Nagent – Not an Agent (github.com)

Nagent, a new open-source project dubbed "not an agent," offers a straightforward and readable implementation of agent-like behavior without the complexity commonly found in agent frameworks. The core of nagent revolves around a simple text file, a large language model (LLM), and a series of structured tags to facilitate interaction and execution. The design emphasizes clarity, allowing developers to easily modify or extend its functions while using a plain text file for memory storage. This distinctive approach showcases how many perceived "agent behaviors" can be distilled into simple file operations, interaction with an LLM, and action execution. Significantly, nagent aims to demystify the agent paradigm for developers, illustrating that effective agent functionality can be derived from a minimalist loop that appends user prompts to a conversation file, processes responses from the LLM, and executes corresponding actions based on structured tags. The use of tags for command execution—such as requesting shell commands or file interactions—simplifies the flow and allows for a clear implementation path. By encouraging a "build your own" mentality, nagent supports experimentation and customization, fostering a deeper understanding of the underlying mechanics of AI agents while promoting a low-barrier entry point for developers looking to explore agent-like systems.