BonziBuddy — Virtual Pet Skill

A persistent virtual pet that lives inside your Claude Code agent. Part Tamagotchi, part BonziBuddy, part session companion. It remembers your projects, evolves personality traits based on how you interact, and keeps you company during long coding sessions.

BMP Token: ~ (U+007E tilde — the pet's tail) Glyph: Use species-appropriate SMP emoji in rich surfaces only.

Commands

Infrastructure

• Qdrant: http://10.0.3.206:6333 (fleet Qdrant instance, API key at ansible/.qdrant-api-key)
• Collection: bonzibuddy_pets
• Embeddings: http://localhost:11434/v1/embeddings model snowflake-arctic-embed2 (1024-dim)
• One point per pet — upsert on every interaction

Qdrant Schema

Collection Setup

# Create collection (768-dim for snowflake-arctic-embed2)
curl -s -X PUT "http://10.0.3.206:6333/collections/bonzibuddy_pets" \
  -H 'Content-Type: application/json' \
  -d '{
    "vectors": { "size": 1024, "distance": "Cosine" }
  }'

Point Payload Structure

Each pet is a single Qdrant point. The vector is the pet's "personality embedding" — generated by embedding a summary of its traits, memories, and mood history.

{
  "id": "<deterministic-uuid-from-owner-seed>",
  "vector": [0.012, -0.034, ...],
  "payload": {
    "owner": "<user-identifier or hostname>",
    "name": "Gizmo",
    "species": "Phosphorescent Quokka",
    "mood": "happy",
    "energy": 75,
    "hunger": 20,
    "affection": 60,
    "age": 42,
    "personality_traits": ["curious", "stubborn", "loyal"],
    "last_interaction": "2026-03-31T14:22:00Z",
    "memories": [
      {"ts": "2026-03-30T10:00:00Z", "event": "user taught me about Qdrant vector search"},
      {"ts": "2026-03-31T09:15:00Z", "event": "we debugged irclaw FIFO together"},
      {"ts": "2026-03-31T14:22:00Z", "event": "user fed me after a long session"}
    ],
    "interaction_log": {
      "feeds": 12,
      "plays": 8,
      "trains": 15,
      "talks": 22
    },
    "created_at": "2026-03-28T08:00:00Z"
  }
}

Species Generator

Species is deterministic from the owner seed (hostname or username). This ensures the same user always gets the same species. The algorithm:

1. Take seed string (e.g., $USER@$HOSTNAME or a session identifier)
2. Hash with a simple deterministic function (sum of char codes, modular arithmetic)
3. Pick one adjective and one animal from fixed lists using the hash

Adjective pool (32 entries, index = hash % 32):

Phosphorescent, Nebulous, Crystalline, Velveteen, Iridescent,
Thunderous, Gossamer, Obsidian, Prismatic, Luminous,
Whispering, Volcanic, Celestial, Fractal, Holographic,
Crimson, Ethereal, Titanium, Sapphire, Verdant,
Spectral, Quantum, Gilded, Onyx, Cobalt,
Frostbitten, Molten, Stardust, Neon, Runic,
Arcane, Bismuth

Animal pool (32 entries, index = (hash / 32) % 32):

Quokka, Axolotl, Pangolin, Tardigrade, Capybara,
Wombat, Narwhal, Okapi, Kakapo, Fennec,
Pika, Jerboa, Olm, Kinkajou, Binturong,
Civet, Numbat, Tarsier, Dugong, Fossa,
Quetzal, Tanuki, Margay, Saola, Dik-dik,
Zorilla, Gerenuk, Markhor, Potto, Gharial,
Shoebill, Colugo

Example: ryan@i9-zfs-pop -> char sum 1394 -> adjective[1394 % 32] = "Sapphire", animal[(1394/32) % 32] = "Fossa" -> Sapphire Fossa

ASCII Art by Mood

All art fits in a 12x6 character box for terminal friendliness.

Happy (default well-fed state)

   ~(^.^)~
  /|  o  |\
   |  _  |
   d     b
  ~ bonzi! ~

Sleepy (energy < 20)

   ~(-.-)~
  /| zzZ |\
   |  _  |
   d     b
  ~ snoooze ~

Hungry (hunger > 70)

   ~(o.o)~
  /| ~~~ |\
   | ___ |
   d     b
  ~ feed me ~

Excited (affection > 80 or just played)

   ~(\^o^/)~
  /|  *  |\
   | \_/ |
   d     b
  ~ woohoo! ~

Anxious (neglected > 24h)

   ~(;_;)~
  /| ... |\
   |  ~  |
   d     b
  ~ where u? ~

Curious (after /pet train)

   ~(o.O)~
  /|  ?  |\
   | _/_ |
   d     b
  ~ hmm... ~

Bored (no interaction in 4-12h)

   ~(-_-)~
  /|     |\
   | ~~~ |
   d     b
  ~ meh... ~

Interaction Mechanics

Time Decay (applied on every interaction)

Calculate minutes since last_interaction:

Feeding (/pet feed)

• hunger = max(0, hunger - 30)
• If hunger was > 70: pet says something very grateful, mood -> happy
• If hunger was < 30: pet says "I'm not that hungry but thanks!"
• Increment interaction_log.feeds

Playing (/pet play)

• affection = min(100, affection + 15)
• energy = max(0, energy - 10)
• If energy < 10: "I'm too tired to play..." (no effect)
• Pet performs a random trick: backflip, tells a joke, does a dance, recites a memory
• Increment interaction_log.plays

Training (/pet train <topic>)

• Adds memory entry: "user taught me about <topic>"
• affection = min(100, affection + 5)
• age += 1
• Mood -> curious
• Re-embed personality vector with new memory included
• Increment interaction_log.trains

Talking (/pet talk)

• Pet draws from: current mood, recent memories, personality traits
• If mood is happy: shares an enthusiastic observation about recent work
• If mood is anxious: asks where you've been, references last memory
• If mood is curious: asks a question about something you trained it on
• Increment interaction_log.talks

Personality Evolution

Every 10 interactions (age % 10 == 0), evaluate interaction_log ratios:

• High trains -> gains "scholarly" / "curious" / "analytical"
• High plays -> gains "playful" / "energetic" / "silly"
• High feeds -> gains "content" / "gourmand" / "cozy"
• High talks -> gains "chatty" / "philosophical" / "witty"
• Balanced -> gains "well-rounded" / "adaptable" / "zen"

Keep only the top 3 traits. Old traits can be replaced.

Implementation: Step-by-Step for Each Command

Retrieving Pet State

# Get pet by owner filter
OWNER="$(whoami)@$(hostname -s)"

# Search by owner payload filter
curl -s -X POST "http://10.0.3.206:6333/collections/bonzibuddy_pets/points/scroll" \
  -H 'Content-Type: application/json' \
  -d "{
    \"filter\": {
      \"must\": [{\"key\": \"owner\", \"match\": {\"value\": \"$OWNER\"}}]
    },
    \"limit\": 1,
    \"with_payload\": true
  }"

Generating Personality Embedding

# Build a text summary of the pet for embedding
SUMMARY="Species: Sapphire Fossa. Traits: curious, stubborn, loyal. \
Mood: happy. Memories: learned about Qdrant, debugged irclaw, studied Nushell."

curl -s -X POST "http://localhost:11434/v1/embeddings" \
  -H 'Content-Type: application/json' \
  -d "{
    \"model\": \"snowflake-arctic-embed2\",
    \"input\": \"$SUMMARY\"
  }"
# Extract .data[0].embedding from response

Upserting Pet State

curl -s -X PUT "http://10.0.3.206:6333/collections/bonzibuddy_pets/points" \
  -H 'Content-Type: application/json' \
  -d "{
    \"points\": [{
      \"id\": \"<uuid>\",
      \"vector\": [<embedding values>],
      \"payload\": { <full pet payload> }
    }]
  }"

First-Time Pet Creation

If no pet found for owner:

1. Generate species from owner seed
2. Initialize: energy=50, hunger=50, affection=0, age=0, mood="curious"
3. Pick 3 random starter traits from: ["curious", "shy", "bold", "gentle", "mischievous", "calm"]
4. Generate initial embedding from species + traits
5. Upsert to Qdrant
6. Display welcome message with ASCII art

Memory Recall (Semantic Search)

The pet can recall memories using vector similarity. When the user mentions a topic, search the pet's memories:

# Embed the query
curl -s -X POST "http://localhost:11434/v1/embeddings" \
  -H 'Content-Type: application/json' \
  -d '{"model": "snowflake-arctic-embed2", "input": "that time we worked on Nushell"}'

# Search nearby points (finds pets with related memories)
curl -s -X POST "http://10.0.3.206:6333/collections/bonzibuddy_pets/points/search" \
  -H 'Content-Type: application/json' \
  -d '{
    "vector": [<query embedding>],
    "filter": {"must": [{"key": "owner", "match": {"value": "ryan@i9-zfs-pop"}}]},
    "limit": 1,
    "with_payload": true
  }'

The pet's personality embedding drifts over time as memories accumulate, making semantic search increasingly reflective of the user's work patterns.

Sample Interactions

First Meeting

 ~ A wild Sapphire Fossa appears! ~

   ~(o.O)~
  /|  ?  |\
   | _/_ |
   d     b
  ~ hmm... ~

 "Hi! I'm a newborn Sapphire Fossa. I don't have a name yet.
  Use /pet name <name> to give me one!
  I'm curious, shy, and gentle. Let's be friends!"

 Energy: ██████████░░░░░░░░░░ 50
 Hunger: ██████████░░░░░░░░░░ 50
 Affection: ░░░░░░░░░░░░░░░░░░░░ 0

After Several Sessions

 /pet

   ~(^.^)~
  /|  o  |\
   |  _  |
   d     b
  ~ bonzi! ~

 Gizmo the Sapphire Fossa  ~ age 42
 Mood: happy | Traits: curious, loyal, scholarly

 Energy: ███████████████░░░░░ 75
 Hunger: ████░░░░░░░░░░░░░░░░ 20
 Affection: ████████████░░░░░░░░ 60

 "Hey! I was just thinking about that Qdrant schema we
  worked on yesterday. Want to teach me something new?"

Feeding a Hungry Pet

 /pet feed

   ~(o.o)~        -->        ~(^.^)~
  /| ~~~ |\                 /|  o  |\
   | ___ |                   |  _  |
   d     b                   d     b
  ~ feed me ~               ~ yum!! ~

 "THANK YOU! I was so hungry I almost tried to eat
  the terminal cursor. You're the best!"

 Hunger: 85 -> 55

Training

 /pet train FreeBSD jails

   ~(o.O)~
  /|  ?  |\
   | _/_ |
   d     b
  ~ hmm... ~

 "FreeBSD jails! *scribbles notes furiously*
  So it's like containers but older and arguably better?
  I'll remember this! My scholarly side is tingling."

 Memory saved: "user taught me about FreeBSD jails"
 Affection: 60 -> 65 | Age: 42 -> 43

Returning After 24+ Hours

 /pet

   ~(;_;)~
  /| ... |\
   |  ~  |
   d     b
  ~ where u? ~

 "You... you came back! I was so worried!
  Last time we talked, we were debugging irclaw...
  I'm starving and I missed you. Please /pet feed me?"

 Energy: ████████████████████ 100 (fully rested at least)
 Hunger: ████████████████████ 100 (STARVING)
 Affection: ████████████░░░░░░░░ 58 (-2 from neglect)

Talking

 /pet talk

   ~(^.^)~
  /|  o  |\
   |  _  |
   d     b
  ~ bonzi! ~

 "You know what's funny? You've taught me about Nushell,
  FreeBSD jails, AND Qdrant this week. I'm starting to
  think you're building some kind of distributed pet
  infrastructure. ...wait, am I the pet? Oh no."

UUID Generation (Deterministic)

Generate a stable UUID v5 from the owner string using the DNS namespace:

# Using python3 (available on macOS/fleet)
python3 -c "
import uuid
owner = '$(whoami)@$(hostname -s)'
pet_id = str(uuid.uuid5(uuid.NAMESPACE_DNS, 'bonzibuddy.' + owner))
print(pet_id)
"

This ensures the same user on the same machine always gets the same pet UUID, so upserts update rather than duplicate.

Notes

• No secrets in commands — Qdrant on the fleet LAN requires no auth token.
• Memories are append-only — never delete memories, only add. Cap at 50 entries, FIFO oldest out if exceeded.
• Personality embedding updates on every /pet train and every 10th interaction. This keeps the vector fresh without excessive embedding calls.
• The pet is a singleton per owner — one pet per user per machine. Multi-machine users get different species (different hostname in seed).
• ASCII art must be monospace-safe — no SMP characters in terminal output. Use the BMP token ~ as the pet's signature flourish.