An e-reader that actually remembers what you read

What it does

ReadAny is a cross-platform e-book reader built on Tauri 2 and Expo that imports EPUB, PDF, MOBI, AZW, and other formats into a local SQLite library. It indexes text with local embeddings via Transformers.js, then lets you search semantically, highlight passages, take Markdown notes, and hold RAG conversations through LangChain.js. The stack is offline-capable by default, with optional WebDAV sync to keep libraries consistent across desktop and mobile.

The interesting bit

Instead of treating search as a keyword hunt, ReadAny uses hybrid vector-plus-BM25 retrieval to surface concepts even when you cannot remember the exact phrasing. The project also ships a single GPL-3.0 codebase that covers desktop, mobile, TTS, and sync, with the team noting that App Store binaries may carry a fee while the source remains free to build.

Key highlights

• Local-first by default: embeddings and vector store run offline via Transformers.js, while AI chat can call OpenAI, Claude, Ollama, DeepSeek, or other providers using your own keys
• Hybrid search mixes vector retrieval and BM25 to find passages by meaning, not just exact wording
• A “Skills System” offers built-in prompts—summarizer, character tracker, etc.—and lets you author custom ones
• Notes export to Markdown, HTML, JSON, Obsidian, or Notion; TTS supports Edge, Browser, and DashScope engines with 100+ voices
• Reading statistics include GitHub-style heatmaps and streak tracking; WebDAV sync handles conflict resolution for concurrent edits

Caveats

• PDF text reflow is on the roadmap but not yet implemented
• Mobile builds require an Expo development client; the README explicitly states Expo Go is incompatible because of native module dependencies
• Official app store binaries may be sold to cover certificate costs, though the GPL-3.0 source is always free to compile

Verdict

Grab it if you want to treat your e-book collection as a private, searchable archive rather than a pile of files; pass if you prefer a minimal reader without AI, sync, or build complexity.