OpenCode - use pretty much any LLM to assist you with Homeassistant

The integrations that allow you to run Claude Code are nice.
I built an experimental solution that can use Mistral, which was fine as well.

But running OpenCode is a step forward, because this way you can switch to pretty much any LLM you like - without having to worry whether there is an app that you can use to help you with Homeassistant.

It got co-authored by Claude Code and Deepseek and is hopefully in a usable state already.

If not: drop me a bug report and it will get fixed if time permits.

Oh, and it lives at Codeberg. So, you have to add the repo manually, because (apparently) there is no way to add a repository via Badge-link.

Installation might need a bit. OpenCode is compiled from source and not pulled via npm (for now).

Link: https://codeberg.org/umrath/hass-opencode

Hi @umrath, sounds great - are you aware of: HA OpenCode - An addon to plug opencode AI into your Home Assistant Are there any differences between the solutions?

I found this afterwards and haven't checked yet. But will do.
Seems the other project is largely abandoned.

I would like to try yours Thoralf. I see it is in such active development.

My goals would be to ask Claude to look through my configuration for errors and suggest improvements. Right now I am using Clause standalone itself and just pasting in and it has been invaluable.

There will be a major change released soon. But this is pretty much experimental.
If you want to stay safe - use 1.2.2 - which is fairly stable.

Edit: 2.1.2 is now stable as well. Still need to compile opencode from source, though. That means installing/updating takes a little bit longer.

This is pretty much one of my major use cases as well.
The spam in the log files is too big to be able to handle this manually and therefore I use a this tool to handle it and find the really relevant stuff.

Works quite good so far.

Major architectural change in progress right now. Most is done and working already - but more to come.
If it fails for you: jump back to stable 1.2.2 or wait a bit until it's fixed.

Edit: v3 should be stable now

What is your definition of "largely abandoned"? :wink:

Oh, sorry. Seems I misinterpreted "I will use this myself, but recommend everyone else to stay away!"
I apologize!

No need to apologize! :slight_smile:
I have that sentence there just to make sure anyone that wants to test has some warning about the potential dangers of using a tool like this. Why not contribute to the tool already created? There is already a healthy number of users, and pooling resources into one project instead of splitting across multiple 100% overlappong tools seem like a better idea?

Ok, fair.
Only problem: The code (as clearly visible in the repo) is pretty much 100% written by Deepseek and Claude (I use both in tandem).
I'm not sure whether you want to deal with that.

But if that's not a showstopper for you: great idea!

Yeah, in principle I have no problem with that, as I myself am doing much the same! :wink:
I just noticed you picked up a lot of the things I have created, so instead of duplicating it would maybe make sense to focus on one tool.

Looking forward to what you come up with! :slight_smile:

I will have a deeper look into your project and start to use it in daily business to see what I actually miss.
In parallel I will let the minions to a gap analyzes to see what I could get ported to your repo.