OiMy — Complete Architecture v8.2

DeepSeek V4 Flash as primary companion (60-70%) · Full context injection · Smarter routing · Opus 4.8 master prompt
Core Principle Architecture Flow GPT-5.4 Mini Routing Logic DeepSeek Context Stack DeepSeek vs Kimi Decision Model Stack Status

Core Principle Change — v8.2

BEFORE (v8.1)
25%
messages handled by DeepSeek (greetings + acks only)
NOW (v8.2)
60-70%
messages handled by DeepSeek (full conversational companion)
The key insight: DeepSeek V4 Flash is a capable model with extended context, full personality prompt, and all user context injected. It doesn't need Kimi K2.6 for most family conversations — only for BCT-guided behavior change, complex medical situations, and multi-constraint planning. The GPT-5.4 Mini routing prompt needs to reflect this reality, not artificially limit DeepSeek to greetings.

Full Architecture Flow — v8.2

User
📱 Telegram @chittiemma1_bot
Any family member's message
Step 1
🛡 Pre-Filter Zero LLM · Deterministic
Emergency keywords · Prompt injection guard · Active thread TTL · Onboarding gate
Step 2
~700ms
⚡ GPT-5.4 Mini Router 17KB cached · intent + depth classification
Classifies intent AND conversation depth. Three paths:
🔧 SCRIPTED (zero LLM)
reminder · calendar · list · weather → action directly
💬 DEEPSEEK (60-70% of msgs)
conversation · advice · opinions · support · casual planning · follow-up
🎼 KIMI/HERMES (30-40% of msgs)
BCT coaching · medical expertise · complex multi-constraint planning
Step 3a
DeepSeek
1-3s
🔥 DeepSeek V4 Flash Opus 4.8 master prompt · full context · personality · 400 tokens
Receives: conversation history + 3-tier profile + Honcho synthesis + Instagram signals + personality prompt. Handles emotional support, casual planning, advice, opinions, follow-up, simple questions. Escalates to Kimi only when it determines it needs specialist handling.
Step 3b
Kimi
15-30s
🎼 Hermes + OiMy Engine Kimi K2.6 · BCT · Archetypes · deep reasoning
Reserved for: BCT-guided behavior change · medical/expert guidance · multi-constraint planning (Ethan allergy + Lily picky + David low-carb + 5 days) · structured goal-setting with milestones · elder care protocols
Step 4
🧠 Memory Write-Back Background · non-blocking · both paths
Honcho · Vector embeddings (text-embedding-3-small) · EntityExtractor (session end) · Opus 4.8 deep inference (once, async)

GPT-5.4 Mini Routing Logic — v8.2

The routing prompt is the architectural linchpin. It decides 60-70% goes to DeepSeek vs 30-40% to Kimi. Must be precise about the boundary.

DeepSeek handles (60-70%)

CategoryExamplesWhy DeepSeek
Conversational / emotional"I'm exhausted," "ugh mornings," "feeling overwhelmed," "good morning," "thanks"No specialized expertise needed. Warmth + context = best response.
Simple food/activity suggestions"what should we have for dinner," "any activity ideas for the weekend," "quick lunch for Lily"With context of allergies/preferences, DeepSeek can handle this well.
Opinions and parenting philosophy"what do you think about gentle parenting," "is screen time really that bad," "should I let him quit soccer"DeepSeek has opinions and can engage at depth. No BCT needed.
Follow-up and continuity"did that work btw," "Lily loved the pasta," "Ethan is still refusing"Conversational context is the main ingredient. DeepSeek excels here.
Simple health / reassurance"Lily has a small cold," "Ethan seems tired today"Context-aware reassurance + one referral sentence = done.
Light planning / logistics"help me think through this week," "what should I prioritize," "remind me about X"Conversational planning doesn't need full BCT pipeline.
Personal connection"you've been helpful lately," "I feel like I keep failing at this," "just needed to vent"This IS the relationship. DeepSeek must own it.

Kimi K2.6 / Hermes handles (30-40%)

CategoryExamplesWhy Kimi
BCT behavior change"I want to build a habit of X," "help me stick to Y," "why do I keep failing at Z"Requires BCT technique selection, stage classification, identity framing
Complex medical"Lily has fever 102 for 3 days + ear pain + not eating," "Ethan had a seizure"Requires expert archetype, mandatory referral, clinical pattern recognition
Multi-constraint planning"week meal plan: peanut allergy + picky + low-carb + 5 days + 30min max," "monthly schedule with school + sports + appointments"Requires PlannerArchetype with constraint satisfaction
Expert domainsIEP questions, occupational therapy, speech therapy, elder medication managementRequires specialized skill archetypes with domain knowledge
Structured goal-setting"I want to create a family fitness plan," "help me set up a reading routine"Benefits from Journey archetype, milestone tracking, BCT

DeepSeek V4 Flash — Full Context Stack

What DeepSeek receives at runtime for every response it handles. Total: ~800-1200 tokens well within DeepSeek's context window.

[SYSTEM PROMPT — Opus 4.8 Master Prompt]
  Oi's identity, personality, quirks, backstory (~300 words)
  How to use injected context (~150 words)
  Tone calibration rules (~100 words)
  Anti-sycophancy rules (~50 words)
  10 BAD/GOOD examples (~200 words)

[FAMILY CONTEXT — injected runtime, T1]
  Child: Ethan (11) — peanut allergy, soccer Tuesdays
  Child: Lily (7) — picky eater (beige foods), ear infection
  Pet: Biscuit (labradoodle)
  David: low-carb, works late 3 days/week

[PROFILE — injected runtime, T2+T3]
  Communication: S-type · Social need: belonging · Attachment: anxious
  Key inferences: achievement anxiety (projects onto Ethan), fear-based
  procrastination, strongest when given one concrete next step

[HONCHO SYNTHESIS — injected runtime]
  "User engages best in evenings. Soccer + school are her main stress points
   this week. Responded well to the chicken idea last session. Tends to
   spiral when given too many options."

[INSTAGRAM SIGNALS — injected runtime, when available]
  Follows: meal prep accounts, family lifestyle, home organization
  Aesthetic: warm/neutral tones, practical vs aspirational
  Recent interest: quick weeknight dinners, kids' sports

[CONVERSATION HISTORY — last 8 turns as messages[]]
  User: ugh mornings are a disaster again
  Oi: Yeah, Mondays do that. What's blowing up?
  User: ethan forgot his bag and lily wont eat
  ...

[USER MESSAGE]
  "what do you think about the whole gentle parenting thing"
Key change from v8.1: DeepSeek no longer gets just entity_context (~150 tokens). It gets the full stack above (~800-1200 tokens) — same richness as Kimi K2.6, just handled by a faster model with the right prompt. The prompt explicitly teaches DeepSeek HOW to use each context section.

DeepSeek vs Kimi — Decision Framework

DeepSeek V4 Flash owns
✅ The relationship — knowing who they are
✅ Emotional resonance and presence
✅ Memory-based warmth ("how was that call with your mom?")
✅ Opinions on parenting philosophy
✅ Simple suggestions with context
✅ The daily texture of the relationship
✅ Quick actionable advice
✅ Crisis de-escalation (I see you, here's one thing)
Kimi K2.6 owns
✅ BCT technique selection and delivery
✅ Clinical-level symptom assessment
✅ Multi-constraint planning
✅ Structured habit/goal architecture
✅ Expert domain guidance (OT, speech, IEP)
✅ Journey tracking and milestone work
✅ Anything requiring specialist archetypes

Model Stack — v8.2

Model% of messagesLatencyWhat it handlesPrompt author
openai/gpt-5.4-mini100% (classifier)700msIntent + depth classification — runs on every messageHand-crafted 17KB prompt with expanded DeepSeek routing
deepseek/deepseek-v4-flash60-70%2-4sAll conversational turns — with full personality, context, memory, Instagram signalsOpus 4.8 — master companion prompt
moonshotai/kimi-k2.630-40%15-30sBCT coaching · medical · complex planning · expert archetypesOpus 4.8 SOUL.md + Sonnet 4.6 archetype prompts
anthropic/claude-opus-4-8Once/user async30-60sDeep profile inference (30-50 inferences, all layers) + prompt engineering
text-embedding-3-smallEach turn bg200msVector memory

Status: Live vs In Progress

FeatureStatusNotes
Architecture v8.2 routing design🔄 In ProgressOpus 4.8 writing master DeepSeek prompt + expanded routing logic now
DeepSeek full context injection (history + Honcho + Instagram)🔄 In ProgressWill replace current entity-facts-only injection once Opus completes
GPT-5.4 Mini routing prompt expansion🔄 In ProgressDeepSeek domain: light greetings → full conversational companion
DeepSeek personality/backstory/quirks/opinions🔄 In ProgressMaya-inspired + Claude-style. Rendering now by Opus 4.8.
Instagram enrichment (FullContact + Apify)⚠️ Needs keyFullContact silent lookup wired. Add FULLCONTACT_API_KEY to env.
Honcho workspace⚠️ Needs setupCreate workspace at app.honcho.dev → activates dialectic memory
NPS 7-8 consistently⚠️ In progressPeak 7 achieved. v8.2 prompt changes are the key unlock.
Hermes SOUL.md v2✅ LiveOi backstory, 8 anti-sycophancy rules, 7 warmth rules, 4 relationship modes
3-Tier Profile Router✅ LiveT1+T2+T3 context injection into every Kimi call
deep-inference-v6.txt (Opus 4.8)✅ Live467 lines, 11 layers, 9 frameworks
EntityExtractor✅ LiveGPT-5.4 Mini batch, session end
Rhythm Engine✅ Live8am + 8pm Hetzner crontab
OiMy v8.2 · 2026-05-31 · Note: DeepSeek master prompt pending Opus 4.8 completion — v8.3 will include final prompt + test results