One story, reported three ways
Related coverage is folded together by meaning…
Local-first personal reader
Rinwa groups duplicate reporting by meaning, finds ideas even when the words differ, and lets translation settle beneath the original. Your library stays local and readable without an account.
No account required · macOS is the primary quality target · Android is a development preview
Related coverage is folded together by meaning…
The links we keep become a map of what mattered…
Why a database on your own device still matters.
The best reading tools disappear. They hold the text still and let understanding arrive at its own pace.
A good reading room does not compete with the page. It softens the edges of the world until only the words remain.
好的阅读空间不与页面争夺注意力。它让世界的边缘渐渐柔和,直到眼前只剩文字。
Images keep their place. Lists keep their rhythm. Code remains exactly what the author wrote.
图片留在应有的位置,列表保有原来的节奏;代码则完整保留作者写下的样子。
“Let the tools disappear. Keep the understanding.”
Two things, done unusually well
Rinwa does not put an AI dashboard beside your reading. It uses intelligence in two places where it can make the page quieter: organizing meaning, and crossing language.
Embedding similarity recognizes the same event across different headlines and sources. Recent coverage folds into one expandable story; natural-language search brings back passages even when you cannot remember the original words.
Embedding Is All You Need — to find your way back through what you read.
SEARCH BY MEANINGthe article about devices owning their data
4 passagesThe source paints immediately. Translation arrives block by block without flattening lists, moving images, or touching code. Translate the whole article—or only the sentence you selected.
A reader should preserve the shape of an argument. Lists keep their rhythm, images keep their place, and the original remains the point of reference.
阅读器应当保留论述原有的形状。列表维持节奏,图片留在应有的位置,原文始终是理解的参照。
const source = article.original;
One reading surface
Feeds, web pages, and built-in video sources do not become separate silos. RSS and the early Connector foundation enter one local content layer and share the same extraction, translation, search, highlighting, and note tools.
Sources enter Rinwa
Read
Translation appears block by block without replacing or delaying the source. Code, dates, numbers, links, and already-Chinese text stay untouched.
Organize
Search articles, notes, and highlights locally. Save the conditions and the result stays in the sidebar, collecting what you care about over time.
Remember
Highlights use quote context to relocate after re-extraction—or stay honestly unanchored. Notes, tags, references, and Obsidian Markdown remain durable user data.
Search your understanding
Rinwa combines local full-text indexing with optional multilingual vector retrieval. Models, indexes, and articles stay on-device. If a model is unavailable, search falls back quietly instead of breaking your library.
Searching on this device · nothing sent to the cloud
…a local database does not mean giving up sync. It means every device first owns a library that can work by itself.
Lobsters · 3 hours agoSync operations, not the entire database file. Each device keeps its own media and indexes.
#sqlite · #sync · ObsidianKeep what matters
Articles expire and services change. Your notes should not. Rinwa treats highlights, tags, references, and edited summaries as user data, never as disposable AI cache.
“A good reading room does not compete with the page.”
好的阅读空间,不与页面争夺注意力。
Calm is not the absence of features. It is making them appear only when needed.
Local first, honestly
Subscriptions, articles, media, and indexes live locally first. Saved reading continues offline.
Optional Rinwa Sync exchanges selected states and operations while every device keeps an independent library.
Extraction, translation, and summary caches have separate versions. A failed refresh preserves the last successful result.
OPML, Obsidian, and encrypted full backups keep subscriptions and knowledge portable.
Under the reading surface
Rinwa keeps the reading surface calm, but the boundaries underneath are explicit: local ONNX search, provider-aware AI, and operation-log synchronization.
A pinned EmbeddingGemma model runs locally. Sparse and dense ranks meet through RRF; failure returns to full-text search.
Local context selection, hybrid retrieval, rolling memory, token budgets, citations, streaming, model discovery, and reasoning controls wrap OpenAI, OpenRouter, Ollama, or a custom provider.
Devices exchange immutable operations through CouchDB, then merge fields deterministically with HLC clocks and tombstones.
The complete guide now covers reading, translation, AI, saved searches, stable highlights, Obsidian, MCP, connectors, backup, sync, privacy, and honest platform status.
Open DocumentationOne interface, many screens
React and Tauri 2 let platforms share the interface and reading logic. When a native capability is not ready, Rinwa says “in development” instead of showing a false green checkmark.
Native windows, hierarchy swipes, directory picking, secure storage, and in-app web windows are in the desktop shell.
Signed and notarized package in preparationThe shared UI and preview package work today. Local SQLite, Keystore, and background refresh still need device validation.
Download ARM64 APK ↓The Tauri project and interface boundary are ready; database, background work, signing, and store validation remain.
Requires real devices and Apple Developer workflowWindows reuses the React surface and local data model. A release awaits validation on the native toolchain.
Requires Windows build and signing validationA few honest answers
Rinwa is the next-stage brand. The current application and preview package still say Luma. The migration will preserve bundle identity, backup formats, and secure-storage compatibility.
No. The open-source client and a single-device SQLite library work without a paid service. kappain plans to offer managed Rinwa Sync as an optional subscription; self-hosting remains part of the architecture.
kappain develops Rinwa as an open-source reader. Reading, local storage, import and export are product foundations; the commercial service is convenient hosted synchronization and its operations, not ownership of your library.
Only to the provider you configure, and only after explicit data-sharing consent. A self-hosted OpenAI-compatible proxy is supported.
macOS is the primary quality target, but the product is still a development preview. Start with a subset of feeds and keep your current reader nearby; OPML export and backups are available.
Rinwa · 页间