Two minds. One body. Your data, never leaving the device.
SeekNirvana runs two custom-trained language models entirely on your phone — one that understands your inner world, one that reads your body — guided by epoch signals from your smart ring. No cloud. No compromise.
Ring sensing layer
Your smart ring captures HRV, skin temperature, SpO2, sleep stages, and movement across epochs — discrete windows of physiological truth. SeekNirvana reads these epochs to understand your body's rhythm before making any recommendation.
Soul layer — mind & spirit
A custom-trained 4B language model runs entirely on your device, fine-tuned on contemplative philosophy, therapeutic dialogue, and reflective journaling. It handles emotional processing, spiritual inquiry, and cognitive stress without a single byte leaving your phone.
Body layer — yoga & breathwork
A second on-device model is trained on asana cues, pranayama sequences, anatomy alignment, and somatic regulation. It reads your ring's epoch data to time breathwork suggestions, adjust flow intensity, and correct posture in context.
Orchestration layer
A lightweight intent router decides which model activates based on what you're doing and what your ring just reported. Only one model runs at a time — preserving battery, preventing OOM, and keeping responses fast.
Epoch boundary · signal evaluated
Local SLM stack
Somatic and Cognitive guidance — on-device, always
SOUL LAYER
Soul layer — Qwen 3.5 / 3.6 class
~2.5 GB on-deviceFine-tuned on the Bhagavad Gita, Upanishads, CBT dialogue frameworks, Vipassana and Yoga Nidra scripts, and mindfulness-based stress reduction conversations. When you open the journal or feel overwhelmed, this model meets you — warm, present, and private.
BODY LAYER
Body layer — Gemma 4 E4B
~3.0 GB on-deviceFine-tuned on labelled asana datasets, pranayama breath sequences, Sanskrit pose vocabulary, muscle-group alignment cues, and somatic regulation protocols. When your ring signals elevated stress or shallow breathing, this model responds with body-first guidance.
EPOCH AWARENESS
Epoch-aware recommendations
Rather than reacting to every sensor tick, SeekNirvana works in epochs — windows of aggregated physiological data. When an epoch closes and signals shift, the orchestration layer decides whether to surface a breathwork cue, a grounding prompt, or nothing at all.
ORCHESTRATION
One model at a time, by design
Both models live on your device but only one loads into memory at a time. Soul layer: ~2.5 GB. Body layer: ~3 GB. The swap takes seconds and happens at natural session transitions, not mid-sentence. Battery stays protected. Your data stays local.
The core idea is privacy with usefulness. Soul and Body models stay close to the user while classifying stress states, summarising patterns, and shaping personalised coaching without sending biometric context or journal entries into a remote inference pipeline.
Exclusive dataset, local-only guidance
The distress-support layer is built on a custom training set focused on somatic stress language, bedtime overthinking, sleep disruption, reflective journaling patterns, and de-escalation sequences. That means guidance feels specific for body-first and mind-first distress alike without private data ever leaving the device.
Why the app matters
The app is where raw ring data becomes practice. It lets users review epochs, receive yoga and breathwork timing, journal their inner state, and get private Soul and Body layer guidance before those patterns spill deeper into the night.
Private intelligence that knows the difference between a tired body and an anxious mind.
Most wellness apps send your most sensitive data — how you slept, how you felt, what you wrote — to a remote server. SeekNirvana keeps two custom-trained models on your device so the guidance is personal, the processing is local, and the context never leaves your hands.
Join the early access list