Douwe Osinga
Douwe Osinga
Qwen3-14b is just not going to cut for general tool use. if you switch off all the tools and keep only the developer tools, maybe. but for me even asking...
you can't? it uses tools for me, just not very well; what happens if you run that script?
you can use custom providers for this I think. ping if not
you need to add a line like: GOOSE_PROVIDER: openai
yeah, export GOOSE_DISABLE_KEYRING=1 seems to do the trick. however we should default to that if writing to the keyring fails. it seems hard to keep this sort of thing working...
I think that is the right way to do this @taniandjerry - just need somebody to write the code
FWIW this does seem to work if you use Ollama directly rather than through LM Studio
changing this to a feature request - running LM Studio on an Ollama port just means you're missing the Ollama tool emulation for smaller models. good to try this config,...
> I didn't see this on the docs, but would be very useful for agents to be able to specify their model; RooCode does this and it's extremely handy. You...
hey @gugahoi - can you do the https://github.com/block/goose/pull/2717/checks?check_run_id=45941194620 so we can merge this? sorry about the slow reaction and overhead