HN Explorer

Who is using Ollama day-to-day?

3 points by brauhaus ↗ HN
I maintain a Chrome extension that adds an AI support to any input field online.

One of my favorite reviews is a 2-star review asking for “local model support”.

My first reaction was: who installs a 100KB Chrome extension to talk to a 10GB model running locally?

But it did make me curious. Are people actually running Ollama or LM Studio as part of their daily workflow?

10 comments

[ 4.4 ms ] story [ 38.3 ms ] thread
raxxorraxor
Yes, I do. It is quite helpful. For main coding tasks I use Claude, but if my credits vanish or I have enough time to get an answer, I use Ollama extensively. I would recommend for developers to maintain their own AI pipeline as well.

For anything dealing with personal data, like browser inputs, I would exclusively use local models too. Probably still niche, but non-local AI would be a deal breaker for me for both browsers and OS.

brauhaus
I'm curious what the workflow actually looks like for people running Ollama day-to-day.

Do you mostly use it through the terminal, a UI like Open WebUI, or via integrations with other tools?

I’m trying to understand where a browser integration would actually fit - if at all

raxxorraxor
I use Cursor for Claude mostly, I just like it better than the claude console version for some reason. For Ollama I switch to VSCode and use roo code/cline mostly.

I have some a Gitea/Playwright mcp/webservice setup, which works well with both. But for that latter, I tend to restrict it to local models.

manishrana
I am running but it is only useful for very easy conference tasks, or either it needs a very high computing power. Currently I was running on a 32GB Mac Studio M1, and I mostly use it for generating commit messages.
drsalt
if you don't care don't waste your time. that user should make their own 100kb extension.
brauhaus
I went ahead and did it... XP

At the start, I was curious to see if I could. It ended up being a rabbit hole, of course.

I can see some use cases for it. The plan is to dogfood it and validate them.

a_lifters_life
Im building a script for my own dev to spin up droplets on DigitalOcean to run high performing LLMs like H100 - this is all prompted through open web ui (ollama) . I find the models my 12GB GPU will tolerate have too many hallucinations - its ok for diffing for example or basic scripts, but anything somewhat complicated I need 60GB+ GPU which isnt feasible for me to buy..