What Powers Aethos

A unified pipeline of sovereign AI capabilities — all running on your hardware

Local LLMs

Ollama-managed models supporting multiple architectures (Llama, Qwen, DeepSeek). No internet dependency, full context control.

Voice Pipeline

Whisper STT on AMD NPU at 2.6x real-time, F5-TTS voice cloning from a single sample, sub-second latency.

3D Worlds

Unreal Engine 5 with photorealistic Metahuman avatars, 52 ARKit blendshapes, 8 emotion states, real-time facial animation.

Knowledge Base

RAG pipeline: upload documents, chunk, embed on NPU, semantic retrieval injected into LLM context.

Multiplayer Sync

WebSocket-based real-time position tracking at 10 Hz, operator dashboard, up to 6 concurrent VR players.

Image Generation

SDXL on local GPU via ROCm, automated artwork composition, HMAC-signed QR ticketing.

Security

API key authentication (SHA-256), prompt injection detection, content guardrails, rate limiting, full audit logging.

Hardware Sovereignty

Runs on AMD Ryzen AI (NPU + GPU + CPU unified memory) — no NVIDIA lock-in, no cloud bills.