agentmaker

agentmaker is an open-source Python framework designed for building agents equipped with memory and advanced retrieval capabilities. The framework allows users to create an agent with memory in just four lines of code, streamlining the process from installation to deployment. It is suitable for developers seeking to implement agents that can utilize memory, retrieval-augmented generation (RAG), and governance features within a single package.

The core package includes six components, all accessible through a single import, and does not require signing up for managed services or installing additional packages for tracing. agentmaker supports four paradigms—Chat, ReAct, Plan-and-Solve, and Reflection—enabling users to build agents that handle a range of tasks and reasoning strategies. The framework also facilitates the composition of agents into multi-agent systems, expanding its utility for more complex scenarios.

Hybrid retrieval is built into agentmaker, combining vector and keyword search with reciprocal rank fusion and optional reranking. This unified retrieval core supports memory, RAG, and tool search, allowing agents to access and utilize information efficiently. The framework is designed with swappable backends, supporting every major provider as well as local models, giving developers flexibility in choosing their preferred infrastructure.

agentmaker is distributed under the MIT license and can be installed via pip. Its open-source nature and the absence of required managed services make it accessible for a wide range of users, particularly developers interested in experimenting with or deploying autonomous agents in Python environments.