Agent

• Tool calling in a loop

https://langchain-ai.github.io/langgraph/tutorials/workflows/

Small models for function calling

• What small / local models can perform tool calling well?

https://www.reddit.com/r/LocalLLaMA/comments/1iozmns/best_small_model_for_function_calling/

https://gorilla.cs.berkeley.edu/leaderboard.html

• #30: Qwen2.5-14B-Instruct (Prompt)
• #52: Qwen2.5-7B-Instruct (Prompt)

Ollama to access local models

https://ollama.com/library/qwen2.5

ollama pull qwen2.5:14b

from langchain_ollama import ChatOllama

# qwen2.5 - 14b (#30 on BFCL leaderboard)
local_llm = "qwen2.5:14b"
model = ChatOllama(model=local_llm, temperature=0.0)

Multi-agent architectures

• Agents SDK: https://platform.openai.com/docs/guides/agents-sdk
• Separation of concerns