smolagents
Hugging FaceMinimal code-first agent framework from Hugging Face, whose CodeAgent writes actions as executable Python rather than JSON tool calls. The core is about a thousand lines. It runs actions in a sandbox, is model-agnostic across hosted and local providers, and imports tools over MCP and from LangChain.
Verified 2026-08-13 via the huggingface/smolagents repository.
Openness
5 high confidence- license
- Apache-2.0(OSI)
- source
- public(github.com/huggingface/smolagents)
- core-gated
- ungated
Fully OSI-licensed (Apache-2.0), entire library public, no proprietary/managed core, top openness.
- https://github.com/huggingface/smolagents recorded 2026-08-13
Apache-2.0 license, full public source for the code-agent framework, 28.8k stars and 2.9k forks; no enterprise directory and no paid tier
Adoption
3 high confidence653,459 PyPI downloads in the trailing 30 days for smolagents, from the pypistats API, which bands at level 3 (100K-1M) on the software adoption scale - meaningful, but below the heavyweight framework anchors. smolagents is the declared and primary channel, so this measures the product itself rather than standing in for it.
- https://pypistats.org/api/packages/smolagents/recent recorded 2026-08-13
653,459 downloads in the trailing 30 days for smolagents
- https://pypistats.org/packages/smolagents recorded 2026-08-13
the rendered pypistats page for the same package, kept as the original citation
- https://github.com/huggingface/smolagents recorded 2026-08-13
28.8k stars, 2.9k forks (corroborating, not the basis)
Capability
3 medium confidenceDeliberately lightweight: strong on model-agnosticism, tool-use, and sandboxing but thinner orchestration/eval surface than heavier frameworks; no benchmark, scored 3 on feature matrix.
- https://github.com/huggingface/smolagents recorded 2026-08-13
code-action agents, sandboxed execution, MCP and LangChain tool import, multi-model support; no benchmark figure published
Verified 2026-08-13