Post #2961641

As a thought experiment I’ve been building, with the assistance of a coding agent, a “better” version of #openclaw that has a pluggable architecture. Like openclaw it’s coded in typescript. Unlike openclaw I’ve wanted to use typescript features like strong typings, and class inheritance, to try and modularise everything and be opinionated about architectural design and unopinionated about containerisation, implementation and security choices. I’ve been building an approach to orchestrating the agent behaviour over the past two weekends. It includes a PromptTemplate class, a Model class, a Tools class. I was today days old when I worked out all I’m doing is building another version of #langchain. A library that has existed since 2022, and has a typescript implementation. All part of the agentic / LLM / AI learning curve…