AAVATME - PRODUCT SPECIFICATION (the AI that knows you) Draft 27 Jun 2026. Sibling 2 in the AAVAT brand house. Held for Nikki’s review. Companion docs: aavattype_b2c (the brand house), full_pattern_product_spec (the 6-lens engine), aavattype_pricing_research (the 2026 pricing study), the CONCEPT_living_signature doc in ~/Projects/avatar-quiz/full-pattern/site-emerald-v2/.
THE ONE LINE
AAVATtype tells you who you are. AAVATme puts that knowledge to work, every day. It is a personal AI seeded with your validated multi-lens self-model, so it genuinely knows you across traits, values, drives, thinking style, strengths, attachment and your story, and supports you from inside your own psychology instead of generic advice.
WHY IT EXISTS (the gap it fills)
A generic chatbot does not know you, so its help is generic. Everyone is now bolting “AI” onto quizzes, but the AI is only as good as the model underneath it, and almost nobody has a real, validated model of the person. AAVATme is that model put into action. The wedge, and the brand line: an AI that knows you is only trustworthy if what it knows is true. The science is not the boring compliance part, it is the entire reason to trust the AI built on top of it.
THE ASSET IT RUNS ON
One structured profile object per person (the “coded construct” in full-pattern/profile-schema.mjs): HEXACO factor scores with confidence bands, the deeper lens scores (values, drives, thinking modes, strengths, attachment and emotion regulation), headline type plus shade, narrative themes, and a consent block. This single object is the genome. The signature is one rendering of it, the report is another, and AAVATme is the same object turned into a working intelligence. Build once, express many.
THE KEY DESIGN DECISION: TWO DELIVERY MODES
Mode 1 - Bring your own AI (the portable self-file). LAUNCH MODE. AAVATme compiles the profile into two things: a human-readable “About Me” operating manual, and a structured prompt block the person pastes into their own ChatGPT, Claude or Gemini (custom instructions, a project, or memory). No infrastructure for us, instant value, and it is the honest position: we do not trap your data in our app, we hand you the keys so every AI you already use actually knows you. Near-zero marginal cost. Ships now as part of the paid AAVATtype tier.
Mode 2 - Hosted AAVATme (the companion). RECURRING MODE, BUILT LATER. A chat surface we host, pre-loaded with the profile, with persistent memory, the live shadow-watch, guided journaling, and the evolving “Finish” layer of the signature. This is where retention and recurring revenue live. Built on Next.js plus the Vercel AI Gateway (the profile becomes the system context) with a memory store (Supabase). It also doubles as the spine AAVATAI reuses.
Sequence: launch Mode 1 (the self-file, inside the top AAVATtype tier or as an upsell), then grow into Mode 2 (the subscription companion) once the recurring content and memory engine exist.
WHAT IT ACTUALLY DOES (the capability set, by job-to-be-done)
A. Knows you. Loads the full profile as context, for example: “high Emotionality at the 82nd percentile, top values security and benevolence, avoidant-leaning attachment, prevention-focused, under-uses the possibility thinking mode.” It then adjusts tone, pace and directness to the person (a high-Emotionality, avoidant person gets warmth and space, not blunt push).
B. Advises inside your psychology. A decision co-pilot that weighs options against your values and needs, not generic pros and cons, and remembers which past choices aligned with your top values and how they felt. It names the specific trap your pattern tends to fall into.
C. The shadow guardrail (the standout feature). It knows your golden-mean shadow and your stress loop, and watches for them in real time: “that is the third yes this week, full tank or empty one?” A compassionate pattern-interrupt that hands you your own saved scripts in the moment.
D. Translates the world to you. A communication coach that drafts the hard email in a way that is true to you and still lands for the recipient. When the other person is also typed (couples and teams), it reads both models and runs “why did that land badly” post-mortems through both lenses.
E. Grows you. Coaches your under-used facets with small practices (Whole Trait Theory: you have a default and you can flex it), and tracks your growth edges over time.
F. Authors your story (the recurring hook). AI-guided journaling on the narrative layer, reflecting your arc back to you. This is the living value that justifies a subscription, and it is the “Finish” layer of the signature that visibly moves as you do.
HOW IT GETS SMARTER (the honest retention engine, not lock-in)
Re-tests refine the model. With separate consent, the AI’s observations of you update your characteristic adaptations (goals, coping, roles), never the measured trait spine, which stays fixed and honest. Value increases genuinely over time because the model of you sharpens, which is a real reason to stay, not a dark pattern.
THE RELATIONAL LAYER (network effect and upsell)
When two people both hold profiles, AAVATme can model the relationship: a couples mode (“navigate this fight”), a team mode (a manager’s AI briefed on each member’s pattern). This bridges straight into the AAVATtype couples and teams products, and into AAVATAI.
THE AVATAR BRIDGE (to AAVATtwin and AAVATAI)
AAVATme is the text-and-voice intelligence. AAVATtwin gives it a face and presence rendered from the same signature DNA. A warm AAVATme user is the ideal lead for “now meet the avatar version of you,” which is AAVATAI’s wheelhouse, and the consented profile data grounds AAVATAI persona realism. One profile object: the signature, the personal AI, and the avatar, all from the same genome.
TRUST AND PRIVACY (non-negotiable, a launch gate)
Two separate opt-ins behind a hard data wall: one to use your profile inside AAVATme, a separate one for any AAVATAI dataset use. Both off by default. The profile is sensitive psychological data, so: encryption, deletion on request, no training on identifiable data, GDPR, UK and AU compliant. The honesty spine carries through: measured-versus-interpreted stays labelled, the AI never claims clinical or diagnostic authority, and the standing line is “tendencies, not limits.” Mishandling this kills both AAVATtype and AAVATAI, so it is built in from day one.
PRICING AND PACKAGING
Mode 1 self-file: bundled into the top AAVATtype tier (“Full Pattern + AI”, the ~$49 tier), or sold as a standalone upsell on the $29 core. Near-zero cost to deliver, high perceived value, and it is the single most differentiated thing in the funnel.
Mode 2 hosted companion: a monthly subscription. The 2026 consumer comps cluster at $9.99 to $19.99/mo (Character.ai $9.99, Rosebud $12.99, Replika $19.99); the broad market anchors are ~$20 mass, ~$100 power, ~$200 max. Realistic AAVATme consumer band: $12 to $19/mo IF pursued. BUT see the cautionary tales below: lead with the one-time self-file (proven willingness-to-pay $10 to $33 for a personality artifact), treat the subscription as the later “living” layer, and treat the B2B2C identity layer as the real durable play.
BUILD PATH (decade-compression: ship the cheap, high-value piece first)
Phase A (now, near-zero cost): the self-file generator. One function that turns the profile object into (1) a readable “About Me” manual and (2) a paste-in custom-instructions block. Claude can build this directly against profile-schema.mjs. Ships inside the paid tier.
Phase B (next): the hosted chat companion. Next.js plus the Vercel AI Gateway, profile as system context, persistent memory, the shadow-watch prompts, guided journaling. The subscription product.
Phase C (later): the relational modes (couples and teams), the evolving-signature sync, and the avatar bridge to AAVATtwin and AAVATAI.
The reusable core across all phases is the “self-model compiler”: one function, profile object in, operating manual plus system prompt out. Everything else is a surface on top of it.
HONEST LIMITS (do not oversell)
Mode 1 is a seeded prompt, bounded by the host AI’s memory and context limits, and only as good as that model’s adherence. It is real and useful, but it is not a fine-tuned mind. “Translate other people” needs the other person to be typed too, a genuine cold-start dependency. And the evolving model must never be allowed to drift the measured trait spine.
MARKET AND COMPETITIVE POSITION (from the AI-personalisation research, 27 Jun 2026)
The whitespace, stated plainly. No verified product as of June 2026 owns the intersection of: (1) a multi-lens VALIDATED personality profile, (2) a portable user-owned file, (3) automated, living seeding into any AI, and (4) the same profile as the soul layer beneath an avatar. The portability startups (MCP, Mem0/OpenMemory, Supermemory, MeCP) have the transport rail but carry no psychological model. The personality firms have the science but trap it in static PDFs. The big assistants now ship “memory,” but it is vendor-locked, made of raw facts, and psychologically blind (and their March 2026 “export” is just prompt copy-paste). AAVATme can own the missing piece: the content standard for “who I am” that rides on top of MCP.
The closest competitor, and the proof. Depth Profile (depthprofile.com) sells a profile built on validated instruments (IPIP-NEO, ECR-R, RIASEC) that exports in five formats including ChatGPT Custom Instructions and Claude Projects, for $19 one-time plus $9 to $29 add-ons. This is the thin version of our idea, and it both proves the willingness-to-pay AND shows the exact gap: their integration is manual copy-paste, single-purchase, single-pass, not living, not multi-lens-deep, and has no avatar layer. We beat it on depth (six lenses, not three), on automated and living seeding, and on the avatar soul layer.
The science risk, neutralised by framing. Academic work shows that measuring an LLM’s personality is unstable. That critique is about the AI as the subject. Our framing must always be “a validated HUMAN profile steers the AI”: the human is the measured subject, where Big Five and HEXACO are rock-solid (test-retest above .80, far better than MBTI’s 39 to 76 percent reassignment in five weeks), and the LLM is the steered instrument, where prompt steering and persona vectors already work. That sidesteps the risk entirely.
The retention mechanism, the honest one. The proven causal lever for these products is “feeling heard” (De Freitas et al., Journal of Consumer Research, April 2026), and the relief is satiating, not cumulative, which is why daily-return habit loops and memory continuity work. Memory and continuity are the switching cost. Companion apps retain far better than average at D30, but AI apps overall churn worse (“AI tourists”), so stickiness is conditional on a real bond, not automatic. The ethical flag, which matters for a values-led brand: the stickiest mechanisms (validation loops, guilt and FOMO) are also the most manipulative. Depth retains honestly, flattery is hollow. We design for “feeling heard,” never for the dark patterns, and we lean on real validated depth, not Barnum flattery.
The cautionary tales. Two pure-consumer “kind personal AI” products, Personal.ai and Pi (Inflection), could not sustain a standalone consumer companion-utility and pivoted to enterprise or were neglected. The “AI add-on” model is also dying (Notion folded AI into a seat tier, Google bundles into Google One). Implication for AAVATme: do not bet the product on a fragile standalone consumer subscription. Anchor on the one-time profile artifact first (proven $10 to $33 WTP), make it portable and living, and treat the B2B2C identity layer (idea 5: “Sign in with your Self,” other apps request the consented profile over MCP) as the durable moat.
STATUS
Spec drafted 27 Jun 2026. Pending: Nikki’s read and sign-off, the market and pricing section (AI agent in flight), and a decision on whether the self-file ships inside the $49 tier or as a standalone upsell. IP is Nicola’s, Her Frontier is home and licensee, the AAVATAI-facing version pays licensing back to HF and a royalty to Nicola. Sole author, built solo on her Mac, 27 Jun 2026.