Features
Everything Twinkle does, working the moment you save.
Twinkle is an AI knowledge base that captures any source, understands it, and makes it findable: by search, by chat, and from the AI tools you already use. Here is the whole toolkit.
01 / Capture
Save from anywhere, in its full detail.
Twinkle has purpose-built extractors for the sources people save from most, so you keep the transcript, the thread, the comments and the metadata, not just a link.
YouTube
Full transcripts, channel + metadata
TikTok
Audio transcription, carousel vision
Instagram
Reels transcribed, captions, carousels
X / Twitter
Whole threads, authors, quoted posts
Reddit
Post + nested comment trees
GitHub
Readmes, issues, pull requests
Spotify
Tracks, artists, albums, playlists
Hacker News
Posts with the top discussion
Bluesky
Posts, profiles, embedded media
Articles & pages
Clean text, boilerplate removed
Images & screenshots
Vision analysis, OCR, QR, products
Audio & voice
Transcription, speakers, song ID
Plain notes
Thoughts, ideas, quick reminders
01 / Understand
It reads, watches and listens, so you don’t re-read anything.
Every save runs through a two-phase pipeline. First, Twinkle extracts the content and writes a clean summary in seconds. Then it goes deeper: pulling out the people, places, products and ideas, searching the web for context, and re-embedding everything with full meaning.
- Video and audio transcribed automatically
- Images read with vision AI: text, products, QR codes
- Summaries and titles written for every item
- People, places and topics extracted as entities
02 / Find
Search the way you remember things.
Twinkle searches by meaning, not exact words. Describe the gist, like “that thread about cold plunges,” and it matches across transcripts, entities and full text at once. When you’d rather ask than search, chat with your whole library and get answers with sources attached.
- Semantic search over everything you saved
- Chat grounded in your knowledge, with citations
- Deep Dive briefs that expand on any item or topic
- Filter by type, date, source or entity
03 / Connect
A knowledge graph that builds itself.
Twinkle treats what you save as a network, not a pile. It identifies the entities inside your content, merges duplicates, and draws the relationships between them, so a person mentioned in a podcast links to the article that quotes them and the product they launched.
- Nine entity types, from people to products to ideas
- Automatic deduplication and alias tracking
- One-hop relationships you can browse visually
- Collections, by hand or grouped for you
04 / Audio intelligence
Voice memos and recordings, made searchable.
Record a thought, a meeting or a lecture and Twinkle transcribes it, works out who said what, and even recognizes songs. What was a forgotten audio file becomes text you can search and chat with.
- Accurate transcription for long recordings
- Speaker separation for multi-person audio
- Music recognition with links to listen
- Automatic classification: memo, meeting, podcast
05 / Act
Action links turn saves into next steps.
A saved track gets a play button. A product gets its page. A place gets a map. Twinkle figures out the useful next action for each item and adds the link, so your library does something, not just sits there.
- Curated automatically during enrichment
- Play, buy, open, book: matched to the content
- Pulled from the entities Twinkle finds
06 / Everywhere
On the web, on your phone, and over the API.
Capture from a browser, share straight from your phone, or send things in programmatically. Twinkle is a web app and a native iOS and Android app on the same library, plus a clean API and an MCP server for your AI tools.
- Web app (installable PWA)
- Native iOS and Android apps
- REST API with personal keys
- MCP server for Claude, Cursor and ChatGPT
Bring it into your AI tools.
Connect Twinkle to Claude, Cursor, VS Code, Windsurf or ChatGPT over MCP and let them search and add to your knowledge directly.
07 / Questions
More on how it works.
Does Twinkle transcribe videos and audio?
Yes. YouTube, TikTok, Instagram reels and uploaded audio are transcribed automatically, and the transcript becomes searchable text you can chat with. Multi-speaker recordings are separated by speaker, and music is recognized and linked.
What is the knowledge graph?
As Twinkle processes your saves, it extracts entities (people, organizations, places, products, technologies and concepts) and the relationships between them. The result is a browsable graph that shows how everything you have saved connects, instead of a flat list of links.
How is semantic search different from keyword search?
Keyword search needs the exact words; semantic search matches meaning. Twinkle embeds your content as vectors, so a search for “managing burnout” can surface a video that never uses those words but is about exactly that.
Can I organize things manually?
Yes. You can create collections and add items by hand, pin and sort them, and customize icons and colors. Twinkle also suggests and auto-creates collections from related items, so you can do as little or as much organizing as you like.
What are action links?
During enrichment Twinkle works out the useful next step for a saved item and adds a link for it. A saved song gets a play button, a product gets its page, a place gets a map. They are generated from the entities it finds, so your library does something rather than just sitting there.
What does the two-phase pipeline mean?
Every save runs through two passes. The first, in seconds, extracts the content and writes a summary so the item is usable right away. The second goes deeper: extracting entities, searching the web for context, curating action links, deduplicating entities, and re-embedding everything with full meaning.
What can Twinkle do with an image?
Vision AI classifies what the image is (a receipt, business card, menu, QR code, product, screenshot and more), reads the text, extracts structured fields, and resolves products or subjects to official pages. A screenshot of a video can even be resolved back to the original source.
What is Deep Dive?
Deep Dive generates an AI research brief on any saved item or entity: a short summary plus structured analysis, new context, related concepts and suggestions for what to explore next. It turns a single save into a starting point.
Does it work the same on mobile?
Yes. The native iOS and Android apps share the same library and features as the web app, including capture, search, chat and the graph, and they add native capture like sharing straight from other apps and recording voice memos.