Developers building retrieval-augmented generation pipelines who need document ingestion, chunking, and citation support out of the box.

Kernel Memory is a Microsoft open source service and library for indexing documents and answering questions with citations using vector search and LLMs.

Self-hosted as a service or library, or run serverless, with pluggable storage including Azure AI Search, Postgres, Qdrant, and Elasticsearch.

Used when a project needs a ready-made ingestion and Q&A memory pipeline rather than building RAG plumbing from scratch.

It provides pipeline features like pluggable data pipelines, security filters, and long-running ingestion that plain vector DB SDKs lack.

Works with Semantic Kernel and pluggable vector stores, embedders, and LLM connectors including Azure OpenAI and OpenAI.

Open source, MIT licensed, self-hosted, labeled by Microsoft as a research project, actively pushed to as of mid 2026 with roughly 2100 GitHub stars.