# AgentAnufry

An open-source Python constructor for a base autonomous AI agent, with long-term memory, browser automation, a task tracker and a skill system to build on.

AgentAnufry (the agent is named Anufry) is a simplified base construction for a multi-task AI agent in Python, aimed at engineers and hobbyists who want to experiment with building their own assistants rather than use a finished product. It ships with long-term memory, browser automation through CDP and Playwright, a task tracker, and a system of built-in skills.

The project is deliberately kept simple so its modules can be reworked in a short time: swapping the database or memory rules, rewriting built-in skills, adding new internet-facing tools, or changing how modular skills are processed.

It supports multiple LLM providers configured through environment variables, including OpenAI, a local Ollama or LM Studio model, Yandex Cloud, and OpenRouter. Not every provider supports embeddings; when the configured provider does not, semantic memory search is disabled automatically and the agent continues to run without it.

Website: https://github.com/Bednyakov/AgentAnufry