G2

Khalil G.

Technical Communications Contractor

Small-Business (50 or fewer emp.)

9/17/2026

"Minimal Overhead, Maximum Flexibility for Python Agents"

4.5/5

What do you like best about Hugging Face smolagents?

What impressed me most about smolagents was how little framework overhead there is. Defining a tool can be as simple as decorating a Python function and writing a useful docstring, while CodeAgent gives the model much more freedom to work with actual Python instead of getting stuck in repeated structured tool calls. I also liked that I could change the underlying model without having to redesign the agent, which made experimenting with different providers much easier. Review collected by and hosted on G2.com.

What do you dislike about Hugging Face smolagents?

The biggest issue for me is the lack of control and visibility around agent execution. It’s difficult to tell how many tokens a run is consuming, and there aren’t built-in limits or warnings to prevent an unexpectedly expensive run. I also ran into a broken built-in tool because of an API change, which the agent managed to work around but shouldn’t have had to. Multi-agent workflows were another concern: one test ran for more than 15 minutes without producing anything, with no straightforward safeguards to stop it. For production use, I’d want much stronger monitoring and execution limits out of the box. Review collected by and hosted on G2.com.

What problems is Hugging Face smolagents solving and how is that benefiting you?

smolagents makes it easier to experiment with different LLM providers without rebuilding the agent around each one. When our HF Inference API credits ran out, we connected an Unsloth GGUF endpoint instead with a minimal change and kept the same agent logic. That flexibility is useful when testing different hosted or self-hosted models. Review collected by and hosted on G2.com.