---
created: '2026-04-12T05:45:57+00:00'
id: 42ee6aab-1bf5-451d-add5-0316541736e3
keywords:
- ana-ml2
- 10.250.50.54
- VM-102
- 10.250.50.70
- docker-host
- correction
- kokoro
- vibevoice
- parakeet
- llama-swap
- comfyui
- chromadb
- dockge
- port-allocation
- infrastructure-correction
links: []
modified: '2026-04-12T05:45:57+00:00'
summary: 'Correction: GPU inference Docker services (llama-swap, ComfyUI, VibeVoice,
  Parakeet, ChromaDB, Dockge) run on ana-ml2 (10.250.50.54), not VM 102. VM 102 runs
  only Synapse, Vaultwarden, and SillyTavern. Also records Kokoro TTS being added
  to ana-ml2 on port 8765.'
tags:
- infrastructure
- correction
- ana-ml2
- VM-102
- docker
- kokoro
- port-map
title: 'Infrastructure Correction: Docker Host Assignment'
---

# Infrastructure Correction: Docker Host Assignment

> **Created:** 2025-07-11
> **Status:** Active correction
> **Affects:** All Infrastructure-PFI documents referencing Docker service placement

## Summary

Multiple documents in the Infrastructure-PFI set incorrectly state that the Docker host running GPU inference services is VM 102 (`10.250.50.70`). The actual host is **ana-ml2 (`10.250.50.54`)**.

## Corrected Service Assignment

### ana-ml2 (10.250.50.54) — Bare Metal, Docker Host for GPU Inference

| Service | Port | GPU | Image | Status |
|---|---|---|---|---|
| Dockge | 5001 | No | `louislam/dockge:latest` | Running |
| llama-swap | 9292→8080 | Yes | `ghcr.io/mostlygeek/llama-swap:cuda` | Running |
| ComfyUI | 8188 | Yes | `mmartial/comfyui-nvidia-docker` | Running |
| ChromaDB | 8000 | No | `chromadb/chroma:latest` | Running |
| VibeVoice TTS | 8745 | Yes | `eworkerinc/vibevoice:latest` | Running |
| Parakeet STT | 8300→8000 | Yes | `parakeet-stt` (local build) | Running |
| **Kokoro TTS** | **8765** | **Yes** | **`ghcr.io/hexgrad/kokoro-onnx:latest`** | **New — being added** |

### VM 102 (10.250.50.70) — Proxmox VM, Non-GPU Application Host

| Service | Port | Notes |
|---|---|---|
| Synapse (Matrix) | 8008 | Matrix homeserver |
| Element Web | 8080 | Matrix client |
| AIPA Matrix Bridge | 8009 | Appservice bridge |
| Vaultwarden | — | Password vault |
| SillyTavern | — | LLM chat frontend |

## Kokoro TTS — New Service Record

- **Host:** ana-ml2 (10.250.50.54)
- **Port:** 8765
- **GPU:** Yes
- **Image:** `ghcr.io/hexgrad/kokoro-onnx:latest`
- **API:** OpenAI-compatible TTS API (`/v1/audio/speech`)
- **Status:** Being added to the stack

## Documents Known to Contain the Error

The following Infrastructure-PFI documents describe VM 102 as the Docker host for GPU inference services and need correction:

- `Infrastructure-PFI-Project-Index.md` — VM 102 service table lists all GPU services
- Any references to compose files at `/opt/docker/compose/` should be attributed to ana-ml2, not VM 102
- Port allocation tables should be split between the two hosts

## Port Allocation — Corrected

### ana-ml2 (10.250.50.54)

| Port | Service |
|---|---|
| 5001 | Dockge |
| 8000 | ChromaDB |
| 8080 | llama-swap (internal) |
| 8188 | ComfyUI |
| 8300 | Parakeet STT |
| 8745 | VibeVoice TTS |
| 8765 | Kokoro TTS |
| 9292 | llama-swap (external) |

### VM 102 (10.250.50.70)

| Port | Service |
|---|---|
| 8008 | Synapse |
| 8009 | AIPA Matrix Bridge |
| 8080 | Element Web |