Skip to content

Latest commit

 

History

History
100 lines (72 loc) · 2.77 KB

File metadata and controls

100 lines (72 loc) · 2.77 KB

Orbit AI

The local AI assistant — featuring LLM, speech-to-text, and text-to-speech, all running locally with zero cloud dependency — powered by the VOLT AI pipeline.


Modes

Mode Strategy Use Case
Fast Self-consistency Quick answers, general chat
Smart Chain-of-thought Complex reasoning, deep analysis
Code Chain-of-thought Programming, debugging
Data-science Chain-of-thought Analytics, metrics, experiments

Voice Mode

Full voice interaction loop with interruption support:

1. VAD detects speech (Web Audio API)
2. Whisper STT transcribes to text
3. Qwen3 LLM generates response (runs under VRM-managed quota)
4. OuteTTS reads response aloud
5. Loop back to listening

Speaking while Orbit is responding cancels current TTS playback and starts listening again.


AI Model Locations

All models are stored under /opt/volt/ai-models/:

Model Path Type VRAM
Qwen3 1.8B /opt/volt/ai-models/qwen3-1.8b-q4.gguf LLM 2048 MB
Whisper Small /opt/volt/ai-models/whisper-small.gguf STT 512 MB
OuteTTS /opt/volt/ai-models/outetts-q4.gguf TTS 512 MB
Tesseract /opt/volt/ai-models/tesseract.traineddata OCR 512 MB

Architecture

OrbitApp (React)
├── useOrbitChat (hook)
├── useVoiceMode (hook)
├── ChatInput (text + voice mic button)
├── MessageBubble (chat messages + TTS read-aloud)
├── ThreadSidebar (chat history)
├── ModeSelector (Fast / Smart toggle)
└── ModelStatusCard (model status indicator)

VOLT Integration

Component Integration
VRM Registers LLM (2048 MB), STT (512 MB), TTS (512 MB) quotas — all critical priority T1
DWP 4 dedicated workers; burst config: 100ms window, 75% worker pool, 3 max consecutive bursts
VGM Logs GPU metrics after each generation
Unifier OrbitBridge — dedicated unification channel for AI requests routed via the Volt Unifier

Resource Quotas

Resource Allocation
RRAM (LLM) 2048 MB — critical T1
VRAM (STT) 512 MB
VRAM (TTS) 512 MB
VRAM (Tesseract OCR) 512 MB
DWP Workers 4 dedicated
Burst window 100 ms
Burst worker cap 75% of pool
Max consecutive bursts 3

Related



← Back: Documentation Index