kernel-memory

Why we included this project

For developers studying retrieval-augmented systems, this Microsoft reference implementation shows how to build an indexing service end to end. It ingests documents through continuous data pipelines, embeds them, and answers natural-language queries with citations and source links. The code runs as a web service, a Docker container, or a .NET library, and it includes a plugin that works with Semantic Kernel, ChatGPT, and Copilot. Because the repository is explicitly archived and unsupported, treat it as a blueprint for architecture and design rather than production code. The examples around synthetic memory and custom semantic memory pipelines are especially useful for teams exploring those ideas.