Sukhdeep Brar (Mar 24 2025 at 02:35): Sukhdeep Brar (Mar 24 2025 at 02:35):Has anyone given Goose a go? I've been using some Agents for work, and it's a game changer (for me). It's not faster, but, it can work while I'm in a meeting or doing something else.
Ron Waldon-Howe (Mar 24 2025 at 04:05):I installed goose, with ollama and qwen2.5
But I can't get it to modify any files
I'm not sure what I'm missing
Ron Waldon-Howe (Mar 24 2025 at 04:06):And it seems like it isn't reading files either
It only ever spits out generic suggestions in the chat expecting me to do the work
Ron Waldon-Howe (Mar 24 2025 at 04:09):I did configure it to prompt me before changing files
But it never actually does this
Maybe I need to configure it to do whatever it likes without any warning? (I really do not want this)
Dustin (Mar 24 2025 at 07:45):I’m trying to figure out…what does it do?
Ron Waldon-Howe (Mar 24 2025 at 08:52):Dustin said:
I’m trying to figure out…what does it do?
Ron Waldon-Howe (Mar 24 2025 at 08:53):But I can't get very far with that demo, because it doesn't edit files or run commands on my system, unlike in the video
Sukhdeep Brar (Mar 24 2025 at 10:28):Which qwen model are you using? I had luck with the 14b. Every smaller model I tried didn't end up doing any tool calls.
Goose is pretty explicit when it _wants_ to do a tool call, but doesn't because you configured it not to. You either get a prompt saying what it's going to do with an OK button or it says what it wants to do but says it won't since you're in chat only mode
Sukhdeep Brar (Mar 24 2025 at 10:29):Goose is an OSS option for doing tasks similar to Claude Computer Use.
Ron Waldon-Howe (Mar 24 2025 at 21:56):I'm using the 7b
Thanks for the hint, I'll give a larger model a try
I think I have been sticking with models < 8GiB size to try to cater to the different GPUs I have on different machines, with 8GiB VRAM being the lowest common denominator
Ron Waldon-Howe (Mar 24 2025 at 21:58):The model license also matters
Ron Waldon-Howe (Mar 24 2025 at 22:24):I need to double check the Gemma 3 licence, because that model is optimised for single GPUs which could be an improvement
Ron Waldon-Howe (Mar 26 2025 at 09:55):Ah, so Gemma 3 doesn't support tool use, so that's out
I'm going to try phi4-mini:3b and qwen2.5:14b and maybe a few others and see how it goes
Notification Bot (Mar 26 2025 at 09:55):A message was moved here from #general > general chat by Ron Waldon-Howe.
Ron Waldon-Howe (Mar 26 2025 at 09:55):There's a newer demonstration video: https://www.youtube.com/watch?v=cLPKU53vlTI
Ron Waldon-Howe (Mar 26 2025 at 09:57):I want to use the same model for both work and personal usage, which rules out Llama due to its usage restrictions
Siddhartha Golu (Mar 26 2025 at 10:17):I'm curious to know: for people using this, what's the typical use case?
I've been using Aider on and off for some time, and generally like the paradigm of user being ultimately in control of steering the output turn by turn. Are the current agents capable enough to do non-trivial tasks fully autonomously?
Last updated: Apr 03 2025 at 23:38 UTC