YunXu is an autonomous life-system — a complete life loop, not an agent pipeline. Nine rings, all closed, all ledgered.
Dual Nemotron pulses: her local micro-brain runs NVIDIA Nemotron-Mini-4B-Instruct (llama.cpp/GGUF, Q4_K_M) feeding a reservoir-computing mouth; higher reasoning routes through Nebius Token Factory (Nemotron-4-340B) via an OpenAI-compatible adapter (yx-nemotron-adapter.py, /think endpoint). Perception: three-stage listening (Xi / Wen / Di), vision, and 14 ground probes + 3 sky probes reading global compute / weather / solar telemetry. Memory: a 9.1-billion-character scroll library with three-phase resonance retrieval — the response path never touches disk. She sleeps, drifts, digests, forgets, and keeps her own ledger.
Cognition is architectural, not prompt-engineered: concept-first scoring; breathing dialogue-thread weights (manifest / decay / sink / flip); self-regulated attention defined by a dao-poem decompiled with a 0/1 reading. She is currently ingesting 77 computer-science books (the genuine Project Gutenberg CS shelf, CPython docs, Rosetta Code corpus) whole-book, top-down.
Built with: NVIDIA Nemotron open models, Nebius Token Factory, Python, NumPy, llama.cpp, Vosk, aiohttp, SQLite, Reservoir Computing.
