← All work
mirubeanWeb platform built · testing with live models
Screenshot research inbox · 見豆

Your screenshots, finally sorted — and fact-checked.

For anyone whose camera roll is full of screenshots they never look at again

Save a screenshot or a link. mirubean works out what it is, looks it up with sources, files it in the right folder — and answers questions about everything you saved.

Identified
Sony WH-1000XM5

Wireless noise-cancelling headphones, about 30 hours of battery.

[1] sony.com.hk[2] Product review
Shopping · Audio+1
3ways to capture
3export formats
1question box for everything
The problem

Camera rolls fill up with screenshots of products, places, posts and ideas that are never looked at again. Saving is easy; making sense of what you saved is not.

The product
mirubean library of filed cards with a restaurant card open
mirubean iOS restaurant card
mirubean iOS Today screen
Today: new screenshots found, and what Miru is working on.
mirubean iOS processing complete
Each save is read, researched and filed — step by step.
mirubean digest on the web
A digest of what was saved and learnt.

Screens from the mirubean web and iOS prototypes, with sample data.

Inside the pipeline

One screenshot becomes a researched, filed note you can ask about.

You save; Miru does the looking, the research and the filing — and shows its sources.

Before
  • Hundreds of screenshots you never open again
  • Searching for the same thing twice
  • No way to find what you once saw
With mirubean
  • Every save identified and researched
  • Organised into folders automatically
  • Searchable and askable — with sources
01 · CaptureOutput
What you get

Save from anywhere in one tap — screenshots, links, even videos.

Links are parsed on the server; videos are fetched by a media worker and sampled into frames and captions.

link parsingframe samplingcaption extraction
Sources on every card

Facts link back to where they came from, so you can check them.

You approve the structure

New folders are proposed, never forced.

Your library, portable

Export to your own AI tools as SKILL.md, markdown or an MCP server.

What we built
  • A vision agent with tools — zoom, frame sampling, captions and transcripts for videos — rather than plain OCR
  • Cited research through our own search and research engine, so every card shows where its facts came from
  • Automatic filing into a folder tree; the agent may propose a new folder for you to approve
  • Ask across everything you saved, plus a daily or weekly digest of what you learnt
  • Export to your own AI tools: a SKILL.md folder, digest markdown or an MCP server
Built with

We build the AI layer ourselves, on proven platforms that scale.

AI layer
IN-HOUSEVision agent with tools · zoom, frame sampling, captionsIN-HOUSEProprietary search & research engine · citedIN-HOUSEFiling model · folder proposals with approval
Front end
Nuxt 4VueTypeScript
Back end & media
Nuxt server (Nitro)Media workerFFmpeg
Integrations & mobile
IN-HOUSEMCP server · SKILL.md exportFlutter (iOS first)

Developed by P. Chan & Co.

Next case study
SoleHK→
No more shoebox of receipts at tax time.