The box
The computer ZoeyOS runs on, so the same box can be built again.
Read from the machine on 2026-10-07. A case, a cooler, and a power supply are not visible to the software, so they are not listed.
Hardware
| Board | ASUS ROG STRIX X870-I GAMING WIFI. BIOS 1686. |
|---|---|
| CPU | AMD Ryzen 9 9950X. 16 cores, 32 threads. 64 MiB L3. The kernel reports a ceiling of 5756 MHz. |
| Memory | 96 GB, two G.Skill 48 GB DDR5 sticks, part F5-6000J2636H48G. Rated 6000, running at 5600 MT/s. The OS reports 89.7 GiB of that 96 GB. |
| GPU | NVIDIA RTX PRO 6000 Blackwell Workstation Edition. nvidia-smi reports 97887 MiB, driver 595.91.07, CUDA 13.2, link width x16. |
| Disks | Samsung SSD 9100 PRO 2TB, mounted at /data. Crucial T705 2TB (CT2000T705SSD3) holds the root filesystem. |
| OS | Ubuntu 26.04.1 LTS. Kernel 7.0.0-38-generic. |
Software
Versions read from the running programs and their environments the same day.
| Room audio | PipeWire 1.6.2. The microphone is a Seeed reSpeaker XVF3800. Speech recognition listens to that raw input. |
|---|---|
| Ear | rail-stt on Python 3.12.13. TEN VAD 1.0.6.8 is the live detector. Silero VAD 6.2.1 is installed beside it. WeSpeaker 0.0.1 stamps the voice. The recognizer is Audio8 ark-asr 0.6B, int8 ONNX, on CPU, in its own Python 3.12.13 environment (numpy 1.26.4). |
| Voice | Chatterbox Turbo, chatterbox-tts 0.1.7, Resemble AI, MIT. Model ResembleAI/chatterbox-turbo at 749d1c1a. Python 3.14.7, torch 2.14.1, transformers 5.2.0. Perth watermark 1.0.1 is on. Speech leaves at 24 kHz PCM. |
| Announcer | Audio8’s TTS runtime (arktts), Python 3.12.13. Separate from the conversation voice. |
| Language | Both lanes serve Qwen3.8-27B-NVFP4. The reply lane is vLLM 0.30.0. The other lane is vLLM 0.27.1. |
Measured with lscpu, dmidecode -t memory, nvidia-smi, lsblk, findmnt, /etc/os-release, and the version of each running environment.