Save anything. Find it later. Discover what connects.

Twinkle is an AI-powered personal knowledge base. You save links, videos, tweets, images, audio and notes, and Twinkle transcribes, summarizes and organizes them so you can find anything by meaning, chat with your library, and connect it to AI tools like Claude, Cursor and ChatGPT. The idea is simple: you already save a lot, and you lose most of it. Twinkle makes the things you keep findable again, and shows you how they connect.

Almost anything: web pages and articles, YouTube and TikTok videos, Instagram reels, X and Bluesky posts, Reddit threads, GitHub repos, Spotify tracks, Hacker News discussions, screenshots and photos, voice memos and audio files, and plain text notes. If it has a URL or you can see or hear it, Twinkle can usually take it.

A bookmark just stores a link, and a notes app makes you do the organizing. Twinkle reads the content, transcribes the audio, writes a summary, and connects it into a graph you can search by meaning. You never tag or file anything by hand, and your whole library is reachable from the AI tools you already use.

Twinkle turns everything you save into vector embeddings, which capture meaning rather than exact words. So a search for “managing burnout” can surface a podcast that never says those words but is about exactly that. It searches across transcripts, entities and full text at once and ranks by relevance.

As Twinkle processes your saves, it extracts entities (people, places, products, ideas) and links them across everything you’ve kept. Open any item or entity and you’ll see related saves you’d forgotten, plus a “tell me more” that pulls them into a short AI brief. Finding one thing leads you to the next.

Yes. Ask a question in plain language and Twinkle answers from what you saved, with links back to the sources. It is grounded in your own library, so it will not make things up about content you never kept.

Yes. Twinkle runs a Model Context Protocol (MCP) server, so Claude Desktop, Claude Code, Cursor, VS Code, Windsurf and ChatGPT can search your knowledge base and save new things to it directly. MCP access is included on every plan, including Free.

Yes. There are native iOS and Android apps on the same library as the web app, so you can capture on the go, including sharing straight from other apps on Android. The web app also installs as a PWA.

Twinkle uses a strong speech-to-text model and handles long recordings, multiple speakers and background music. Quality depends on the audio, but for clear speech in videos, podcasts and voice memos it is reliable enough to search and quote.

Yes. Every item is isolated to your account with row-level security in the database, so nobody else, signed in or not, can read your library. Privacy was built in from the first migration, not bolted on as a setting.

You can delete any item at any time, and deletion removes it and its derived data. Your content stays yours; we are not in the business of locking it up.

Yes. Twinkle has a REST API with personal keys, so you can capture and search programmatically, plus the MCP server for AI tools. Both are available to use however you like.

No. Twinkle summarizes, tags and connects everything automatically. If you want structure you can make collections by hand, but the whole point is that the filing happens for you.

Yes. The Free plan holds 50 items with no card required and includes AI search, chat, the knowledge graph and MCP access. Pro and Max raise the limits when you are ready for a bigger library.