01 / THE PROBLEM
A personal assistant should work with local materials and retain useful context without becoming an opaque black box.
Noelle explores a local-first assistant architecture that connects a language model to browsing, file analysis, semantic retrieval, and voice interactions. The project is framed as an ongoing personal agent experiment.
02 / DESIGN DECISIONS
Use Ollama to explore local model workflows.
Use FAISS as a building block for search over indexed content.
Use Playwright for controlled web-browsing automation.
Keep the assistant voice separate from the capabilities it is allowed to invoke.