OpenClaw On Proxmox
Run OpenClaw in a dedicated Proxmox container, point it at the existing llama.cpp Router Mode stack, and bring the assistant out through Telegram, Discord, and a controlled web edge.
Published January 22, 2025 · Updated May 9, 2026
OpenClaw On Proxmox
Open WebUI is the browser workbench.
OpenClaw is the point where the same local model stack becomes an assistant that can follow you into Telegram, Discord, scheduled tasks, and lightweight agent workflows without turning the Proxmox host itself into the experiment.
This cluster keeps the infrastructure side in Proxmox because that is the actual question here: create the container, install the runtime, point it at the existing llama.cpp Router Mode endpoint, wire channels, and expose the control surface safely.
The behavior pages handle the living assistant model, so this page stays on the deployment boundary and links forward when the setup is done.
If the question is how OpenClaw itself should behave once it is running, continue into the dedicated OpenClaw section.
How It Fits The Existing Stack
Internet (HTTPS)
-> Cloudflare Edge
-> encrypted tunnel
-> CT 201 cloudflare-tunnel (192.168.50.70)
-> CT 106 openclaw (192.168.50.85:18789)
-> CT 102 llama-cpp Router Mode (192.168.50.45:8012)The important boundary is simple:
- OpenClaw does not need GPU passthrough
- inference stays in the existing llama.cpp container
- the assistant gets its own lifecycle, logs, workspace, and restart policy
That separation is what keeps the AI stack understandable.
Recommended Container Shape
| Setting | Value | Rationale |
|---|---|---|
| Container ID | 106 | next clean service slot in the existing lab |
| Hostname | openclaw | descriptive role |
| Disk Size | 10 GB | runtime, workspace, logs, and upgrades |
| CPU Cores | 2 | enough for the gateway and tool execution |
| RAM | 2048 MiB | comfortable headroom for Node.js and skills |
| Swap | 512 MiB | safety margin for spikes |
| OS | Debian 13 | stable modern base |
| Bridge | vmbr0 | same LAN as the rest of the stack |
| IPv4 | 192.168.50.85/24 | stable internal endpoint |
| Gateway | 192.168.50.1 | router |
| Unprivileged | Yes | no reason to widen the blast radius |
What This Cluster Covers
- OpenClaw LXC Build And llama.cpp Integration - create CT 106, install Node.js and OpenClaw, and point the assistant at the existing Router Mode endpoint.
- OpenClaw Channels, Daemon, And Secure Exposure - configure Telegram or Discord, replace the broken user-daemon path with a system service, and expose the control UI through Cloudflare Tunnel.
Continue Into The OpenClaw Section
Once the assistant is running, the project-specific pages take over:
- OpenClaw Architecture, Agents, And Workflows - the current operating map for commander, specialists, household routing, task-board flows, and voice.
- OpenClaw Workspace, Skills, And Use Cases - shape the workspace, install skills, and decide what the assistant should actually do.
- OpenClaw Voice Transcription With whisper.cpp - the dedicated guide for wiring voice notes through
tools.media.audio, a local script, and whisper.cpp. - OpenClaw Workforce, Delegation, And Review Patterns - the current replacement for the older Mission Control workforce material, with updated delegation and review patterns.
- OpenClaw Operations And Troubleshooting - verification, maintenance, structured debugging, and the failure cases worth learning once.
Related Platform Guides
- llama.cpp Router Mode On Proxmox - the model-serving layer OpenClaw depends on.
- llama.cpp Router Mode On Proxmox - the model-serving layer OpenClaw depends on.
- Cloudflare Tunnel On Proxmox - the controlled edge path for the web UI and control surface.
- Open WebUI And Ollama On Proxmox - the complementary browser-first interface for the same local model estate.
- OpenClaw - the project-specific section once the infrastructure question is solved.