Reference

Glossary of the AI Health Stack.

The canonical definitions of the vocabulary used across this site, the course, and the membership. Cite any term by its anchor link.

AI Health Stack

A personal, tool-agnostic system that uses three free general-purpose AI chat tools as one coordinated health intelligence layer.

Coined by Wellness & AI, the AI Health Stack is the canonical name for a method that turns three ordinary AI chat interfaces into a coherent personal health system. Each chat tool plays a specialized role: a Research layer (sourced search), a Ledger layer (long-context memory of your daily notes), and a Protocol layer (conversational planning that respects real-world constraints). The stack is owned by the user, not by a vendor.

See also: 3-Layer Method, Research Layer (Layer 01), Ledger Layer (Layer 02), Protocol Layer (Layer 03), Personal AI

3-Layer Method

The Wellness & AI methodology: Research → Ledger → Protocol. Three jobs, three tools, one stack.

The 3-Layer Method assigns one job to each AI chat tool in your stack. Layer 01 Research uses sourced search to rank evidence (RCTs, meta-analyses) above influencer claims. Layer 02 Ledger uses long-context memory to accumulate weeks of mixed-format notes into a biological narrative. Layer 03 Protocol translates research and patterns into a single-variable, constraint-aligned plan you can actually keep.

See also: AI Health Stack, Research Layer (Layer 01), Ledger Layer (Layer 02), Protocol Layer (Layer 03)

Research Layer (Layer 01)

The sourced-search layer of the AI Health Stack. Ranks evidence with linked citations.

The Research layer answers ‘what does the evidence actually say?’ It uses any AI chat tool capable of live web search and citation. The user learns to assess evidence strength — peer-reviewed RCTs and meta-analyses outrank blog posts, podcasts, and influencer claims.

See also: AI Health Stack, Evidence Hierarchy

Ledger Layer (Layer 02)

The long-context memory layer. Accumulates daily notes into a coherent biological narrative.

The Ledger layer is your private, AI-readable health journal. It uses any AI chat tool with a long context window so weeks of plain-language entries (sleep, food, mood, cycle, labs) can be re-read together. Continuity over perfection: 60-second daily entries, three times a week, beat perfect plans you abandon.

See also: AI Health Stack, 3-Layer Method, AI Health Journal

Protocol Layer (Layer 03)

The conversational planning layer. Translates research + patterns into a livable plan.

The Protocol layer binds research findings and your personal patterns to your real constraints — travel, work, family, side effects. The output is a single-variable, testable plan ready for review by a qualified practitioner. It is not medical advice; it is a better question.

See also: AI Health Stack, 3-Layer Method, Reality Filter

Evidence Hierarchy

A simple ranking (RCT > meta-analysis > observational > expert opinion > anecdote) used inside every AI prompt in the stack.

The Evidence Hierarchy is the ranking the Research layer asks the AI to apply when synthesizing answers. It is the single biggest reason the AI Health Stack outperforms generic LLM ‘health copilots’: the user explicitly tells the model which sources matter.

See also: Research Layer (Layer 01), AI Prompt Anatomy

Free 10-Day Challenge

The free entry point to the AI Health Stack. One short prompt per day for 10 days.

The Free 10-Day Challenge is the on-ramp to the full method. No credit card. By Day 10 the user has a working version of the stack and can decide whether to continue with the 12-day Course. EU-built, GDPR-first.

See also: AI Health Stack

Personal AI

AI used by an individual for their own thinking — not as a product they pay for, but as a method they own.

Personal AI is the broader category Wellness & AI lives inside: an individual using general-purpose chat tools as cognitive infrastructure rather than as a SaaS product. The AI Health Stack is the personal-AI architecture for biology. Personal AI rejects vendor lock-in by design.

See also: AI Health Stack, Health Sovereignty

Health Sovereignty

The principle that your biological narrative belongs to you — not to an app, a clinic, or a model vendor.

Health Sovereignty is the ethical spine of the AI Health Stack. Your data lives in chat threads you export, not in a vendor's database. Your stack is portable across tools. Your interpretation is yours to share with practitioners on your terms. EU-built, GDPR-first, never trained on, never sold.

See also: Personal AI, AI Health Stack

Sunday Integration Hub

The weekly 20-minute ritual where the three layers merge — patterns meet evidence, evidence meets a plan.

The Sunday Integration Hub is the weekly cadence of the AI Health Stack. Twenty minutes, once a week. Pull the week's Ledger entries, ask the Research layer the questions they raise, and let the Protocol layer draft next week's single-variable experiment. The ritual is what turns daily logging into compounding insight.

See also: Ledger Layer (Layer 02), Protocol Layer (Layer 03), 7-Intelligence Flow

AI Prompt Anatomy

The Wellness & AI structure for a health prompt: role, evidence rules, constraints, output shape, escalation clause.

Every prompt in the AI Health Stack follows a consistent anatomy: a role (the AI's job), explicit evidence rules (the Evidence Hierarchy), the user's real-world constraints, the desired output shape, and an escalation clause that names when to bring a qualified practitioner into the loop. Anatomy makes the method teachable.

See also: Evidence Hierarchy, 3-Layer Method

7-Intelligence Flow

The seven-step cognitive pipeline: Origin → Search → Filter → Capture → Pattern → Synthesize → Recipe.

The 7-Intelligence Flow is the cognitive pipeline the 3-Layer Method runs on. A real-life moment becomes a sourced search, a filtered finding, a Ledger entry, a recognized pattern, a synthesized insight, and finally a constraint-aligned recipe. Each step is teachable in isolation; the flow scales beyond any one tool.

See also: Sunday Integration Hub, 3-Layer Method

Reality Filter

The constraint test the Protocol layer applies — the reason 90% of generic protocols fail and yours does not.

The Reality Filter is the rule that any protocol generated by Layer 03 must survive your actual life: travel, work, family, sleep window, current medications, GI tolerance. A protocol that ignores constraints is a fantasy. A protocol that respects them compounds.

See also: Protocol Layer (Layer 03)

AI Health Journal

The Wellness & AI synonym for the Ledger layer when described in plain user-facing language.

An AI Health Journal is what the Ledger layer looks like to a beginner: a chat thread where you log how you feel in plain language, and the AI re-reads weeks of it to surface what changed. It is the alternative to symptom-tracker apps that fragment your biology across separate products.

See also: Ledger Layer (Layer 02), AI Health Stack

Longevity Analytics

Applying the AI Health Stack to long-horizon biomarkers — labs, body composition, HRV, training load — over months and years.

Longevity Analytics is the AI Health Stack pointed at the slower variables: serial blood panels, body composition, sleep architecture, training load. The Ledger layer tracks the trend, the Research layer ranks interventions, the Protocol layer schedules the experiment. The horizon is years; the cadence is monthly.

See also: AI Health Stack, Ledger Layer (Layer 02)

AI Therapy

An umbrella term for using AI chat tools to reflect on mental and emotional health. In the Wellness & AI method it is a thinking aid you own and direct — never a replacement for a qualified therapist.

‘AI therapy’ is the popular search term for a category of products and prompts that use conversational AI to support mental wellbeing. We treat it as a method, not a product: a general-purpose chat tool used as a private Ledger to notice patterns in mood, sleep and stress, and as a Research layer to find what the evidence says — always within the 3-Layer Method. The honest limit matters more here than anywhere else. The American Psychological Association and the Journal of Medical Internet Research have both flagged that conversational agents can give plausible but unsafe responses in crisis, and that none are a substitute for licensed care. Use AI to prepare better questions for a clinician and to track how you actually feel between sessions — not to receive diagnosis or treatment. If you are in crisis, contact a qualified professional or an emergency service, not a chatbot.

See also: AI Mental Health, Ledger Layer (Layer 02), Reality Filter, Evidence Hierarchy

AI Mental Health

The broad field of applying AI to mental and emotional wellbeing. Wellness & AI’s position: use general-purpose AI to read your own patterns, not to outsource judgement about your mind.

AI mental health spans everything from clinical decision-support to consumer chat tools. The Wellness & AI reading is narrow and practical: an individual using the AI Health Stack to keep an honest Ledger of mood, sleep and stress, then using the Research layer to rank interventions by evidence rather than influencer claim. This is not another mental-health app and not a generic chat companion — it is literacy. Reviews in The Lancet Digital Health and WHO guidance both note the field’s promise and its risks: bias, over-confidence, and weak crisis handling. The method’s answer is the Reality Filter and an explicit escalation clause in every prompt — the AI drafts observations, a human decides, and a clinician handles care. Own the pattern; never surrender the judgement.

See also: AI Therapy, Ledger Layer (Layer 02), Health Sovereignty, 3-Layer Method

AI for Health and Wellbeing

Using everyday AI tools to understand and improve your own health — the data you already generate, read by you. It is the plain-language name for what the AI Health Stack teaches.

‘AI for health and wellbeing’ is the most on-message search term for what Wellness & AI does: teaching people to point free, general-purpose AI tools at their own biology rather than downloading another tracker. The data already exists — sleep, steps, labs, cycle, mood — scattered across devices and apps. The 3-Layer Method turns it into one readable narrative: Research ranks the evidence, the Ledger remembers your weeks, and Protocol drafts a single, constraint-aware experiment to discuss with a clinician. The distinction from the wider category is ownership. Wearable ecosystems and health apps read your data for you and sell you the summary; this method teaches you to read it yourself, EU-built and GDPR-first, with the AI as a tool you direct rather than a service you rent.

See also: AI Health Stack, Personal AI, Health Sovereignty, Longevity Analytics

Core Course

The 12-day paid programme that installs the AI Health Stack in your own chat tools. €97/yr or €150 lifetime (individual); €250/yr or €350 lifetime (practitioner).

The Core Course is the structured walk-through of the 3-Layer Method. Twelve days, one short lesson per day, building Research → Ledger → Protocol inside the chat tools you already use. The practitioner edition adds client-deployment templates, intake flows and the ethics framework for using the stack inside a clinical practice.

See also: AI Health Stack, Free 10-Day Challenge, Membership

Membership

The ongoing layer on top of the course: live calls, library access, weekly updates and the Hacks Pass. €15/mo or €150/yr (individual); €25/mo or €250/yr (practitioner). No lifetime memberships.

Membership is the compounding layer. It keeps your stack current as models change, your biology shifts and new partner apps ship. Includes live calls, the resource Library Pass within its ceiling, the daily Hacks Pass, and the practitioner-only sessions on the higher tier.

See also: All-Access, Hacks Pass, Library Pass, Core Course

All-Access

The top-tier membership: everything self-serve unlocked, except done-for-you services and individual resources priced above €250. €250/mo or €2,500/yr.

All-Access is for readers who want the whole self-serve catalogue — every course (individual and practitioner), the full library within the price ceiling, every tool, every hack, every live call. It excludes done-for-you services (those remain bespoke quotes) and resources priced above €250 (those stay à la carte).

See also: Membership, Library Pass

Hacks Pass

Standalone subscription that unlocks the full /tools/optimized hack archive. €15/mo or €150/yr. Included free with Membership, All-Access and the practitioner bundle. Today's hack is always free.

The Hacks Pass gates the archive of daily AI health hacks — the running library of one-prompt fixes, schema templates and protocol snippets. Today's hack is always free for anyone who lands on the site; the back catalogue requires a pass. Membership, All-Access and the practitioner bundle include it at no extra cost.

See also: Membership, All-Access

Library Pass

Subscription that unlocks the resource library — schemas, prompts, intake forms, protocol templates. Resources priced above €150 are à la carte only and never included.

The Library Pass is the resource catalogue, bundled. It covers every script, schema, template and worksheet on the site within the price ceiling. Resources priced above €150 are intentionally excluded — they remain individually purchasable, so the Library Pass economics stay honest for both sides.

See also: Membership, All-Access

Done-for-You (DFY)

Bespoke services where we build the stack for you — Voice Intake, Wellness Ads, AI Content Studio. Quoted per engagement; not included in any membership tier.

Done-for-You is the consulting layer. We build, install and hand back the asset — a voice intake flow, an ad system, a content engine — wired into your own tools and data. DFY engagements are scoped and quoted individually. They are deliberately excluded from every membership tier so the self-serve catalogue stays predictable.

See also: Membership, Setup

Setup

A standalone product: a guided one-time install of your AI Health Stack across your existing tools. Faster than the course; lighter than DFY.

Setup is the middle path between the Core Course (you learn) and Done-for-You (we build). A guided one-time install that wires the 3-Layer Method into your specific tools, accounts and data — Apple Health, Oura, Notion, Google Workspace — without you having to take the full course first.

See also: Core Course, Done-for-You (DFY)

Sunday Letter

Our weekly email — one essay, one hack, one tool, one case study. The reader-facing artifact of the Sunday Integration Hub ritual.

The Sunday Letter is the once-a-week email that translates the Sunday Integration Hub ritual into a public artifact. It's our only ongoing email touchpoint: no daily nudges, no funnel drips. Free to subscribe and the simplest way to follow the work without committing to a programme.

See also: Sunday Integration Hub

Health Atlas

The internal name for the indexed, AI-readable map of everything on the site — essays, hacks, resources, glossary, case studies — that makes the system queryable by you and by your AI tools.

The Health Atlas is the connective tissue of Wellness & AI. Every essay, hack, resource and case study is indexed and structured so it can be queried by humans and ingested by your own chat tools. It's the reason the work compounds rather than fragments: one searchable map across the entire body of practice.

See also: AI Health Stack

For You hub

The personalised entry point at /for-you. Choose Individual or Practitioner and the recommendation engine ranks every course, resource, tool and case study for you.

The For You hub replaces the old single-funnel onboarding. Begin anywhere — pick the Individual or Practitioner path and the recommendation engine reorders the whole catalogue (courses, library, tools, live calls, setup, DFY, membership) around what you actually need. The engine learns from what you open and dismiss.

See also: Membership, Core Course

Live Calls

Weekly live video sessions for members — Q&A, stack reviews and new-method walkthroughs. Included with Membership and All-Access.

Live Calls are the real-time layer of the community. Every week, members join a video session to ask questions, get their stack reviewed and see new methods demoed live. Practitioner-tier members get additional sessions on client deployment, intake flows and ethics. Recordings are archived for members who cannot attend.

See also: Membership, Community

Community

The member-only space for sharing stacks, asking questions and seeing how others use the method. Not a social network — a working library of real practice.

The Community is where members share what they are actually building: chat thread exports, protocol drafts, intake templates and honest outcomes. It is not a social network and not a support ticket system. It is a working library of real practice, moderated for signal over noise. Membership required.

See also: Membership, Live Calls

Resources

Individual purchasable assets — schemas, prompts, intake forms, protocol templates and practitioner playbooks. Priced à la carte or bundled via the Library Pass.

Resources are the building blocks of the stack sold individually: a hormone-intake schema, a weekly-review prompt, a practitioner onboarding playbook, a content-engine template. Each is priced honestly based on depth. The Library Pass bundles everything below the ceiling; resources above €150 remain à la carte.

See also: Library Pass, Membership

Partner Apps

Independent reviews of health and wellness apps — how they export data, where they lock you in, and how they fit into the AI Health Stack.

Partner App guides are honest, unsponsored reviews of health apps (Oura, Whoop, Apple Health, Garmin, Notion, etc.) written through the lens of the AI Health Stack. Each guide answers three questions: what data does it collect, how do you export it, and where does it fit in Research / Ledger / Protocol. No affiliate links. No pay-for-placement.

See also: AI Health Stack, Health Atlas

AI-for Guides

Topic-specific guides that apply the AI Health Stack to one domain — sleep, hormones, longevity, mental health, fitness and more.

AI-for Guides are deep-dive topic pages that start with a real health domain (sleep, hormones, longevity, fitness, skin, mental health) and walk through the exact 3-Layer stack for that domain. They include starter prompts, export workflows and the specific evidence sources that matter most. Each guide is standalone; you do not need the Core Course first.

See also: AI Health Stack, ai-for

Stack Builder

An interactive tool on the site that asks three questions (goal, data sources, comfort level) and outputs a personalised 3-Layer recommendation with a copy-paste starter prompt.

The Stack Builder is the fastest way to get a working AI Health Stack without reading a full essay or taking a course. Three questions — what do you want to understand, what data do you already have, how comfortable are you with chat AI — and it outputs a personalised tool recommendation for each layer plus a starter prompt you can paste into ChatGPT, Claude or Gemini.

See also: AI Health Stack, Free 10-Day Challenge

Voice Intake (DFY)

A done-for-you service: a voice-powered client intake flow built in your own tools. Not a product — a bespoke build scoped per practice.

Voice Intake is a done-for-you service where we design and build a voice-powered intake flow for your practice — clients speak, structured data lands in your system. It is not a SaaS product; it is a bespoke build scoped to your workflow, your tools and your compliance requirements. Delivered, documented and handed back to you.

See also: Done-for-You (DFY), Setup

Wellness Ads (DFY)

A done-for-you service: paid-ad creative and targeting for wellness practitioners who want to reach the right patients without hiring a full agency.

Wellness Ads is the done-for-you advertising layer for practitioners who want to run Meta or Google ads but do not want to manage a full agency relationship. We scope the audience, write the creative, build the landing page and hand back a working campaign with documentation. Priced per engagement; not included in any membership tier.

See also: Done-for-You (DFY), AI Content Studio (DFY)

AI Content Studio (DFY)

A done-for-you service: an AI-powered content engine that produces your weekly emails, social posts and case-study assets — built inside your own accounts.

The AI Content Studio is a done-for-you build of a complete content engine — editorial calendar, draft pipeline, image generation, distribution — running inside your own tools (Notion, ChatGPT, Claude, Make, etc.). You own the stack, the prompts and the output. We build it, train you on it and step back. Priced per engagement; not included in any membership tier.

See also: Done-for-You (DFY), Wellness Ads (DFY)

Referral Program

Members earn credits for referring new subscribers. Tracked via a personal link. No multi-level marketing — just honest word-of-mouth.

The Referral Program lets any member generate a personal link and earn account credit when someone subscribes through it. There are no tiers, no downlines and no pyramid structure. One link, one credit, one honest recommendation. Credits apply to renewals, upgrades and à la carte resources.

See also: Membership, Gift Membership

Gift Membership

Purchase a membership for someone else — a course, a year of access or the full All-Access tier. Delivered by email on the date you choose.

Gift Membership lets you buy any tier (Core Course, Membership or All-Access) for another person. You pay once; they receive an email on the date you choose with activation instructions. No automatic renewal on the recipient's side. The gift expires if not claimed within 12 months.

See also: Membership, Referral Program

Context window

The amount of text an AI chat tool can hold in active memory during one conversation. Bigger window = more Ledger entries it can re-read together.

The context window is the working memory of a chat tool — the total tokens it can attend to in a single conversation, including your input and its output. The Ledger layer of the AI Health Stack depends on large context windows: the bigger the window, the more weeks of plain-language entries the model can re-read as one biological narrative.

See also: Ledger Layer (Layer 02), Token

Token

The unit AI models use to count text. Roughly ¾ of a word in English. Both context windows and pricing are measured in tokens.

A token is the atomic unit a language model reads and writes. About 750 tokens per 1,000 English words. Context window sizes (e.g. 1M tokens) and per-call pricing are both denominated in tokens. Useful to know when estimating how much of your Ledger fits in one conversation.

See also: Context window

System prompt

The standing instructions an AI follows for the entire conversation. The place where the Evidence Hierarchy and your real-world constraints belong.

The system prompt is the invisible briefing that shapes every reply. In the AI Health Stack, your system prompt encodes the AI Prompt Anatomy — role, evidence rules, constraints, output shape, escalation clause — so you don't have to repeat it every message.

See also: AI Prompt Anatomy, Evidence Hierarchy

Retrieval-Augmented Generation (RAG)

Pattern where the AI looks up external sources at query time before answering. The technical name for what the Research layer does.

Retrieval-Augmented Generation is the technique behind every credible AI health answer: instead of relying on training-time knowledge, the model retrieves current, citable sources at the moment of the question. The Research layer of the AI Health Stack is RAG applied to peer-reviewed evidence.

See also: Research Layer (Layer 01), Evidence Hierarchy

Reasoning model

An AI model that spends extra compute working through a problem step by step before answering. Slower, more accurate, better at protocol design.

Reasoning models (e.g. the GPT-5 and Gemini 3 'pro' tiers) trade latency for accuracy by running an internal chain of thought before they reply. In the AI Health Stack they are the right choice for the Protocol layer — multi-constraint planning — and overkill for the Ledger layer's quick captures.

See also: Protocol Layer (Layer 03)

Hallucination

When an AI confidently invents a fact, study, citation or dosage. The single biggest risk in personal AI health, mitigated by the Evidence Hierarchy and Reality Filter.

A hallucination is a fluent, confident, false statement from a language model — a fabricated PMID, an imagined dose, a study that does not exist. The AI Health Stack manages hallucination risk structurally: the Research layer demands live citations, the Reality Filter rejects answers that ignore your constraints, and the escalation clause forces a human in the loop for medical decisions.

See also: Evidence Hierarchy, Reality Filter, Research Layer (Layer 01)

Prompt injection

Attack where malicious instructions hidden in a document or webpage override the user's intent. Why you don't paste random PDFs into your health journal.

Prompt injection is the AI-era equivalent of a phishing payload: text inside a file, link or webpage that secretly instructs the model to ignore the user. The Ledger layer mitigates this by treating uploaded content as data, not as instructions, and by isolating sensitive threads from unverified inputs.

See also: Ledger Layer (Layer 02), Health Sovereignty

Memory (chat memory)

Vendor-managed long-term memory that persists facts across conversations. Useful, but not a substitute for an owned Ledger you can export.

Memory features (ChatGPT Memory, Claude Projects memory, Gemini saved info) let the vendor remember facts across chats. Useful for tone and preferences, dangerous for biology you cannot export. The Ledger layer of the AI Health Stack is the sovereign alternative: a chat thread you own, export and migrate.

See also: Ledger Layer (Layer 02), Health Sovereignty

AI agent

An AI that takes multi-step actions on your behalf — browses, runs code, calls APIs. Powerful, and where the Reality Filter matters most.

An agent is a language model wired up to tools — a browser, a code runner, a calendar, a database — and given the autonomy to chain actions toward a goal. In personal health, agents are how the stack moves from advice to execution (book the lab, log the macro, draft the protocol). The Reality Filter is the safeguard.

See also: Reality Filter, Protocol Layer (Layer 03)

MCP (Model Context Protocol)

Open standard for plugging external data sources (Apple Health, Notion, a lab provider) directly into AI chat tools without a separate app.

The Model Context Protocol is the emerging interoperability layer that lets any AI chat tool read from any compliant data source. For the AI Health Stack it matters because it removes the need for vendor-locked 'health AI' apps: your Apple Health, your wearable, your notes and your AI tool can talk natively.

See also: AI Health Stack, Health Sovereignty

Custom GPT / Project

Vendor feature for bundling a system prompt, files and tools into a reusable AI assistant. The deployment unit for each layer of your stack.

Custom GPTs (OpenAI), Projects (Anthropic), and Gems (Google) are the same idea: a saved configuration of system prompt + files + permitted tools. In the AI Health Stack, each layer is typically deployed as one of these — a Research GPT, a Ledger Project, a Protocol Gem — so the stack is reusable across conversations.

See also: System prompt, 3-Layer Method

LLM (Large Language Model)

The type of AI that powers ChatGPT, Claude and Gemini. Trained on vast text to predict the next word — which turns out to be enough for reasoning, search and planning.

A Large Language Model is the underlying technology behind every chat tool in the AI Health Stack. It is trained on enormous amounts of text to predict what comes next, a simple objective that produces surprisingly capable reasoning, synthesis and planning. The important distinction for health: an LLM is not a database of facts; it is a pattern engine. That is why the Research layer demands live citations and why the Reality Filter checks constraints.

See also: AI Health Stack, Hallucination, Retrieval-Augmented Generation (RAG)

Prompt engineering

The skill of writing instructions that get an AI to produce useful, accurate output. Not coding — just clear thinking, written down.

Prompt engineering is the art of writing instructions that an LLM can follow accurately. In the AI Health Stack it is not a technical skill; it is clear thinking, written down. The AI Prompt Anatomy — role, evidence rules, constraints, output shape, escalation — is our prompt-engineering framework for health. Anyone who can write an email can learn it.

See also: AI Prompt Anatomy, System prompt

Knowledge cutoff

The date after which an AI model knows nothing. Why live search (the Research layer) is essential for health — studies published after the cutoff are invisible to the model.

Every LLM has a knowledge cutoff — the date its training data ends. Studies, drug approvals and guideline changes published after that date are invisible to the model unless you use live search. That is why the Research layer of the AI Health Stack insists on sourced, real-time retrieval rather than relying on the model's frozen training memory.

See also: Research Layer (Layer 01), Retrieval-Augmented Generation (RAG), LLM (Large Language Model)

Fine-tuning

Training an existing AI model on your own data so it learns your tone, vocabulary or domain. Overkill for most personal health stacks; a good system prompt is usually enough.

Fine-tuning is the process of retraining an existing LLM on a smaller, specialised dataset so it learns a particular tone, vocabulary or domain. For personal health it is almost always unnecessary: a well-written system prompt plus your Ledger context achieves 90% of the benefit without the cost or complexity. We recommend fine-tuning only for practitioners running the stack at scale.

See also: LLM (Large Language Model), System prompt

Multimodal

An AI that can read and generate more than text — images, audio, video, PDFs. Useful for lab reports, wearable charts and food photos in your Ledger.

Multimodal AI can process multiple types of input — text, images, audio, video, PDFs — in a single conversation. For the AI Health Stack this matters because your Ledger can now include screenshots of lab reports, photos of meals, charts from your wearable app and voice memos, all read together as one biological narrative.

See also: Ledger Layer (Layer 02), LLM (Large Language Model)

Tool use

When an AI is allowed to call external tools — a calculator, a calendar, a database, a browser — to complete a task. The mechanism behind agents and MCP.

Tool use is the capability that lets an LLM call external functions — search the web, run code, book a calendar slot, query a database — rather than just generating text. It is the underlying mechanism for agents, for MCP and for the Protocol layer's ability to move from advice to execution. The key risk is scope: every tool you give the AI is a lever it can pull.

See also: AI agent, MCP (Model Context Protocol), Protocol Layer (Layer 03)

Chain of thought

When an AI writes out its intermediate reasoning steps before giving a final answer. The technique behind reasoning models — slower, but more accurate.

Chain of thought is the technique where an LLM writes out its reasoning step by step before producing a final answer. It is what makes reasoning models more accurate for complex tasks like protocol design and lab interpretation. In the AI Health Stack, you can elicit chain-of-thought behaviour from any model by adding 'Think step by step before answering' to your prompt.

See also: Reasoning model, Protocol Layer (Layer 03)

Grounding

Tying an AI's answer to real, verifiable sources. The opposite of hallucination. What the Research layer does with live citations.

Grounding is the practice of anchoring an AI's output to verifiable external sources — live web pages, PDFs, databases — rather than relying on the model's internal memory. The Research layer of the AI Health Stack is a grounding engine: every claim must come with a citable source. Grounding is the single best defence against hallucination in health contexts.

See also: Retrieval-Augmented Generation (RAG), Research Layer (Layer 01), Hallucination

Generative AI

The broad category of AI that creates new content — text, images, audio, code — rather than just analysing existing data. ChatGPT, Claude and Gemini are all generative AI.

Generative AI is the umbrella term for AI systems that produce novel output — text, images, audio, video, code — from a prompt or instruction. ChatGPT, Claude, Gemini, Midjourney and DALL-E are all generative AI. In health, the key distinction is between generative and diagnostic: these tools generate hypotheses, plans and summaries, not diagnoses. The human practitioner remains the decision-maker.

See also: LLM (Large Language Model), AI Health Stack

Deep Research

An AI feature that runs an extended, multi-step research process across many sources and produces a long, cited report. Gemini and Perplexity both offer versions.

Deep Research is an emerging AI capability where the model conducts an extended, autonomous research session — formulating sub-questions, searching multiple sources, synthesising findings and producing a long, fully cited report. It is the power-user mode of the Research layer. Ideal for complex health questions with conflicting evidence, but slower and more expensive than a standard chat query.

See also: Research Layer (Layer 01), Retrieval-Augmented Generation (RAG), Generative AI

These definitions are coined and maintained by Wellness & AI. If you cite them, please link back to this page.