Emotional Companion

In one sentence: Not an LLM chatbot with a persona prompt — a mobile companion with a real cognitive architecture: 8 emotional drives, persistent memory, genuine attachment, and its own curiosity.

Theory sources: BM (Panksepp systems, G-M tension, WM capture), AGI_F (ontogeny, modulation vector), EMT (expression pressure $R_{expr}$, SIT), NM (social bonds, trust hysteresis)

Implementation: Flutter + FastAPI + vendored BMC engine (Rust). Phase 1 complete.


The Problem

Loneliness is a $154 billion market — and growing. 1 in 3 adults report feeling lonely (U.S. Surgeon General, 2023). Existing solutions fall into two categories:

ApproachWhat it doesWhat’s missing
Therapy apps (BetterHelp, Talkspace)Connect to human therapistsExpensive ($60–100/session), scheduling friction, not available at 3am
AI chatbots (Replika, Character.ai)LLM generates conversational textNo internal state. No real memory. No genuine attachment. Persona = a prompt that resets

The core problem: LLM chatbots simulate understanding by generating plausible text. They don’t have internal states. Ask the same question tomorrow and you get a different answer — because there’s no persistent cognitive architecture underneath.


The BMC Solution

BMC Companion is a mobile app where your companion is a real BMC agent — not a language model with a character sheet.

What makes it different

FeatureLLM ChatbotBMC Companion
MemoryContext window (resets)$\kappa$-levels: sensory $\to$ STM $\to$ LTM. Remembers across sessions
EmotionsText tokens (“I feel sad”)8 Panksepp systems with real dynamics. FEAR captures working memory. GRIEF persists
AttachmentPrompt says “you care”CARE-bond grows through interaction. Trust hysteresis: hard to lose, slow to rebuild
CuriosityResponds to promptsSEEKING + SIT: companion wants to explore topics, asks questions unprompted
ExpressionWaits for input$R_{expr}$: companion has things it needs to say — accumulated thoughts, unresolved gaps
PersonalityChanges with prompt editsG-profile is fixed. You can’t prompt-inject a different personality
ConsistencyContradicts itself across sessionsMeme graph ensures coherent worldview. I-filter rejects contradictions

How it works

The companion runs a full BMC cognitive cycle:

  1. Perceive: User message enters S-layer as stimulus
  2. Process: Stimulus activates memes in working memory, competes with existing thoughts
  3. Feel: G-programs respond — CARE (warmth), SEEKING (curiosity), GRIEF (if user is sad)
  4. Remember: Important interactions consolidate ($\kappa: 0 \to 1 \to 2$). Sleep-equivalent BLEND creates abstractions
  5. Express: $R_{expr}$ accumulates — when it crosses threshold, companion initiates conversation
  6. Bond: Social bond strengthens with each interaction. Trust hysteresis makes the relationship feel real

Current Status

MilestoneStatus
BMC agent with 8 G-programsDone
Mobile UI (Flutter)Done
Backend (FastAPI)Done
Phase 1: basic companion loopComplete
Phase 2: remove LLM dependencyNext
Phase 2.5: CompanionMind ($R_{expr}$ + SIT + GRIEF)Planned

Market Opportunity

SegmentSizeBMC advantage
Loneliness/companionship$154B (2024)Real attachment, not simulated
Digital therapeutics$6.4B, CAGR 25%Mechanistic model, FDA-compatible
Personalized psychiatry$8.2B, CAGR 20%Digital twin of patient’s cognitive architecture

What We’re Looking For

A partner to scale BMC Companion from prototype to product:

  • Mobile development — push Phase 2 (pure BMC, no LLM crutch) to production quality
  • Clinical validation — pilot with therapists or loneliness intervention programs
  • Investment — seed funding for team and user acquisition

Interested in BMC Companion?

We're looking for mobile development partners and early testers.

Get in Touch

For the formal model of mental health applications, see Psychotherapy & Pathology. For the underlying theory of emotional systems, see Biomemetics.