OpenClaw vs Hermes: AI Agents for Home Assistant with Gavin Campbell – HGG680

Posted on by

What is HGG680 about? Jim Collison is joined by Gavin Campbell for Home Gadget Geeks 680 to compare OpenClaw and Hermes as AI agents for Home Assistant, with practical discussion around local versus cloud LLMs, privacy, token costs, Mattermost, Docker, Unraid, and the hardware choices that matter in a real home lab.

Audio Only
Apple PodcastsGoogle PodcastsPlayer EmbedShareLeave a ReviewListen in a New WindowDownloadSoundCloudStitcherSubscribe on AndroidSubscribe via RSSSpotifyAudioboomSpreakerOther Options

Quick answer: HGG680 compares OpenClaw and Hermes as practical AI-agent options for Home Assistant and home-lab workflows. The main takeaway is to start with read-only monitoring, clear approval gates, isolated containers, and small specialized agents before letting automation make changes.

Products, Platforms, and Entities Mentioned

  • OpenClaw: AI-agent workflow platform used here for monitoring, routing, and guarded automation experiments.
  • Hermes: Gavin Campbell’s container-friendly agent setup for Unraid and home-lab tasks.
  • Home Assistant: Smart-home platform discussed as a natural place for read-only AI observation before any control workflow.
  • Docker and Unraid: Isolation and deployment layers that make agent experiments easier to contain and recover.
  • Mattermost and local LLMs: Supporting tools for chat-driven workflows, routing, and lower-cost local inference.

Episode Overview

In this episode, Jim and Gavin look at how AI agents can make Home Assistant more conversational and more useful, while still requiring thoughtful choices around hosting, models, security, and cost.

The conversation covers local versus cloud LLMs, privacy, cost, performance, token economics, Mattermost integration, Docker and Unraid deployment tips, and hardware requirements for running AI at home.

The big question is practical: when does it make sense to run an AI assistant close to your home automation stack, and when is a cloud model still the better tool for the job?

Key Takeaways

  • OpenClaw and Hermes both point toward more conversational home automation, but they approach the problem differently.
  • Local LLMs can improve privacy and control, while cloud models often provide stronger performance and easier access to newer capabilities.
  • Token usage, context windows, and repeated automation checks matter when an AI agent runs in the background.
  • Docker and Unraid can be useful deployment surfaces, but persistent storage, networking, backups, and permissions need attention.
  • Mattermost can act as a practical notification and command surface for home-lab agents.
  • Hardware planning still matters, especially when deciding whether local inference is realistic for a given workload.

Featured Topics

OpenClaw and Home Assistant

Gavin explains how OpenClaw can sit alongside Home Assistant to provide a natural-language interface for smart-home awareness, automation, and control. The discussion includes context handling, plugin-style extensibility, and the value of keeping home-lab workflows adaptable.

Hermes Agent Comparison

Hermes brings a different approach, with a focus on simpler setup and chat-driven interaction. Jim and Gavin compare where Hermes may fit well, especially for listeners who want practical notifications and control through Mattermost.

Local vs Cloud LLMs

The episode weighs the strengths and limits of running models locally against using cloud-hosted models. Privacy and data control favor local models, while cloud models can provide stronger performance, larger context windows, and easier access to newer capabilities.

ConsiderationLocal LLMsCloud LLMs
PrivacyMore control over home dataData is processed by a provider
CostHardware up front, lower API spendOngoing token-based usage
PerformanceDepends on local hardwareAccess to larger hosted models
ReliabilityCan work without cloud availabilityDepends on internet and provider status
Model choiceLimited by memory and computeBroad access to current models

Token Economics

Jim and Gavin discuss how context windows, conversation history, tool output, and repeated automation checks can affect token usage. For a home-lab agent, managing what the model sees can matter as much as choosing the model itself.

Docker, Unraid, and Hardware

The conversation includes practical setup guidance for Docker and Unraid environments, including persistent storage, networking, resource allocation, backup planning, and using tools like Can I Run AI? to evaluate hardware for local inference.

Mattermost Integration

Mattermost comes up as a useful surface for notifications, collaboration, command-style interactions, and logging. For home-lab users, chat can become a practical way to keep AI agents visible without adding another dashboard.

Why Gavin Moved from OpenClaw to Hermes

One of the most useful parts of the conversation is Gavin explaining why his own workflow shifted from OpenClaw experimentation toward Hermes. OpenClaw impressed him early because it could figure things out quickly, but updates, security changes, and occasional breakage made it harder to keep stable. Hermes became attractive because it was built with containers in mind, making it easier to run separate agents in Docker on Unraid without one experiment disrupting the rest of the setup.

Backups, Guardrails, and Approval Before Action

Jim and Gavin also spend time on the practical safety side. Gavin shares the hard lesson that an agent can make a mess if it has too much freedom, including a Home Assistant dashboard problem that required restoring from backup. The takeaway is simple: let agents observe first, recommend second, and only act after they have earned trust and there is a clear rollback path.

Specialized Agents Instead of One Giant Assistant

The strongest idea in the episode may be agent specialization. Gavin describes building focused agents for specific jobs, including lawn care, pool maintenance, fitness accountability, finances, and Home Assistant. Giving each agent a narrower role keeps memory cleaner, expectations clearer, and the advice more useful than asking one large assistant to remember and manage everything.

Home-Lab Side Quests: Local Models, Drives, and Chia

The later part of the show widens into familiar home-lab territory: local LLM strategy, what older hardware can still do, hard-drive warranty frustrations, and Chia as a possible future automation target. Those side topics reinforce the larger theme of the episode: AI agents become useful when they are tied to real systems, real constraints, and real maintenance work.

Chapters

Guest

Gavin Campbell joins Jim for another practical home-lab and home-automation discussion. Gavin is part of the HomeTech.fm community and regularly brings hands-on experience with smart-home systems, networking, and emerging technology.

FAQ: OpenClaw, Hermes, and AI Agents for Home Assistant

What is this episode about?

HGG680 is about using AI agents with Home Assistant, with a comparison of OpenClaw and Hermes plus discussion of local models, cloud models, Mattermost, Docker, Unraid, and home-lab hardware.

What is the difference between OpenClaw and Hermes?

The episode frames OpenClaw and Hermes as different approaches to AI-assisted home automation. OpenClaw leans into extensible agent workflows, while Hermes is discussed as a practical chat-driven option for notifications and interaction.

Should Home Assistant users run AI locally or in the cloud?

It depends on the job. Local AI can help with privacy and control, while cloud AI may offer better performance, larger context windows, and easier access to newer models.

Why does token cost matter for home-lab agents?

Agents can use tokens quickly when they keep long histories, read tool output, or run repeated checks. Managing context is important for both cost and reliability.

Can Mattermost be used with Home Assistant AI agents?

Yes. The episode discusses Mattermost as a useful place for notifications, commands, logging, and collaboration around AI-assisted home-lab workflows.

Resources Mentioned

Join the Conversation

Are you experimenting with AI agents in Home Assistant, or are you keeping AI separate from your smart-home stack for now?

What would you trust a local AI agent to do in your home lab, and where would you draw the line?

Leave a comment and join the conversation.

Full show notes, transcriptions available on request, audio and video at https://theaverageguy.tv/hgg680

Join Jim Collison / @jcollison for show #680 of Home Gadget Geeks, brought to you by the Average Guy Network.

WANT TO SUBSCRIBE? http://theAverageGuy.tv/subscribe

Join us for the show live each Thursday at 8pmC/9E at http://theAverageGuy.tv/live