Show HN: Coding agent with algebraic memory (VSA) instead of RAG (github.com)

A new coding agent named Raidho has been announced, which utilizes a dual-model architecture to enhance coding efficiency and memory retention. Unlike traditional models that operate in a single loop, Raidho employs a high-performance model for reasoning and a cost-effective one for execution, complemented by a durable memory system that persists across sessions. This innovative approach allows developers to choose their preferred providers for each task, offering flexibility and enhancing cost efficiency—demonstrated by a significant reduction in operational costs compared to conventional methods. The significance of Raidho lies in its unique implementation of Vector Symbolic Architecture (VSA) instead of the common Retrieval-Augmented Generation (RAG), enabling it to store and recall facts with reduced memory footprint. Additionally, the agent incorporates features such as auto-distillation for efficiency improvement over repeated tasks and a council mode that facilitates consensus building between different AI providers. With a straightforward setup process and the capability to integrate with various AI service providers, Raidho represents a promising advancement in coding agents for the AI/ML community, highlighting the potential for greater flexibility, lower costs, and enhanced performance in coding tasks.