An Obsidian-Style Editor Where the AI Stays Home

What it does Reor is a desktop markdown editor that chunks and embeds every note into a local vector database. It surfaces related notes via vector similarity, answers questions using RAG against your corpus, and supports semantic search—all without requiring an internet connection for the core engine. You pick a single directory on your filesystem, and the app works entirely within it.

The interesting bit The project treats the human as a second RAG generator: while the LLM retrieves context to answer queries, you can toggle a sidebar in the editor to see related notes “retrieved” from your own corpus, effectively cross-referencing your current thought against past ones. It is an unusual framing that blurs the line between author and model.

Key highlights

• Runs LLMs and embeddings locally via Ollama, Transformers.js, and LanceDB; cloud APIs are strictly optional.
• Stores everything in a local directory with an Obsidian-like markdown editor.
• Supports both local Ollama models and OpenAI-compatible remote endpoints if you want to mix sources.
• Open source under AGPL-3.0.

Caveats

• Importing notes from other apps is currently manual: you must copy markdown files into the chosen directory, and frontmatter may not parse correctly.
• The README notes that integrations with other apps are “hopefully coming soon,” so migration helpers are still missing.

Verdict Developers and researchers who want a private, local second brain with AI-powered linking should take a look; if you need polished import pipelines or collaborative cloud sync, this is still a solo, filesystem-bound experiment.