October 2025 Summaries

Engineers working with large language models (LLMs) often find themselves constrained by the limitations of prompt improvements, prompting some to explore the innovative concept of memory servers. A memory server acts like an external hard drive for an LLM, storing and organizing knowledge in a way that enhances long-term data retention and context comprehension. The author shares their experience of setting up a memory server using the Claude LLM and a Neo4j knowledge graph, which allowed the system to build a social graph of colleagues and retrieve detailed information about people and projects efficiently. The implementation improved accuracy and reduced prompt-writing time, although challenges such as transparency in fact verification and privacy remain, with data running locally on the user’s laptop to mitigate some concerns. The project highlighted the parallels between human and infrastructure data models, demonstrating the value of graph databases in visualizing the deep interdependencies within data systems, and underscored the educational benefits of experimenting with such technologies for engineers interested in advancing AI and automation.

Infrahub is a platform designed to organize and manage infrastructure data by integrating devices, services, and various components into a structured system, offering a centralized way to connect data, logic, and automation. With its change management workflow inspired by Git, users can create branches to safely test and implement changes without affecting the main environment, ensuring full visibility and version history. Infrahub's workflow includes a feature called Proposed Changes, allowing teams to review, discuss, and approve updates in a controlled manner, while enforcing mandatory approval policies in its Enterprise edition for stricter governance. The platform also supports built-in and user-defined checks to ensure changes adhere to specific rules, such as naming conventions, enhancing consistency and reducing errors. Furthermore, Infrahub extends its branch and version control capabilities to schema changes, offering a comprehensive approach to infrastructure management that combines the principles of software development with infrastructure operations.

AI agents are increasingly interacting with infrastructure similarly to human users, necessitating a shift in system design towards schema-driven frameworks that cater to both AI and humans. This approach emphasizes providing AI with clear, structured data and predictable paths to improve their usability and effectiveness. Schema-driven design, as exemplified by GraphQL, allows for efficient data curation and abstraction, enabling AI to navigate complex systems without human intervention. This is crucial for improving AI performance, as demonstrated by Infrahub, a data management platform that utilizes flexible schemas, a GraphQL query engine, and a knowledge graph model. These components help AI agents understand context and dependencies while maintaining human oversight through git-native workflows. By adopting these AI-ready frameworks, systems can better accommodate the unique needs of AI agents, paving the way for more effective and efficient automation solutions in the future.