# Smolagents Минималистичный фреймворк Hugging Face для ИИ-агентов с кодом и вызовом инструментов - Findings24: - Сайт: - Цены: Freemium - Категории: AI Ассистент Кода, AI Инструменты Разработчика ## Что такое Smolagents? Smolagents is a minimalist AI agent framework from the Hugging Face team. Its stated purpose is to help developers build agents that can interact with the real world using only a few lines of code, while keeping the framework small and approachable. The site centers the product on code agents: instead of producing only JSON or text actions, the agent writes and executes Python code snippets to complete tasks. Smolagents also supports traditional tool-calling agents, sandboxed execution options, and integration with models and tools from the Hugging Face ecosystem and beyond. ## Что умеет Smolagents? ### Compact core framework Smolagents is described as a minimalist framework with about 1,000 lines of core code in \`agents.py\`, which keeps the abstraction layer small and the codebase easy to inspect. ### Code agent workflow The framework emphasizes agents that write and run Python code snippets to carry out actions, rather than relying only on JSON or text-based tool outputs. ### Efficiency-oriented execution The site says code agents can reduce steps and LLM calls by about 30% compared with standard tool-calling methods, and can perform better on complex benchmarks. ### Sandboxed code execution Smolagents supports sandboxed environments such as E2B for safer execution of generated code, giving developers an isolated place to run agent actions. ### Model and Hub integration The framework integrates with Hugging Face Hub for sharing and loading tools, and it can work with Hugging Face models through Transformers as well as OpenAI, Anthropic, and others through LiteLLM. ### Multiple agent styles In addition to code agents, the site says smolagents also supports traditional tool-calling agents that use JSON or text blocks when that format is a better fit. ## Варианты использования ### Code-first agent development Build agents that can take actions through Python code, especially when the task benefits from reusable functions, object handling, or more complex control flow than simple tool calls. ### Custom tool workflows Add custom tools such as travel planning, web search, or image generation to an agent and share those tools through the Hugging Face Hub. ### Safer code execution Run generated agent code in a sandboxed environment when you want a safer boundary around execution during development or deployment. ### Multi-model setups Use supported model providers from Hugging Face, OpenAI, Anthropic, or other LiteLLM-compatible sources when you want flexibility in model choice. ### Tool-calling alternatives Choose traditional tool-calling agents instead of code agents when JSON or text-block actions fit a specific workflow better. ## Часто задаваемые вопросы ### What is Smolagents? Smolagents is presented as a minimalist AI agent framework from the Hugging Face team. It is designed to let developers define an agent, add tools, and run it with relatively little code. ### What kinds of agents does Smolagents support? The site says smolagents supports code agents, where the agent writes and executes Python code snippets, and also supports traditional tool-calling agents that use JSON or text blocks. ### Is Smolagents free? The pricing page shows a free Starter plan and a paid Pro plan. Starter includes 5 devices, 1 month of cloud retention, unlimited notifications, and basic integrations; Pro is listed at $12 per month with unlimited devices, 1 year of cloud retention, unlimited notifications, advanced integrations, and priority customer support. ### How does Smolagents handle tools and execution safely? The site describes sandboxed execution options such as E2B for safer code execution. It also shows examples built around custom tools and Hugging Face Hub sharing. ## Краткие сведения - **Category**: AI Agent Framework - **Brand**: Smolagents - **Source domain**: smolagents.org - **Primary users**: Developers building agents - **Core workflow**: LLM-written Python code and tool use - **Pricing**: Free Starter plan and paid Pro plan