Show HN: PMB – local memory for coding agents that shows if it is used (pmbai.dev)

PMB, a new local memory tool for coding agents, has been introduced, enhancing how these agents manage and utilize context during coding tasks. Unlike traditional approaches, PMB automatically injects relevant memory before the model makes any decisions, eliminating the need for prompts or tools that require remembering external commands. The system operates with sub-millisecond retrieval speeds and employs a hybrid recall mechanism that combines BM25 algorithms, dense vector searches, and entity graphs, all optimized to deliver ranked results efficiently. This architecture allows PMB to maintain high performance without hindering the agent's responsiveness. Significantly, PMB empowers developers to work without recalibrating their context during transitions between coding sessions or different tools, as the agent retains a continuous memory across various platforms. Each decision, lesson, or event in PMB is tracked, scored, and visualized in an intuitive, color-coded graph, enabling users to prune non-essential information effectively. With its robust local architecture, PMB empowers users with complete ownership of their memory files, ensuring that sensitive code and data remain on local machines. This innovation could lead to more productive coding experiences, as agents will no longer require repetitive context inputs, ultimately streamlining workflows in AI/ML development.