Constella provides local memory for AI agents

Constella is an open-source application that builds a personal knowledge base from your local files, offering semantic search, a knowledge graph, and an MCP interface that lets AI agents such as Claude Code retrieve information without uploading it to the cloud. The app scans directories, Obsidian notes, PDF, DOCX, Markdown files and images, splits them into small chunks, creates embeddings and stores them in LanceDB, while the meta-data and the graph itself are kept in SQLite. The result is a semantic search that returns contextual matches rather than simple keyword hits, and a graph that enables queries like “what is related to what”. Constella supports both local and cloud models and provides an MCP interface so AI agents can use your knowledge base as long-term memory.

The full project flow is: files → chunking + embedding → LanceDB + SQLite → knowledge graph → RAG / MCP / AI agents. All components run on your machine, with no separate server required.

The code is available on GitHub, and the course “AI Agent Development” offers hands-on practice of similar architectures. In the navigation interface you can find sections on jobs, tasks and interviews, as well as a dataset library for data scientists.