How Next of Kin works, what it does, and what it doesn't.
Last reviewed: July 2026 | Maintained by: Next of Kin
Next of Kin is a knowledge tool, not a medical system. This page explains how it works in plain language — what happens to the information you share, how the AI uses it, and where the limits are.
When a caregiver uses Next of Kin, they build a knowledge base about the person they care for. That knowledge base is made up of everything they choose to share: journals, voice recordings, documents, photos, interview transcripts, notes.
When a carer asks a question — “What should I do if Daniel seems upset?” — Next of Kin will search that knowledge base and return an answer. The answer will be drawn from what the family actually shared, not from general care guidance or clinical databases.
That's the core of it. Everything else is in service of that moment.
Everything you upload is stored in a secure database provided by Supabase, a GDPR-compliant infrastructure provider. All data is encrypted at rest and in transit.
Your profile and its contents are only accessible to:
We do not have standing access to your profile. We do not browse or read what you have shared.
Next of Kin uses a technique called Retrieval-Augmented Generation, or RAG. Here is what that means in practice. Capture and secure storage are live today; the retrieval and answer steps below were validated in a full proof of concept over our founding family's real material in July 2026, and are now being built into the product.
Step 1 — Ingestion. When you upload a document, record a voice note, or complete an interview, the content is transcribed where needed and stored securely. As the knowledge base comes online, it will be broken into passages, each converted into a mathematical representation called an embedding — a way of capturing the meaning of the text so it can be searched by meaning rather than just by keyword.
Step 2 — Storage. Those embeddings will be stored alongside the original text in a vector database. The original text is never modified or summarised at this stage.
Step 3 — Retrieval. When a carer asks a question, we will convert that question into an embedding and search for the chunks in the knowledge base that are most relevant to it. We will retrieve the closest matches.
Step 4 — Generation. The retrieved chunks — the actual words from the family's journals, interviews, and notes — will be passed to a large language model along with the carer's question. The model will generate an answer using those specific passages as its source material.
Step 5 — Citation. The answer will include a reference to where the information came from. A carer will be able to see that an answer came from Ruth's journal entry from March 2024, for example — not from a generic care database or from the AI's general training.
The system is designed so the AI can only answer from what has been shared. If a family hasn't mentioned something, the AI doesn't know it. It will not invent information, guess, or fill gaps with general care advice. If a question cannot be answered from the knowledge base, it says so — we verified exactly this behavior in our July 2026 proof of concept, where questions the material couldn't answer were met with a plain statement of the gap instead of a guess.
The AI does not:
Next of Kin uses AI services provided by Anthropic (Claude) and OpenAI (Whisper for transcription). When content is processed — a voice note transcribed, a document's text extracted, and — as the knowledge pipeline comes online — passages embedded — it passes through these providers' systems.
Both Anthropic and OpenAI have data processing agreements that govern how they handle content sent to their APIs. We are in the process of formalising Business Associate Agreements with both providers.
We do not use your content to train AI models. We do not permit our AI providers to use your content for model training.
When you record a voice note or complete a voice interview, the audio is sent to OpenAI's Whisper service for transcription. The resulting text is stored in your knowledge base. The original audio file is also stored.
Whisper is a speech-to-text model. It does not analyse the content of what you say — it converts speech to text. The content analysis happens at the retrieval and generation stage described above.
To be clear about our limits:
Answers from Next of Kin will be suggestions, not instructions. They are informed by what a family has shared and should be treated the way you would treat advice from a knowledgeable colleague — useful context, not a directive.
A carer who receives an answer about how to help someone in distress should use their own professional judgment alongside that answer. Next of Kin supports carers; it does not replace them.
AI tools can sometimes generate confident-sounding answers that aren't true. This is called hallucination, and it is a known limitation of general-purpose AI. Next of Kin is built to prevent it.
When the knowledge base is ready, a carer will be able to ask questions in plain language — and the answer will come from what the family actually shared: cited to the source, in the parent's own words. We validated this approach in July 2026 in a full proof of concept over our founding family's real material. It answered from the family's words with every citation traceable, and — just as important — when we asked it questions the material couldn't answer, it said so plainly instead of guessing. The carer-facing chat is now in development, being built carefully rather than quickly.
The knowledge base is your family's voice. Not the AI's imagination.
This page is reviewed whenever we make a material change to how the technology works — a new AI provider, a change to how data is stored, a new processing step. The review date at the top of the page reflects the last time it was checked for accuracy.
If you have a question about something not covered here, write to us at hello@nextofkin.ai.