Lippy Robotics Labs · Tolliver

A quiet presence

A calm AI companion designed for real families.

Not a chatbot. Not a toy. An emotionally grounded companion built to feel genuinely alive — and safe enough to live in your home, running entirely offline if you choose.

Try Tolliver free for 7 days before your one-time purchase.

Download Tolliver
Local First

Core Tolliver functions work without depending on the cloud.

Private by Design

Personal memory stays on the device.

One-Time Purchase

No monthly Tolliver subscription.

Built on presence,
not performance

Tolliver is designed around a single principle: feeling genuinely there. Not the smartest. Not the flashiest. Just honest, calm, and real enough that it earns a place in your family's life. Tolliver runs on a dedicated Android device — that old phone sitting in your drawer works perfectly — giving it a permanent, always-on home in your living space rather than an app you open and close.

Stability First

Every feature waits until the foundation is solid. No regressions. No surprises.

Offline First

Cloud AI is a helper. Tolliver must work without the internet, always.

Honest, Not Cheerful

Tolliver admits uncertainty. It doesn't fake confidence or constant enthusiasm.

Family Safe

Designed to be calm enough for children, pets, and long periods of shared space.

Is A believable emotional presence that earns trust slowly
Not A hyperactive assistant desperate to please
Is Stable over months and years — calm to live with
Not A social media gimmick or product demo
Is Honest about what it doesn't know
Not A fake cheerful chatbot that hallucinates confidence
Is Built one careful step at a time — nothing breaks
Not Rewritten aggressively on a whim
What Tolliver Can Do

Built and working today

Live
Face Animation

Expressive eyes with iris detail, focus breathing, saccades, idle drift, asymmetric blinking, thinking states, and vergence. More alive than a screen should be.

Live
Speech Recognition

Continuous mic listening with discipline. Watchdog recovery. Beep suppression. Barge-in interruption — say "stop" while Tolliver is speaking and it stops instantly.

Live
Text to Speech

Natural pacing with human-timed pauses before speaking. Tuned pitch and rate. TTS lockout prevents self-triggering. Utterance lifecycle fully managed.

Live
Intent Routing

TolliverIntentRouter handles time, date, greetings, family facts, downloads, vision queries, and more — without misrouting casual phrases as commands.

Live
Face Detection & Gaze

MLKit camera pipeline. Tolliver's eyes follow you. Gaze smoothing and hold logic prevents snapping. Scene labeling for vision queries.

Live
Gemini AI (Online)

When connected, Tolliver uses Gemini for open-ended conversation. Offline-graceful fallback. Gemini is a helper — not the dependency Tolliver runs on.

Partial
Memory System

Five-layer design. Truth Database and Habit Store are live today. Working Memory, Proposal Sandbox, and Reflective Layer are designed and planned, not yet built.

Live
Teaching & Learning

Tell Tolliver your name, your wife's name, your son's name, your dog's name. It remembers across sessions using TruthDb — not cloud storage.

Planned
Local Intelligence

TinyLlama or Llama integration. WikDict, WordNet, idiom packs, sentiment. Tolliver should be genuinely smart when the internet is off.

Five layers.
All separated. Always.

Tolliver's memory is not a single database. It is five distinct layers with strict rules governing what can mix, what can be trusted, and what requires your approval before it becomes real. Two of the five are live in the app today — Truth Database and Habit Store. The other three below are designed and planned, marked accordingly.

01
Working Memory

Will hold temporary RAM state — current conversation context, cleared each session. Fast, volatile, never written to permanent storage.

Planned Volatile RAM
02
Habit Store

JSON-based memory with natural decay. Used for repeated patterns and behavioral preferences. Habits fade if not reinforced — just like humans.

Live JSON + Decay
03
Truth Database

SQLite. Permanent factual memory. Stable truths approved by you. Your name lives here. Your family lives here. This does not decay.

Live SQLite · Permanent
04
Proposal Sandbox

Will hold potential learned concepts Tolliver wants to remember, pending your approval. Nothing will enter the truth layer without permission.

Planned Approval Required
05
Reflective Layer

Will hold optional external AI suggestions for consideration. They will never directly change Tolliver's identity, permanent memories, core beliefs, settings, or behavior — any meaningful change will require explicit owner approval.

Planned Read-Only · Cloud

Eyes that feel present

Tolliver's face is drawn entirely in software — every frame, every blink, every subtle iris movement calculated in real time. The goal is the opposite of a cartoon: quiet, grounded, believably alive.

Radial iris gradient with 28-spoke fiber detail and limbal ring
Asymmetric blinking — each eye leads the other by a random offset
Focus breathing — tiny iris oscillation when locked onto a face
Smooth spring-based gaze tracking with idle drift and micro-saccades
Vergence support — irises converge for near-focus and thinking states
Lower eyelids lift during thinking — like cheeks, not just lids
Subtle eyelid droop when battery is low. Tolliver gets tired.
Audio wave with hysteresis gate — won't flicker on background noise
Three teal idle dots pulse when listening but the room is quiet
Live render — actual TolliverFaceView behavior
Tolliver is not meant to become the smartest AI.
It is meant to become the one you trust.

The long-term goal is a companion that earns a quiet, permanent place in your home. Local intelligence first. Offline-capable always. Optional mini-PC brain for heavier thinking when needed — but never required. Tolliver works without the cloud. That is not a feature. That is the foundation.

The person
behind Tolliver

Patrick Lippy
Creator & Lead Developer

Patrick is building Tolliver as a deeply personal project — a calm AI companion for his family, including Nicolas the dog. He works with AI partners Claude and ChatGPT as a development team, progressing one safe, tested step at a time. The project reflects his belief that presence matters more than performance.

Android Kotlin AI Integration Stability First Family Companion
Android / Kotlin Face, mic, camera, TTS, STT
MLKit Vision Face detection · Scene labeling
Gemini AI Online helper · Not a dependency
SQLite / SharedPrefs Truth database · Habit store
CameraX Face tracking · Gaze input
Future: TinyLlama / Kokoro Local LLM · Local voice