████████╗██╗ ██╗███████╗
╚══██╔══╝██║ ██║██╔════╝
██║ ███████║█████╗
██║ ██╔══██║██╔══╝
██║ ██║ ██║███████╗
╚═╝ ╚═╝ ╚═╝╚══════╝
██████╗ ██████╗ ██╗ ██████╗ ██╗ ██╗██╗███╗ ██╗
██╔══██╗██╔═══██╗██║ ██╔══██╗██║ ██║██║████╗ ██║
██║ ██║██║ ██║██║ ██████╔╝███████║██║██╔██╗ ██║
██║ ██║██║ ██║██║ ██╔═══╝ ██╔══██║██║██║╚██╗██║
██████╔╝╚██████╔╝███████╗██║ ██║ ██║██║██║ ╚████║
╚═════╝ ╚═════╝ ╚══════╝╚═╝ ╚═╝ ╚═╝╚═╝╚═╝ ╚═══╝
BRICK'S AFTER HOURS
// self-hosted voice chat — no leash, no guardrails, all dolphin //
THE VOICE LOOP
YOU SPEAK
→
WHISPER
→
DOLPHIN
→
ELEVENLABS
→
YOU HEAR
→
∞
// INFRASTRUCTURE
picass0 (brain)Tesla T4 · 16GB VRAM · g4dn.xlargeoffline
Rocky (ears+mouth)Windows 11 · mic + speakersonline
Tailscale mesh100.127.18.29 ↔ 100.101.108.7connected
Hugging Facetoken cached · API readyready
ElevenLabsTTS API · voice TBDneed key
// MODEL PICKS (16GB VRAM budget)
| model | quant | vram | vibe |
|---|
| dolphin-llama3-8B | Q4_K_M | ~6GB | ★ top pick — spicy, uncensored |
| Hermes-2-Pro-Mistral-7B | FP16 | ~14GB | smart, less filtered |
| dolphin-mistral-7B | Q4_K_M | ~5GB | OG dolphin, solid |
| Mistral-7B-Instruct | FP16 | ~14GB | vanilla, good LoRA base |
// THE STACK
BrainOllama → dolphin-llama3-8B on picass0
Earsfaster-whisper STT on Rocky
MouthElevenLabs TTS API
FlavorHF LoRA adapters via peft/mergekit
GluePython voice loop on Rocky
// BUILD ORDER
1Get the brain online
Wake picass0 · install Ollama · pull dolphin · test REST API
2Give it a mouth
ElevenLabs API key · pick a voice · TTS script
3Give it ears
Install faster-whisper · mic capture · STT script
4Wire the loop
Full cycle: listen → transcribe → infer → speak → repeat
5Season to taste
LoRA shopping · voice clone · wake word · memory
// COSTS
picass0 runtime~$0.53/hr on-demand
ElevenLabsfree tier 10k chars/mo · $5/mo for 30k
Hugging Facefree
Ollama / Whisperfree / open source
// VOICE PREFERENCES
vibegay guys / twinks
sourceElevenLabs voice library · community uploads
stretch goalvoice clone
built by brick // .brock.1777698487 // "hey dolphin"