sure, the agent will not do any operation without your approval. The agent first show you what they plan to do in terms of automation/dashboard and only if you approve it it will deploy it
I missed some flexibility, I use the dirt cheap GLM Coding Plan with GLM-4.6 for 6 dollars a month, anyway long story short, you can find and install it here, it works as described but only if you have a GLM Coding Plan
This plugin is dangerous (which, like, I totally could have guessed). I asked it to make a dashboard for me and it completely rewrote my config.yaml erasing everything. Excellent example of the unexpected side effects of an AI Agent. Recovered from backup. Terrible idea.
Against my better judgement I may continue to experiment with it; I came here looking for a way to give an AI the contents of my home to better make dashboards, maybe I’ll figure out a good way to do that with this.
Yes and quite honestly it’s not been stressed enough. If you’re new to LLMs and see openclaw it’s shiny… Yes great capabilities…
But we need to be CRYSTAL CLEAR - ANY llm with direct access to your config directory is dangerous. This… This is just like if I had decided to jump on my Harley barefoot with shorts, no helmet and decided to run high speed in through a busy construction zone laying down a gravel base…
Yeah I can probably do it but there’s also an extremely high chance 'll fall get crunched burned on the pipes or wipe the heck out with zero safety gear.
Many. Many Many people do not seem to realize this is possible with ANY LLM given direct write access to your HA install.
You’re risking the entire setup. Yeah it’s that bad.
Just be glad it didn’t export your secrets file… (hint yes it ab-so-freaking-lutely can)
THIS ^^^ is the exact reason why I do not currently endorse ANY llm that has direct access to write to your config directory. If you know what you’re doing great… do put on the damn helmet. Also glad you’re back up.