// the find
reorproject/reor
Private & local AI personal knowledge management app for high entropy people.
Reor is an Electron desktop app that keeps your notes in local markdown files and runs AI features (RAG Q&A, semantic search, auto-linking) entirely on your machine via Ollama and LanceDB. It's for developers who want the AI note-taking experience without their notes leaving their laptop. Think Obsidian with a vector database bolted on.
The architecture is honest: embeddings live in LanceDB next to your files, Ollama handles inference, nothing touches an external server unless you explicitly configure a remote API. The dual-generator mental model (LLM retrieves for Q&A; human sees the same retrieval in the sidebar) is a clever way to frame what RAG actually does in a PKM context. Cross-platform Electron with a forked BlockNote editor means the writing experience is decent and not just a textarea. Supports OpenAI-compatible APIs as a drop-in, so you can swap Ollama for a remote model if your hardware can't handle local inference.
The BlockNote editor is vendored wholesale into src/lib/blocknote rather than taken as a package dependency — that's a maintenance trap; upstream bug fixes won't reach them automatically and the fork diverges silently. Last commit was May 2025 and the Discord announcement still reads like an early-access pitch, so the velocity question is real. No note graph visualization (Obsidian's killer feature for the target audience), and the README punts on frontmatter support with 'may not parse correctly' — that will break anyone importing from Obsidian or Logseq. The AGPL license means any SaaS wrapper is immediately a compliance problem.