lumi. uses artificial intelligence to power most of its core features. This notice explains exactly which features use AI, which AI systems are involved, what those systems can and cannot do, and what that means for you as a user. Transparency here is intentional — you deserve to know when a machine is interpreting your words.
1 — AI is central to how lumi. works
lumi. would not function without AI. Its core value — turning unstructured voice entries into meaningful patterns and reflections — is only possible through machine learning models. This is not incidental use of AI; it is the product.
Two AI systems are involved: OpenAI's Whisper (speech-to-text transcription) and Anthropic's Claude (analysis, generation, and synthesis). Both are third-party services. Neither is developed or operated by lumi. (Sarah-Jane Barton trading as lumi.).
2 — What AI does in lumi.
Every AI-powered feature in lumi. is listed below, along with which model produces it and what it actually does.
OpenAI Whisper
Transcription
Converts your voice recording to text. The transcript is the foundation for everything else in the app.
Claude Haiku
Entry tagging
Assigns your entry to one or more life areas, scores sentiment and energy, and identifies goal mentions.
Claude Haiku
Intention detection
Identifies specific things you said you would do — flagged for pattern alerts if unaddressed after 21 days.
Claude Haiku
Daily questions
Generates one context-aware question each morning based on your recent entries, goals, and life areas.
Claude Haiku
Weekly signals
Produces a short ambient observation about your week — 2–4 sentences, generated every Sunday.
Claude Haiku
Pattern alerts
Checks whether an intention has been resolved, mentioned, or gone quiet — surfaces an alert if unaddressed.
Claude Haiku
Voice profile
Builds a statistical model of your writing style — formality, rhythm, vocabulary, recurring phrases — over time.
Claude Sonnet
Monthly summary
Synthesises all entries from the month into a structured reflective summary, written to match your own voice.
Claude Sonnet
Onboarding extraction
Interprets your onboarding answers to suggest life areas, extract initial goals, and seed your voice profile.
Claude Haiku
Life context synthesis
Converts your optional background answers into a concise first-person context statement used to frame summaries.
Claude Sonnet
Session summaries
Generates a structured pre-session summary from recent entries for therapy or coaching sessions (if Sessions is enabled).
Claude Haiku
Action suggestions
Suggests small between-session actions based on generated session summaries and notes.
3 — What AI-generated content is and is not
This is important
All AI-generated content in lumi. is produced for personal reflection only. It is not professional advice of any kind — not medical, psychological, therapeutic, legal, or financial. AI outputs may be inaccurate, incomplete, or not representative of your actual state, intentions, or experience. You should never make important decisions based solely on AI-generated content from lumi.
Specifically:
- Sentiment and energy scores are algorithmic assessments of language — not clinical evaluations of your mental or physical health
- Life area tags are AI inferences — they may not always correctly identify what a given entry is really about
- Intention detection is pattern matching — the AI may flag things as intentions that were not, or miss intentions that were subtle
- Pattern alerts are triggered algorithmically after 21 days of silence on an intention — they are not personalised recommendations or clinical prompts
- Monthly summaries are AI interpretations of your entries — they may reflect the language you used rather than the experience you had
- The voice and writing profile is a statistical model — it may not accurately represent you as a person, and should be read as an observation, not a definition
- Session summaries are generated from journal entries — they are not a substitute for professional clinical notes or a therapeutic record
4 — No human review
No human — including anyone at lumi. — reads your journal entries or reviews AI-generated content produced from them. All processing is automated. The AI systems that power lumi. are operated by Anthropic and OpenAI respectively; their staff may have access to API data for trust and safety purposes under their own policies, but this is incidental to lumi.'s operation and is not routine review of your content.
The distress detection system — which checks entry transcripts for signs of crisis before AI processing — is also fully automated. It does not involve human review of any entry, and it does not alert any person.
5 — AI models used
lumi. currently uses the following AI models. This may change as models are updated or replaced — this notice will be updated accordingly.
- OpenAI Whisper — speech-to-text transcription. openai.com/research/whisper ↗
- Anthropic Claude Haiku — fast, lightweight analysis and generation tasks
- Anthropic Claude Sonnet — substantive synthesis tasks requiring depth and voice fidelity
Anthropic's models are accessed via their API. Anthropic does not use API data to train their models by default. See Anthropic's privacy policy ↗ for current data handling commitments.
6 — Regulatory context
AI regulation is developing quickly. lumi. is designed with transparency and user control as foundational principles — both because it is the right approach and because it aligns with the direction of travel in UK and EU regulation.
EU AI Act
Came into force August 2024. Provisions phasing in through 2026. Requires transparency about AI-generated content and prohibits certain high-risk uses. lumi. is designed to comply with transparency obligations — this notice is part of that.
UK AI regulation
The UK government's principles-based approach places transparency and explainability as core expectations. lumi.'s approach — clearly labelling AI outputs, disclosing models used, and avoiding AI in crisis moments — is consistent with these principles.
UK GDPR
Where AI outputs are used to make decisions that significantly affect users, Article 22 rights apply. lumi. does not use AI to make automated decisions about users — all AI outputs are reflective and informational only. Users retain full agency.
lumi. does not use AI to make any automated decisions that produce legal or similarly significant effects on users. AI outputs in lumi. are informational and reflective — the user always decides what to do with them. This is intentional design, not a compliance workaround.
7 — Your rights in relation to AI-generated content
You have the right to:
- Override AI-generated life area tags on any entry
- Disregard any AI-generated output — summaries, signals, questions, and pattern alerts are prompts for reflection, not conclusions
- Delete any entry and all AI-generated content associated with it
- Delete your account and all associated data, including your voice and writing profile, at any time
- Request information about how AI has processed your data — contact lumi. using the details below