The short version
- What you create in your own organisation is yours. All of it can be exported at any time, and deleted or scrubbed.
- What you contribute to someone else’s organisation or workspace belongs to that space. By accepting the invitation you agree they may keep it.
- We don’t track who said what. That is deliberate, and it means we cannot scrub a single remark automatically. If you want something removed from another organisation’s record, ask its admin.
- Kair works from the group’s perspective. Summaries and patterns speak about what the group discussed, never about what a named person said.
Your own organisation
When you create an organisation and work inside it, you are the owner of everything in it: recordings, transcripts, summaries, topics, documents, the topic graph built from them, and the memberships around them. Nothing about that is locked in. You can:- Delete whole sessions, with their audio, transcripts and summaries.
- Export your data at any point, as a machine-readable archive (personal, workspace or organisation).
- Erase your data through a data request, reviewed by the organisation’s admin, which in your own organisation is you. Audio and transcripts are deleted, the topic history and topic graph are rebuilt without them, and what has to remain for the record is blanked with a scrub marker, so anyone who meets the gap can see why it is empty.
Joining someone else’s organisation or workspace
Being invited to another organisation or workspace is a different relationship. You are a guest in a shared space, and the space belongs to whoever runs it. By accepting an invitation and taking part, you implicitly give that organisation or workspace permission to keep what you provide. That is what makes collaboration work: a workshop’s record can’t fall apart because one participant leaves. This means:- The organisation, not you, decides how long its record is kept.
- Leaving a workspace does not automatically remove what you contributed to it.
- If you want your contributions removed, you ask the organisation’s admin. They review the request and, if approved, the erasure runs. See Data subject requests.
Why we can’t scrub one comment automatically
Kair does not record which person made which remark in a group conversation. A session is stored as a conversation, not as a list of per-person statements, and the topics and patterns we derive from it are attached to the group’s discussion rather than to individuals. That is a privacy feature, and it has a trade-off. Because no record ties a particular comment or topic to you, the system cannot go and find “everything you said” in another organisation’s data and remove it on its own. If you want a specific comment, or the topic it fed into, taken out of an organisation’s record, ask that organisation’s admin. They know the context and can decide what to remove.Where a member has claimed a speaker label in a session, erasure does remove that claim and the sessions they themselves own. A claim is something you choose to make; without one, nothing links a remark to you.
Built for the group, not the individual
Kair exists to help groups work out where they agree, where they differ, and what to do next. People take part more freely when nobody is quoted back at them, so the platform is designed to talk about topics and patterns, never people.- Summaries describe what the group discussed. The prompt forbids names, contact details, addresses, exact ages and anything that singles someone out. People appear as a role (“the head teacher”) or as “a participant”. Claimed speaker names and labels that slip through are replaced with
[participant]. See Session summaries. - Topics and entities are about subjects, not people. The topic graph holds places, organisations, concepts and themes. People are never entities in it. See Topic history.
- Patterns belong to the conversation. Divergence, convergence and intensity are reported for the group. Where the source carries no speaker labels there is no attribution at all, and a point is recorded as “this point was made”. See Moderator agent.
- At transcription. A named-entity recognition (NER) model runs on the transcription worker and replaces detected people with
<PERSON>before transcript segments are saved, alongside a redactor for emails, phone numbers, IBANs and similar identifiers. Repeat mentions of a detected name, including German genitives (“Annas Vorschlag”), are caught too. See Transcription. - At summarisation. The prompt forbids names, and claimed speaker names and labels are scrubbed to
[participant]. - In the topic graph. People are never stored as entities.