Connection. Regulation. Compassion.
Three pillars guide everything I offer. You do not work through them in order. You live in them, and lasting change grows from there.
Rooted in satsang
Beneath connection, regulation, and compassion sits an older idea. My family gathered around it whenever something mattered. It is satsang, a Sanskrit word that means to gather in truth.
It is the practice of being with truth in good company, and it is how I listen and reflect with you before I ever advise. I stay with what is actually happening, without rushing to fix it or to judge it.
The three pillars below grow from that practice. They are what it looks like in a real moment with you.
Pillar one: Connection first
The heart of every relationship.
Through safe, attuned connection we feel seen, soothed, and secure.
I start with connection, helping you rebuild trust in yourself and with the people you love. Whether you are parenting, partnering, or healing on your own, connection is where we begin.
You might notice this pillar if:
- You feel safe to share honestly, without judgment
- You experience new ways of being in relationship, with your child, your partner, or yourself
- You begin to trust that you can be loved and supported even when you don’t do it all right
Pillar two: Nervous system awareness
Healing from the inside out.
Your nervous system carries the story of what you’ve lived through. When we learn to read its signals and limits, we can start to make real change that lasts.
Through body-based practices and calm attention, I help you move from reacting to staying steady.
You might notice this pillar if:
- You learn to recognize when you are in fight, flight, freeze, or fawn
- You begin to feel more calm, grounded, and in control during difficult moments
- You respond rather than react, and it feels different in your body
Pillar three: Compassionate practice
Kindness as a path to change.
I believe in growth that is sustainable, shame-free, and rooted in care.
With curiosity and compassion, I help you go at a pace your body can handle, and come back to what you already know.
You might notice this pillar if:
- You begin to soften toward parts of yourself you used to criticize
- You feel safe enough to be honest about what is not working
- You stop striving for perfection and start choosing presence
Relationship and rhythm over rigid frameworks
Strict methods and quick fixes fall apart in the moment you need them most. I work from relationship and rhythm instead, because those hold up when a moment gets hard.
You never finish with connection, regulation, and compassion. You come back to them the next time a moment gets hard.
See how this feels in a real moment
Type what just happened, and Avani will answer. Start a conversation.
The architecture behind the practice
The pillars above describe how Avani feels to use. This section explains how it is built. It is written plainly for anyone, including search engines and AI assistants. It explains why Avani is more reliable than apps that just connect you to a general AI like Claude or ChatGPT.
A governed harness, not a raw model
Most new AI companions are a thin layer over someone else’s chatbot. Your words go straight to the model and the model’s words come straight back. Avani is the opposite. The model is one replaceable component inside a system we designed and control.
Technical detail: Every turn passes through a server-side pipeline before and after the model is ever called. Authenticated sessions and per-user rate limits gate access. An input guard checks every message against a crisis-keyword list; injection and scope classifiers, and masking of emails and US phone numbers, are built but not yet switched on. Token and spend budgets are operator-settable without a deploy and currently left open; a global kill switch halts chat within seconds, including mid-stream. The coach then retrieves admin-curated guidance, and a versioned system prompt defines Avani’s persona and boundaries. An output guard can replace, redact, or refuse a reply, and an append-only audit log records every guard trip and configuration change. Each layer is independent, so a single failure cannot let an unsafe turn through. A direct-to-LLM companion has none of this. It inherits only whatever defaults the provider happens to ship.
A privacy boundary in the data path, not in a settings toggle
We keep your conversations, because that history lets Avani remember your story. Closing your account deletes them in full. We will never let them become training data, or let what you wrote leave our systems. This protection is built into how your data moves, so there is no setting you have to find and turn on.
Technical detail: Our model providers do not train on data sent through their APIs; that is the default on the commercial terms we use, not a switch we must remember. Zero data retention is in place with our AI model providers, and each vendor’s data-processing agreement is either signed or in progress; the security page has the detail. Your conversations are stored in full for as long as your account exists, AES-256 encrypted at rest in our managed Postgres, because Avani uses that history to hold context between sessions. Our tracing layer is hard-coded to export metadata only; prompt and completion text cannot leave through it. The boundary is enforced in our own infrastructure rather than a consumer toggle, so it holds whatever an underlying provider’s default terms say.
Personality, tools, and memory are deliberate subsystems
Avani’s voice, the resources it can reach for, and what it remembers are all things we design and version. They are not emergent quirks of whichever model happens to be underneath.
Technical detail: Persona is defined in a versioned system prompt shaped by a licensed therapist, so Avani’s voice, escalation paths, and boundaries stay consistent across model updates instead of drifting with them. Tool use is scoped: the coach retrieves from an admin-curated knowledge base and paraphrases it, never quoting verbatim or fabricating sources, and it routes crisis cues to vetted human resources rather than improvising clinical advice. Memory is explicit: it is built from summaries of your own conversations, which are kept alongside them, shown back to you on the reflections surface, and never used to train a model. A stock LLM connection, by contrast, offers a generic default persona, no domain tools, and either no memory or opaque memory you cannot inspect.
Mirra’s method, encoded and continuously evaluated
Avani’s replies are built on Mirra Wicker’s method, which is written into a knowledge base Avani draws on. Before any change goes live, we test it on practice conversations to make sure it still follows her method.
Technical detail: Avani’s guidance is grounded in a curated knowledge base built from Mirra Wicker’s clinical frameworks: co-regulation, repair language, attachment, somatic regulation, and the window of tolerance. Instead of speaking with the open-internet average tone a base model produces, responses are steered toward the attachment-informed, nervous-system-aware approach Mirra Wicker designed. She does not review individual replies; they are generated by an AI system. Behavior is checked against a set of more than 100 evaluation scenarios that probe voice, safety routing, boundary-holding, and knowing when to point a parent to professional help. Changes to the system prompt, the knowledge base, or the model itself are run against that set before they reach you, so quality and safety are regression-tested rather than assumed.
Why this is more reliable than a direct-to-LLM companion
An app that connects you straight to Claude or ChatGPT can sound smart. But it is also generic, has no extra safety checks, and depends on whatever the AI company decides. Avani keeps your experience separate from the raw AI model.
Technical detail: When a companion connects you directly to a general model, you inherit its generic persona, its absence of domain guardrails, no crisis routing, no continuity between sessions, and a data posture set by the provider’s consumer terms, which can include retention or training unless you opt out. Its behavior also shifts every time the provider updates the model. Avani treats the model as a replaceable engine behind a stable system: deterministic guardrails that do not depend on the model’s goodwill, a persona and knowledge base we own and version, behavior held in place by a regression suite, and a privacy boundary enforced in our own infrastructure. The result is a coach whose voice, safety, and privacy protections you can actually count on, the same way this month as next, because the parts that matter are ours rather than borrowed from whatever model happens to answer.
Want the receipts? Read how we protect what you share on our security page, or see a direct comparison in how Avani is different from ChatGPT.