And wrote an article about it. Let me know if you think it is a good or bad idea to run HA in production for a business.
Any other people running a museum or similar public venue on HA?
Edit: updated article URL
And wrote an article about it. Let me know if you think it is a good or bad idea to run HA in production for a business.
Any other people running a museum or similar public venue on HA?
Edit: updated article URL
Interested to read the article, but the link you posted doesn't work for me.
I think it's a great idea, with a couple of caveats.
First and foremost: Who is going to maintain it, long-term? Even if the business needs aren't very dynamic, the software is. This is the biggest problem I've seen with any software - for that matter, any technology - in any business. Buying and installing is only the beginning. Ongoing maintenance, changes and repairs are important, too.
Second: Is it the right tool for the job? Assuming you're doing facility management, HA probably is. But HA can do many different things. And for many of them, there are better options out there. Always look at the requirements. What problem are you trying to solve? Pick the best solution for that specific problem.
Your link is pointing to localhost and does naturally not work outside your own computer.
I'm curious how much time is spent in the gardens, and how much tinkering with HomeAssistant? I'm assuming automating is the tradedoff and the two counterbalance the time available with job interest and satisfaction. Is your time voluntary and is the Botanical Gardens management aware and have robust succession and contingent management strategies should your support abruptly cease? Is it a one person labor of love or part of a broader team approach, well documented and supported by multiple people?
Wow, I pointed it to my dev instance, how dumb. I updated the url, thanks for pointing out.
I address both in the article. Regarding long term maintenance: There's a heavy bet on LLM's (in this case Claude). There is an example of how that is working in the article. They also use HAShop for consultancy when something really breaks and can't figure it out. The hardware and networking layer is bit harder, but there is an external IT company that can help out.
Is it the right tool for the job? I think currently it is, because I know of no tool that can integrate all our systems in a single pane of glass. And for that, HA is amazing.
Thanks for the update. I read the succinct article - well presented.
I note with unease that you've swapped dependency of one individual and a small number of vendors for dependency on AI, and the recent abrupt withdrawal of Anthropic products (by US authorities you have no legislative or financial clout over) that you may increasingly rely on is cause for concern, as is the growing trend to start to embed AI dependency and the associated token consumption you have to pay for as an ongoing cost overhead, payable forever, far beyond when your contracts with the vendors expire. You may now have access to your data, but the dependency on how to robustly use it is still out of your grasp and control - you have swapped one devil for another.
Hardware has not become generic enough for robust AI support yet, as evidenced in the flood of threads in these support forms marking new releases, and the AI bots are still working out how to update their knowledge bases to discard obsolete information from the decision strategy trees garnered here.
The reliance on AI (not to be confused with LLMs) will result in deskilling, and your reliance on those external vendors like HAShop will become your other critical path bottleneck in times of contingency. Who will have the skills to ask the right questions?
Sure, use AI for advice, but don't rely on it yet - those rare plants are too valuable to leave to some hallucinating slop generated by a rogue LLM machine trained on bad code examples in user forums, reddit, and vendor sponsored influencers keen to make a livelihood, regardless of truth.
HomeAssistant is your tool as there is nothing quite like it out there in the commercial world. The fact it is open source, often supported by unpaid volunteers of varying skill levels, often vibe coded addons, leaves a big market gap that the commercial world will recognise as a great need.
Witness the rise of the Matter standard, a meeting of vested interests, where the consortium struggled for years to arrive at a common platform, with the participants already adding extensions to make their ecosystem exclusive (looking at you Apple), and the struggle by everyday users to make it just work by plugging it in, as promised, years later. Plug-and-play is actually plug-and-pray. Even AI struggles with this.
May your plants thrive and your monthly token bill be minimal is my wish to you.
Thank you for your thoughts and I mostly agree. But there are some nuances:
Home Assistant is not responsible for process control, it is merely a UI on top of existing systems. This still means it's a dependency, as people get comfortable in using it for telemetry, but it is not a point of failure for keeping our collection alive, this responsibility still lies with the vendors.
Although Claude is a dependency, the switching cost to another AI platform are pretty small. Creating a memory export and integrating to another AI platform using MCP is trivial.
Token cost is minimal: Claude has a non-profit discount, which makes this very affordable for our small team. I know prices can shoot up, but on the other hand, current models have proven to be good enough for the task, so I see no reason to keep doing this on frontier models in say, two years. It could even be taken in-house on a local model at some point in the future. What feels like rocket science today is easy in a few years.
I know AI is far from perfect and Home Assistant is not there in terms of UX for the average user, but this whole project is a bet on the future based on the trajectory of both dependencies. A somewhat risky bet, but thusfar with real measurable impact on energy usage and employee experience.
You bring up some good points. But to that last point, from what I can tell the new twist on that old saying is "plug and pay," not "plug and pray." Obviously that's what every vendor is working toward, and I don't see the purveyors of Matter hardware going any other way.
My rosary beads are well worn.
This is spectacular. Well thought out and I like how you're using AI to democratize telemetry analysis across a broader skillset.
Appreciate the post. I think it surprises people to see low risk high reward successful deployments of AI and home assistant.
I agree to be on the cutting edge of this you have to outsource a bit of control to say anthropic or whomever, but you're absolutely right that the cost is plummeting and in 5 years you may be running your own local AI. Don't let fear of the future keep you from building something awesome today.
My generation fretted that when "new developers" stoped learning assembler and could no longer do mental math in hex. Ha.
That two leaf plant is incredible. That's redwood level of survivability. I want to visit.
Thank you! There even is a Giant Redwood in the garden. It's funny to see how many people literally hug that tree on a daily basis.
Question,
How have you secured the ha installation against the big bad internet?
Since with all respect, ha is not really... secure enough to run a business imho
I disagree. HA is audited and fixed regularly:
Cloudflare tunnel and strong passwords is enough imo. And 3-2-1 backups.
Amazing.
I’d like to see this home assistant in action! When is an open day?
For monitorring dedicated monitoring tools like zabbix / nagios / icinga / checkml also come to mind worth investigating. There are dozens of plugins for such open source tools. And these tools are also build to scale and stay robust.
Basically any (open source) infrastructure monitoring tool that your (II) organization already uses, could be made adapted to also monitor the systems in the building.
Probably such tools that are designed for these tasks might run a bit more robust on the long run if you can integrate into the organisation wide monitoring tooling.
Your department technical people would then need to get their own dashboard on the central monitoring tool to see, (and manage and be able to script more integrated alerts for devices depending on eachother on the systems) you are responsible for. Alerting for critical stuff can than probably be easily integrated into the probably already existing 24/7 duty.
And home assistant probably can also integrate with zabbix or something to still have a more human friendly view for the less technical staff that still need to have a view on things.