what is agentmaker fr?

agentmaker is a Python framework for building AI agents, systems where a language model can call tools, remember past conversations, and search through documents, with the pitch that four lines of code get you an agent that already has memory. It works with many AI model providers, including OpenAI, Anthropic, Google Gemini, DeepSeek, Moonshot, Zhipu, local models run through Ollama or vLLM, and any service that copies the OpenAI style API. The framework covers a wide set of agent building blocks. Any typed Python function can become a tool an agent can call just by adding a one line decorator, and it can also connect to external tool servers using the Model Context Protocol. For remembering information, it includes local search that combines meaning based and keyword based matching without needing an external database, a full pipeline for retrieval augmented generation with source citations, and long term memory that pulls out and updates facts over time. Everything defaults to a local SQLite database so there is nothing extra to set up, though each storage piece can be swapped out for something else later. It also includes safety and reliability features: guardrails that check input and output, an approval step before risky tools run, and the ability to record every step of a run to a file or database. A built in debugging tool, called Trace Detective, can look at a failed run and use an AI model to point out the first thing that went wrong, the likely cause, and a suggested fix, either printed to the terminal or shown in a small local web page. Installing it requires Python 3.12 or newer, done with a simple pip install command, with optional add ons for things like web search tools, document loading for retrieval, and OpenTelemetry tracing. The project includes sixteen runnable example scripts and is licensed under MIT.