I host my email, just not at home but on a rented server.
Here I documented it all https://wiki.gardiol.org/4-e-mail-hosting
It works, its been working for decades now. Never had significant issues…
Email is quite resilient. Servers keep trying to send again and again messages for hours before they give up… So a backup server is not even really required.
Well my small private instance is there!
I selfhost my.chatbox (and agent) on my hardware at home.
Open models are surprisingly good and any thing you tell them stays on your premises, so no privacy risks.
You need some hardware and performance are not as chatgpt or Claude.
Yeah, it’s good, but it’s not a replacement for your own intelligence. Treat them as a free assistant which seems competent but probably says shit to please you…
Always double check and apply your own intelligence.
I find this comparison bad for two reasons.
First, BI is alive, AI is not alive.
Second, a leech has no positive reason to exist and do no good to anyone, probably the only exception is some bird feeding on them, which could feed on other critters. AI is only a tool, and as such can have good use when used properly and not abused.
I tried stalwart. Could not run reliably on the server I rent. Too slow.
Instead postfix + dovecot just fly and barely used any resources.
Maybe tools like stalwart have use for bigger email servers, but for my use case it’s just massive overkill. Easy to setup tough.
Yes you do.
First of all, there are services that will refuse to work properly on http only. Specially if you want an SSO like authelia, will not work over http
Second, modern browsers will complain when using plain http if the website has forms to fill or passwords to write … very annoying, but it’s the way it is.
Third, having the “this site is not secure” broken lock is not very nice for users. At best they might complain, at worse they will get used to “unsecured” websites and .ight end up easier prey for fake sites in the future. Yes I mean wife or kids. These are things normal users should not see as normal…
Last, if you expose anything to the web, please never ever do that without proper https with certificates from let’s encrypt.
Why 1.0.0? 0.10.0 feels more… Binary
Good idea… Will try that next time
True, but currently my prefill speed sucks: some 300-400 t/s. I guess going 8x 8x on the cpu lane will improve that, hopefully by quite some… Losing it again in favor of 24gb extra vram?
That’s the question …
Guess need to wait for the raiser board and test actual speed improvement for prefill.
True… The advantage in vram probably outweighs the rest …
I guess a test is the only answer.
Need to wait for the raiser board to arrive and test …
Yes, carefully check AI output is a must, always.
But still tests even if not perfect are better than no tests, so…
I have relaxed the rate limiter on the proxy, try now
It seems it’s being overwhelmed somehow… Refresh does help.
Yes, my use case seems to be accepted indeed.
Can you elaborate on using obsidian as a memory store?
You are welcome, Indeed my original wall of text was typical of AI bots…
Well, rules are important and knowing them is too. Ignoring them can then be a choice, otherwise it’s just ignorance…
But yes I fully understand your point.
Ram is overkill. 32gb maybe even 16gb would be enough. The CPU too, I used to have an Intel i7 from 10 years ago and it worked. Upgraded because that was the cheapest upgrade I could do.
The GPUs are mandatory, at least to get 30t/s…
Thank you! These days it’s good not to master English enough to be mistaken for an ai bot
My mother language is Italian, not English. But I spent some time living in the US so maybe there is that at my advantage.
Cool!
So it was 1.0.0 after 0.9.0 after all…