Show HN: Looplet – a 0-dep agent loop you own (github.com)

Looplet, a new framework for creating and controlling LLM (Large Language Model) agents, has been launched, offering developers unprecedented control over agent loops. Unlike traditional agent frameworks that act as black boxes, Looplet exposes each step of the agent's decision-making process through an iterator, allowing users to observe, modify, and direct the flow at every point without being locked into specific dependencies or complex abstractions. This means developers can seamlessly transition from a simple prototype to a production-ready agent within the same framework, utilizing ordinary Python code for maximum flexibility. The significance of Looplet lies in its "zero-dependencies" design, which not only simplifies deployment but also offers practical debugging and evaluation features that are typically cumbersome in other frameworks. Each action taken by the agent is encapsulated in a "Step" object that can be printed, saved, or used for regression testing, providing a clear audit trail of the agent's operations. With built-in hooks for inserting custom logic and managing context dynamically, Looplet is tailored for those seeking to build robust and easily modifiable LLM agents while retaining complete ownership of the underlying processes.