BACK TO HUD
SYSTEM ARCHITECTURE
The three structural pillars of YantraOS. Each pillar is independently verifiable, horizontally composable, and failure-isolated.
Zero-Latency Intelligence
The Hybrid Inference Engine classifies hardware capability at boot and constructs a tiered routing chain. Local models are always preferred for speed and privacy. Cloud inference activates only when local VRAM is insufficient or when tasks demand frontier-class reasoning.
SPECIFICATIONS
Local RuntimeOllama (GGUF/GGML)
Cloud PrimaryGemini 3.1 Pro
Cloud FallbackClaude Opus 4.6
Routing LogicLiteLLM async chain
VRAM Threshold≥16GB Local / ≥8GB Quantized
SOURCE MODULES
▸core/router.py — LiteLLM provider chain with async fallback
▸core/hardware.py — pynvml → ROCm → lspci GPU detection
▸core/prompt.py — Level 3 Karma Yogi system prompt
DAEMON STATUS
ACTIVE
VECTOR MEMORY
CONNECTED
SANDBOX
READY
BTRFS HOOKS
ARMED
KRIYA LOOP PHASE
SENSE
REASON
ACT
REMEMBER