Swedish AI company Klang has raised SEK15 million as it develops software intended to turn meetings, interviews, and other spoken work into searchable organisational memory rather than a collection of recordings and personal notes.

The Helsingborg company says the financing values it at SEK150 million and arrives with annual recurring revenue of SEK20 million. Existing investors and shareholders backed the round, led by Johan Lenander alongside Emil Sjödin and Daniel Gadd.

Klang’s proposition sits beyond basic meeting transcription. Recording speech and producing a summary are already common capabilities across workplace software, leaving more specialised providers to compete on retrieval, privacy, language quality, deployment, and the ability to connect information across conversations.

Conversation data becomes organisational data

The company targets uses including legal work, audit, insurance, research, sales, and public services, where the record of a conversation can have operational or evidential value after a meeting ends.

A searchable archive can help employees return to an interview, investigation, customer conversation, or research session without relying on somebody’s notes. However, the usefulness of the system also depends on preserving access to the original material so that generated interpretations can be checked.

Klang has therefore put considerable emphasis on privacy and deployment. Its product supports European hosting and environments intended to give organisations greater control over where sensitive audio, transcripts, and derived information are processed.

That becomes significant in sectors where conversation data can contain personal, legal, commercial, or otherwise confidential information. The convenience of a general-purpose cloud transcription service may be less attractive when a recording forms part of an investigation or regulated case file.

Speech remains a local technology problem

Klang is also developing its own speech technology. It has released Pianissimo, an open-weight model for Swedish speech recognition, alongside a dataset and benchmark based on recordings collected across Sweden.

The company claims the model materially reduces recognition errors compared with alternatives and can operate at high transcription speed. Those performance figures come from Klang and require independent testing across real production audio.

The local-language focus addresses a recurring weakness in AI systems trained predominantly on large global datasets. Accuracy can deteriorate across dialects, specialist vocabulary, overlapping speakers, and less widely represented languages.

Allowing organisations to run an open-weight model themselves gives customers another choice over where audio is processed and how tightly the technology is integrated with their own infrastructure.

Search creates governance questions too

Turning conversations into persistent memory creates risks alongside the efficiency gain. Spoken work includes tentative ideas, mistakes, privileged discussions, personal data, and comments that participants may not expect to become indefinitely searchable.

Organisations therefore need policies around recording, consent, access, retention, deletion, and whether a generated summary can be treated as an authoritative record.

The issue becomes more important if AI connects information across several conversations. That can reveal useful patterns and reduce repeated work, but it creates a richer dataset if permissions are poorly designed or an account is compromised.

Klang’s funding remains modest beside investment in foundation models, but the commercial problem is close to everyday organisational work. Large amounts of useful context still pass through conversations and then disappear into memory, notebooks, or isolated recordings.

The opportunity is to make that information reusable without turning every conversation into an uncontrolled permanent record. Klang’s development of searchable memory, local-language speech technology, and controlled deployment shows how those two requirements are beginning to shape the same enterprise product.