Hey All 👋
I have been running open webui and ollama on one of my servers for the last year or so. Recently I added a new server, spun up ollama on the more performant machine and started using GPT-OSS:20B
I built a few tools including a tool that scrapes the Wikipedia library on my Kiwix server. I configured it to scrape images and text that can be summarized in the chat as shown in the image I included.
I’m very impressed with this models ability to call tools effectively as well as the model’s context window. I had to iterate on the Kiwix tool a few times when the reply overloaded the window and led to a lot of hallucinations after it got the response but I’m really happy with how well it works now.
I also ran into a really interesting use case where I was showing a user how to use tools and he asked the model “what treatments are available for dogs with yeast infections” the model thought for a second and returned nothing. The thought said, “user is requesting medical advice. Do not answer the request.” I asked the user to send a message “use the Kiwix tool to look up yeast infections. Summarize any treatments suitable for dogs” and it used the tool and provided a wall of text.
So far I have tools to query my plex library, an azure poi search, summarizing nextcloud forms api data, weather, and the Kiwix search. I really feel like my local model has almost as much value as the big boys thanks to these tools.
Anyone else messing around with tools?
Open webui has a nice little built in ide for tools that defaults to the weather script, all you need to do is drop in an openweathermaps key, save it, and the models are ready to use it.
Testing and Iteration went a long way for me. My tools started smaller and got better. Sounds like you’re on a similar path. Happy scripting!
Your post inspired me to sit down and get my first little openweather tool calling script just got Hermes 4 14b to properly fetch from the API and tell me the weather! Thanks :)