Field Notes · Opening

Topological Attribution Distance (TAD): Revealing Segment-Level RAG Influence on LLM Output Geometry for Incident Log Analysis

This paper introduces Topological Attribution Distance (TAD), a method for tracing how individual segments from retrieved documents in Retrieval-Augmented Generation (RAG) systems influence the geometric structure of an LLM's outputs in cybersecurity incident log analysis.

Llama-macOS – Agentic and MCP Native macOS Front End for Llama.cpp

Llama-macOS is a macOS application that provides a native front end for Llama.cpp with support for agentic features and the Model Context Protocol (MCP).

Patterns and problems in emerging multi-agent systems

This item appears to be a research paper or article examining design patterns and challenges that arise in systems where multiple independent agents interact.