description for review by your data protection officer.

Request the compliance documentation

Klangskript serves two worlds with the same infrastructure.

Whether you are a student transcribing a lecture or a company reviewing an international team meeting: the underlying security architecture is identical. What differs is the feature set and billing model.

For students, content creators and freelancers

Fast, precise transcription and translation without legal compliance requirements — but with the same level of data protection as our enterprise solution.

  • File upload: Upload audio and video files, process in the background, retrieve results
  • High transcription quality: faster-whisper large-v3 as the standard model
  • Hallucination-free translation: Our base variant uses classic translation models (Opus-MT) instead of generative LLMs — your transcript is translated, not rewritten
  • Premium option: Structured summaries, chapter marking and filler word removal using European-licensed open-weight models
  • Pre-paid model: Buy credit in advance, billed per audio second, no surprises
  • No subscription, no minimum term

Subtitling YouTube videos and podcasts · Transcribing lectures, seminars and interviews · Dictation and voice note transcription

For companies and law firms with confidentiality and compliance requirements

Booked sessions with a dedicated GPU instance, highest model quality and contractually guaranteed § 203 StGB compliance.

  • Dedicated session booking: Reserve timeslots with 10 minutes lead time — exclusive GPU instance for your session only
  • Two-phase processing: Live transcription during the conversation (latency under 2 seconds), high-precision post-processing with full conversation context after the session
  • Two modes: Precision (recommended, post-processing with maximum quality) or Simultaneous (live translation for real-time communication)
  • Structured outputs: Automatic generation of meeting minutes, case notes or interview protocols from templates
  • Premium AI models: Gemma 4 26B MoE for translation and structuring, faster-whisper large-v3 for transcription, pyannote 3.1 for speaker recognition
  • Contractual guarantees: Data processing agreement under GDPR Art. 28, § 203 StGB commitment declarations for the entire subcontractor chain
  • SLA-backed subscriptions: For companies and law firms with regular demand

Confidential board and team meetings, including foreign-language ones · Client meetings, witness statements, court transcripts · Conversations with international partners, candidates or customers

The B2B pipeline is currently in pilot operation. We onboard selected companies and law firms personally.

Medicine: doctor-patient conversations — in development

For communication with non-German-speaking patients, we are developing a solution with additional clinical quality assurance. It supports the conversation — it replaces neither the doctor nor the interpreter. We will announce availability and feature scope once our quality evaluation and regulatory assessment are complete.

Reservations are being accepted.

Investigative journalism: On-premise appliance

For newsrooms with source protection requirements under § 53 StPO, we are developing an on-premise solution: the entire Klangskript infrastructure as a pre-configured appliance that runs directly in your editorial network. Audio never leaves your premises.

Available in Phase 3. Reservations are being accepted.

How Klangskript works technically.

We believe trust is built through transparency. That is why we describe our architecture openly.

The Orchestrator

  • Location: Germany (currently Nuremberg)
  • Technology stack: React, FastAPI, PostgreSQL, Redis
  • Function: User management, billing, cloud resource orchestration
  • Persistent data: exclusively metadata (account balance, session IDs, billing data) — never audio or transcription content

The AI compute nodes

  • Location: German and European data centres only
  • Hardware: NVIDIA H200 (141 GB VRAM) for B2B, NVIDIA L4 (24 GB VRAM) for B2C
  • Lifetime: only for the duration of the respective session
  • Storage: exclusively volatile RAM, no persistent disk for user data
  • Network isolation: WireGuard tunnel as the only connection to the orchestrator

AI models in use

Model Function Licence Execution
faster-whisper large-v3 Transcription MIT European GPU
pyannote 3.1 Speaker recognition (diarization) MIT European GPU
Opus-MT (Helsinki-NLP) Translation (B2C basic) CC-BY European GPU
Gemma 4 E4B Structuring (B2C premium) Gemma Licence European GPU
Gemma 4 26B MoE Premium translation & structuring (B2B) Gemma Licence European GPU

All models run as self-hosted instances in our infrastructure. No external API calls to third parties. No telemetry to model providers.

Compliance chain

  • Orchestrator provider: German company, standard GDPR contract
  • GPU provider: operated entirely within Europe, no US parent
  • Data processing: Contractual chain under GDPR Art. 28 across all subcontractors
  • Special obligations for B2B: § 203 StGB commitment declarations for all relevant sub-processors

What we do not do

  • No use of your data for model training
  • No disclosure to third parties — except where legally required
  • No advertising or analytics trackers, no cookie banner on this website
  • No use of US-based CDNs for sensitive content

Frequently asked questions

How does Klangskript differ from providers like Otter.ai, Descript or Trint?
The services mentioned primarily use US cloud infrastructure and are therefore subject to the US CLOUD Act. For German professionals bound by confidentiality, their use is legally problematic to impermissible. Klangskript processes exclusively in Europe and contractually commits to § 203 StGB.
Can I use Klangskript for transcribing and translating company meetings?
Yes. If your company records internal meetings (e.g. via Zoom or Teams), you can upload the recording to Klangskript afterwards. You receive a complete transcript with speaker recognition and a translation into your preferred language. Please ensure that recording and processing of the meeting has the consent of all participants.
Can doctors already use Klangskript?
Not yet. Use in medical practices — particularly communication with non-German-speaking patients — is in development. We will only bring this capability into practices once our clinical quality evaluation and regulatory assessment are complete. We accept reservations by email; we deliberately do not promise a date.
Is my data used to train AI models?
No. We exclusively use pre-trained open-weight models (Whisper, Opus-MT, Gemma). Your data is processed and then completely and irrevocably removed from volatile memory. There is no training dataset to which your recordings could contribute.
What happens technically when a session ends?
The GPU instance is shut down and irrevocably deleted. The memory is actively overwritten before release. Session metadata (start/end time, audio seconds consumed) is retained for billing — content is not.
Can Klangskript be compelled by US authorities to hand over data?
No. Neither our company nor our infrastructure partners are subject to US jurisdiction. German and European authorities may request information within the framework of legal provisions — but where professional secrets are concerned, the corresponding rights of refusal to testify and seizure protections apply without restriction.
How fast is live transcription?
Under 2 seconds from the spoken word to text display on your device. Precise post-processing of a 30-minute conversation takes between 30 and 90 seconds depending on language and complexity.
Which languages are supported?
Transcription: 90+ languages via Whisper large-v3. Translation in the B2C basic pipeline: common European language pairs via Opus-MT. Premium translation with Gemma 4: 35+ languages out of the box, 140+ languages with reduced quality guarantee.
Can I use Klangskript already?
Not yet. Klangskript is still before market launch. This website is informational only; we will announce here as soon as the platform is available.
Who is behind Klangskript?
Klangskript is a project by Markus Schlüter (Schildow, Germany). Registration as a sole proprietorship under the name “Souveräne Systeme” is in preparation.
Can I transcribe YouTube videos or podcast episodes via a link?
No, deliberately not. We exclusively support direct upload of files that you own or are authorised to process. This eliminates legal grey areas around downloading third-party content — consistent with our principle of not imposing compliance risks on our customers.

Why Klangskript exists

For many years I've worked with Linux and self-hosted software — out of the conviction that you should keep control over your own data. Klangskript grew out of that: confidential conversations — internal meetings, client matters, trade secrets — don't belong in a US cloud, but in an infrastructure bound by European law and confidentiality.

What mattered to me was that this protection isn't reserved for corporations with their own IT department — but is also available to mid-sized companies and small firms, without their own servers and without a lengthy setup.

— Markus Schlüter, Founder of Klangskript

Interested in Klangskript?

Klangskript is still before market launch. For questions about the project, the architecture or future collaboration, you can reach us by email.

Get in touch