// the take
A prototype Obsidian plugin that generates draft paragraphs by retrieving relevant notes from your vault (via OpenSearch keyword search plus a semantic embedding index) and feeding them to an LLM as context. Built for Obsidian users who want retrieval-augmented drafting over their own notes rather than a generic AI writing tool, and for anyone curious about a minimal local RAG pipeline.
Combines keyword search (OpenSearch/BM25) with semantic retrieval rather than relying on embeddings alone, which is the right call for note corpora where exact terms matter. Clean separation of concerns: a local FastAPI retrieval service does the heavy lifting while the Obsidian plugin is a thin TypeScript client, so your notes and embeddings never leave your machine. The write-up accompanying the repo explains the retrieval design decisions in more depth than the README does, which is useful if you want to adapt the approach rather than use it as-is.
No commits since June 2024 — over a year stale, and the two open TODOs (Claude support, whole-document context) were never picked up, so it's effectively abandoned as a maintained project. It's explicitly a prototype: only OpenAI is wired in for generation, there's no error handling discussion, and the retrieval quality depends on tuning you'd have to do yourself. Setup is heavy for what it does — Docker, a running OpenSearch container, and a separate semantic-index build step before the plugin even works, which is a lot of infra for a personal note-taking tool. Single-maintainer project with no visible tests beyond a CI lint workflow, so expect to debug it yourself if something breaks on a newer Obsidian or Python version.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →