hallow elk
Hi,
I’m using Unsloth Studio for domain-specific models (CAD/HVAC use cases).
Do you plan to support:
• tool calling / function calling
• agent-style workflows (multi-step reasoning + actions)
This would make Unsloth much more useful for real-world engineering applications.
prisma hinge
@hallow elk we're working on exposing an api endpoint from unsloth studio for GGUFs using llama-server, that way you can call the model from your own external workflow, claude code or even openclaw
hallow elk
Thanks but integrated workflow into studio for two things like router littlellm or python or some script language and for orchestration LangGraph becouse this is only way to get better resolution with small custom fine tuned llm on hardware with limited resources , problem is not so much tool cooling but use multiple fine tuned llm for specific task in orchestrated manner
prisma hinge
we removed litellm because of the hack, we had it in recipes
yes but we're not an agentic platform. that would be a completely different product imho
and if we ever do it, we're probably not gonna use langgraph..
hallow elk
I was thingking you version of simple littellm and simple langraph just basic one for testing models but for deployment no there are different product but for testing and proof of concept to have all in one IDE
prisma hinge
we already have tool calling , web search etc in chat
but you can't implement agent-style workflows reliably in a compacted way
imho
having said that
will think of a way, but definitely plugging in litellm or langgraph is not where we're heading
until then you can simply call the model and test it in an agentic framework via the openAPI compatible endpoint we're working on (should be out next week)