Helsingborg-based Klang.ai has raised 15 MSEK (€ 1.3 million) in its latest funding round. The startup develops its own language models for transcribing, summarizing, and analyzing conversations, with a focus on keeping sensitive data local and European rather than flowing to US cloud providers.
Alongside the funding, the company has released Klang Pianissimo, the first model on its new streaming endpoint. Until now, transcribing Swedish meant recording the full audio and waiting for the batch to finish. With the new endpoint, text arrives while people are still speaking, and the finished transcript is ready when the stream ends.
This makes Klang one of the few companies in Europe set up to train and release open-weights models.
Pianissimo is trained for Swedish as it is actually spoken, dialects included. In Klang’s benchmarks, it processes an hour of audio in about a second, with every token pinned to 80 milliseconds of audio. The endpoint is compatible with the OpenAI realtime API, meaning developers already streaming audio to a US provider can switch with just a base URL and an API key. Audio is processed on Klang’s own hardware in Sweden, and nothing is stored or used for training.
“Pianissimo can transcribe over 40 hours of audio in a minute. Its hard to imagine how much audio that really is,”
Said Mattias Fält, Head of AI & Co-Founder at Klang.ai
The investment just raised is from business angels such as Johan Lenander, Emil Sjödin and Daniel Gadd. It comes as human conversation becomes a core information source in an AI-driven workplace, while many existing tools still send sensitive speech data to large US cloud services. Klang.ai positions itself as the European alternative, ensuring that information stays under local control.
The new capital will support continued product development and international expansion across the Nordics, Benelux, DACH, and Japan, as well as around ten new hires across research, development, and marketing.
