Mirror · product roadmap
The product's path bottom-up: what's already standing on its own feet, what's being worked on right now, what's next — and where it all ultimately leads.
Tap a card — it opens like doors and shows the details.
client feature therapist feature both sides
The core of the product. The client picks a therapist persona and keeps a free-form diary — by text or voice. Between replies the system keeps the conversation in a natural rhythm, without checklists or questionnaires.
A smart dialogue-management system has been built in, closer to the format of a real therapy session: not every reply is a probing question. Sometimes the therapist just listens, sometimes it mirrors the person's state out loud, naming what it noticed, sometimes it genuinely deepens the topic. A living conversational rhythm instead of the "emotion + question" template on every reply. Tuning on live dialogues is ongoing.
After sessions, a client portrait is assembled: recurring themes, active requests, progress. Not dry statistics, but a clear picture of what's happening with a person.
MAC (Metaphorical Associative Cards): the bot generates an abstract image for the client's current state, and a short reflective dialogue is built around it — "what resonates with you here." The card is saved in the client's history alongside regular sessions.
Every model reply costs money. Yaloms are the in-app currency: charged for every LLM call, each user has a balance/limit. Needed for monetization and protection against uncontrolled spend. A previous attempt (the add-yaloms branch) is being redone from scratch with lessons learned.
The therapist fills out a questionnaire by voice or text: price, location, format of work, direction, experience, contacts. AI assembles a profile card from that — the therapist can accept it as-is or edit it manually. Moderated in the admin panel before publishing.
The client briefly answers questions about their request — location, budget, gender preference, direction, what the problem is. The LLM cross-references this with the client's full profile (session history, portrait, patterns, active requests) and with every therapist's profile card, and computes a % fit for that specific person — with a personal reason addressed directly to them ("this will help you with...", not generic phrasing). The result is a paginated match list with the therapist's direct contacts.
The therapist manages their clients right in the bot: adds a client, writes notes on them, requests an AI report on a client or a recommendation on which techniques to bring into a session, at any time. A basic version is already working.
Diary entries and client correspondence in the database need to be encrypted so their content is unreadable — not only in case of a leak, but to the app's own administrators too. Personal stays personal. required
Moving LLM request processing to local servers in the user's region — Belarus, Russia, Kazakhstan — instead of a single cloud model. Client data doesn't leave their country's jurisdiction, latency is lower, and dependency on a single external provider is reduced.
We'll aim for the same level of protection people are used to from messengers like Telegram: minimal data stays on the server, most stays on the user's side, active sessions can be viewed and logged out of at any time. While the product is in "home app" mode this is intentionally deferred — priority returns once there are real users.
The system detects a client's significant pain points — a childhood incident, a relationship pattern — and keeps them as "bright memories" (there can be several at once, not just one). Future dialogues draw on them, the way a therapist keeps a client's key theme in mind across sessions. strong differentiator
Continuation of phase 1: testing the rhythm on live sessions before moving on to therapist memory.
Access without Telegram, a richer interface — a future entry point for the therapist marketplace as well.
The next step after Mini App — standalone apps (iOS, Android, a full web app) not tied to Telegram as the only entry point. Requires multi-channel registration (several sign-in methods for one account), which the system is already preparing for at the yaloms level.
For well-known therapists — books, articles, recordings of public sessions. For Belarusian therapists — their scientific publications, notes, audio or text records with clients (with consent). This forms the basis for a RAG "likeness" of a specific therapist — which can be brought into a client dialogue on its own or blended with the main persona.
Expanding the persona lineup — more "character" options for a therapist to fit different clients. Beyond the four standard personas, we'll generate several neuro-personalities of well-known Belarusian therapists — and see how diary sessions unfold in their voice.
Connecting a diary client with a real, live therapist — from matching to joint work between offline sessions.
When offline work is already underway, the therapist sends a "link" invite with the client's consent. The client sees: "therapist N is offering to become your official therapist" — and confirms. This unlocks joint functionality around real sessions.
The client keeps journaling, and the system knows the meeting schedule. 24 hours before a session it offers to send the therapist a short brief on the week — what happened, concise metrics like anxiety level. The client sees exactly what will be sent and confirms it themselves.
The system learns which techniques worked with which clients. The therapist sees their effectiveness by topic. This data raises the recommendation weight for similar future clients — and aggregated analytics across the whole platform shows which techniques work in which situations, forming the basis of recommendations for other therapists.
Shapes architecture decisions already.
The diary is both a sensor of state and a measure of outcome. A therapist/intervention is matched to a client's profile on the principle of evidence-based medicine: "this is what helped clients like this." key feature
A web dashboard: a pre-session brief, pattern search across the client's full history, detection of gestalt contact-interruption mechanisms. A second market — B2B.
The more data accumulates about clients, their therapy, and sessions, the more precisely the system finds past clients similar to a new one. And recommends the exact techniques that actually helped those similar cases, not generic advice. The same principle as YouTube's recommendation system — except instead of videos, it's therapy techniques that work.
Connecting wearables and health services — sleep quality and duration, stress level, other medical indicators, and eventually uploading medical test results to track trends. Complements the "Healing Navigator" — the body's state as another signal alongside the diary, for a more precise match and a better read on the effect of working with a therapist.
Why this exists
The industry has already taught AI to recognize photos, then voice and tone. But how the human psyche changes and transforms in real interaction — in sessions with a therapist, in a diary, not in a textbook description — nobody has that yet. Mirror, at scale, is exactly that dataset.
Which therapists and what depth we connect to the system directly colors its future techniques and character. And we use this dataset not as a data warehouse, but as a tool: to guide people toward a more conscious, happy, and fulfilled life.
AI still can't truly feel, and may never learn to. But a tool purpose-built to study human feeling — pain, joy, happiness — doesn't exist either. Digitized therapist↔client sessions are exactly that kind of data.
We believe a feeling AI is the next milestone in the evolution of intelligence. Whoever assembles that dataset faster and deeper will shape how the future intelligence — American, Chinese, or European — understands and feels people. This isn't just a product — it's participation in the race to define what the mind of the future will be.
We're building the neuro-brain of psychological help — an intelligence trained on the most effective therapy experience, to become a real therapist for everyone.