← sources

Home Lab private setup: combine frontier model for reasoing with local model for execution. Does this exist?

source: reddit (reddit:selfhosted)published 2026-11-09 11:16ingested 2026-11-10 02:39

retained text

Hi all,

  
I've searched online and also tried to scout chatting with ChatGPT and Claude but I didn't find anything.

  
I'm looking for existing projects (or prior art) for the following setup:

\*\*Goal\*\*

A homelab container where a cloud frontier model (Claude) does the reasoning and writes a plan, and a local model (Gemma 4 via Ollama running on a 12gb rtx gpu) executes it. The local agent must NOT get a free shell.

\*\*The executor is not fixed.\*\* I already run Hermes Agent and n8n, so those are candidates, but I'm happy to use anything that fits: another agent framework (Atomic Agents, Pydantic AI, LangGraph, smolagents...), a plain Python service with Instructor, an MCP client, or a workflow engine. What matters is the safety model, not the tool.



\*\*Requirements\*\*

no private data should go to Antropic. 

\*\*Environment\*\*

Proxmox (LXC + a Home Assistant VM), Unraid NAS with Docker, Ubuntu server, Home Assistant, Ollama (Gemma 4 26B-A4B; I get valid JSON only with think:false). Hermes Agent and n8n are already installed on the NAS.

  
Thanks in advance

document fetched 2026-11-10 00:28 · canonical https://reddit.com/r/selfhosted/comments/1xrp2t9/home_lab_private_setup_combine_frontier_model_for