LIVE
Google Introduces "Gemini 4 Carbon" as Next Step in Its Model Lineup28/10/26 · Google|DeepMind unveils Genie 4, a new generation of world model27/10/26 · Google DeepMind|Anthropic files IPO prospectus, with risk disclosures taking center stage26/10/26 · Anthropic|Robot policy trained only in simulation matches top real-world systems26/10/26|OpenAI unveils GPT-6.1 Terra, a model built for computer use26/10/26 · OpenAI|Google's AI gives evasive answers about who won the 2020 US election28/10/26 · Google|NVIDIA introduces Nemotron 3.5 Transcribe, a promptable multilingual transcription model28/10/26 · NVIDIA|OpenAI and ServiceNow deepen partnership on enterprise workflows28/10/26 · OpenAI|Failed startups' data finds a second life training AI models28/10/26|Anthropic expands AI for Science program to 20 foundations and research institutions28/10/26 · Anthropic|Anthropic shares early results on using Claude to accelerate drug development27/10/26 · Anthropic|GEPA-Curator: Filtering Failures to Better Train LLM Agents27/10/26|Google Introduces "Gemini 4 Carbon" as Next Step in Its Model Lineup28/10/26 · Google|DeepMind unveils Genie 4, a new generation of world model27/10/26 · Google DeepMind|Anthropic files IPO prospectus, with risk disclosures taking center stage26/10/26 · Anthropic|Robot policy trained only in simulation matches top real-world systems26/10/26|OpenAI unveils GPT-6.1 Terra, a model built for computer use26/10/26 · OpenAI|Google's AI gives evasive answers about who won the 2020 US election28/10/26 · Google|NVIDIA introduces Nemotron 3.5 Transcribe, a promptable multilingual transcription model28/10/26 · NVIDIA|OpenAI and ServiceNow deepen partnership on enterprise workflows28/10/26 · OpenAI|Failed startups' data finds a second life training AI models28/10/26|Anthropic expands AI for Science program to 20 foundations and research institutions28/10/26 · Anthropic|Anthropic shares early results on using Claude to accelerate drug development27/10/26 · Anthropic|GEPA-Curator: Filtering Failures to Better Train LLM Agents27/10/26|
Toolofficial publisher

Tools smolagents (Hugging Face)

Tool : A precise action the AI can call · requires huggingface

Tools declared with a simple decorator, code execution by the agent.

Minimalist and elegant; the agent writes Python, so sandboxing is required.

Docs
Overall score
6.5 /10
Security: one caveat, see the card
Utility
7.0
Reliability
6.0
Security
One caveat
Ease
5 min · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Building an AI agent capable of using tools usually requires a lot of infrastructure code. smolagents reduces this complexity to a few dozen lines by providing a minimalist Python library, fully open source under the Apache license.

The core principle relies on the CodeAgent: instead of calling tools via JSON, the agent directly writes Python code to chain actions, nest functions, use loops or conditionals. Tools are declared with a simple Python decorator, making them immediately available to the agent. A second agent type, ToolCallingAgent, remains available for scenarios where JSON-based tool calling is preferred.

The library targets Python developers who want to prototype or deploy agents without depending on a specific model provider: it supports Hugging Face models, OpenAI, Anthropic, LiteLLM, local Transformers and Ollama. Tools can come from an MCP server, LangChain, or a Hugging Face Space.

However, smolagents is not suited for people without Python experience: there is no consumer-facing graphical interface, and managing security when executing agent-generated code remains the developer's responsibility, requiring a sandbox environment (Modal, E2B, Docker, Blaxel) for production use.

What it offers

  • ·Tool declaration via a simple Python decorator, with no additional configuration.
  • ·CodeAgent: the agent writes and executes Python code to chain actions and tools.
  • ·ToolCallingAgent: support for classic JSON or text-based tool calling.
  • ·Compatibility with many model providers: Hugging Face, OpenAI, Anthropic, LiteLLM, local Transformers, Ollama.
  • ·Integration of tools from MCP servers, LangChain, or Hugging Face Spaces.
  • ·Share and load agents and tools from the Hugging Face Hub as Gradio Spaces.
  • ·Support for multimodal inputs: text, image, video, and audio.
  • ·Command-line tools (smolagent, webagent) to run an agent without writing boilerplate code.

In practice

Price
Free
Commercial use
Commercial use: yes
Available on
sdk-python
Host
huggingface
Licence
Apache-2.0
FR interface
No

Security and data

Rating
Security: one caveat, see the card
Capabilities
readswritesexecutesnetwork
Your data
Data stays local
Maintenance
Active

Links and repository checked on 25 October 2026 · added on 1 December 2024

How to install Tools smolagents (Hugging Face)

1. Make sure Python is installed on your machine. 2. Install smolagents with default tools: pip install 'smolagents[toolkit]'. 3. For OpenAI or Anthropic, add the litellm option: pip install 'smolagents[litellm]'. 4. For local models, add the transformers option: pip install 'smolagents[transformers]'. 5. In production, configure a sandbox environment (Modal, E2B, Docker, or Blaxel) before letting the agent execute code. 6. Import CodeAgent and a model in your Python script to run your first agent.
pip install 'smolagents[toolkit]'

Official documentation ↗

In the same category

Whole category Orchestrate →