Private books · local models · visible citations

Knowledge that
stays in the room.

Local Knowledge Terminal turns reviewed books and dictionaries into calm, multilingual knowledge cards on a Raspberry Pi—without making the cloud your library.

Working browser software; the $250 sprint uses existing hardware and does not include a hardware product.

Concept product view of the Local Knowledge Terminal showing its current multilingual word card interface
Concept visual grounded in the current display prototype and working interface; hardware configuration may vary.
LOCALPrivate corpora and model inference
CITEDSources come from retrieved records
EN · 日 · 中Multilingual cards with reading aids
PI 5An 8 GB target, not a cloud dependency

A terminal, not another chat tab

Built around durable cards.

The browser interface is the first display. E-paper, microphone, and audio remain separate adapters, so the knowledge core does not depend on one piece of hardware.

01

Ground the answer

Retrieve from a private book or reviewed dictionary first. Citations are attached from those records, not invented by the language model.

02

Compose one card

Question, Answer, Word Card, Word Origin, Root, and Affix remain distinct modes with one clear teaching point per screen.

03

Keep it local

SQLite stores accepted knowledge. A local Qwen model prepares bounded artifacts; private books, indexes, and model files stay outside the public repository.

Local Knowledge Terminal word origin graph interface

Words with a memory

See where a word came from.

Word Origin uses a directed ancestry graph rather than a paragraph pretending to be one. Nodes can distinguish book evidence from model context, while the primary card stays readable.

  • English meaning remains primary.
  • Japanese and Chinese can add pronunciation guides.
  • Root and affix views reuse accepted atomic evidence.

Related work you can inspect

The collection work is already public.

The fit sprint is new, but it is grounded in years of organizing technical, multilingual, and book-length material into durable reading tools. These projects show the relevant craft without pretending that they are paid LKT client engagements.

Physics archive

Susskind lectures

A maintained collection of transcripts, subtitles, notes, TeX, course PDFs, and a public web reader.

Inspect the archive →

Reproducible editions

PocketPolyglot

A Studio, CLI, and TeX pipeline for aligned multilingual books with pinyin, furigana, and reusable source data.

Inspect the pipeline →

Public output shelf

LinguaLeaf

A browsable shelf of multilingual reading editions, including Chinese classics and history in several formats.

Browse the shelf →

Related public work is evidence of collection handling and publishing experience, not a claim that those projects were produced through this paid sprint.

Who the pilot may fit

A small, private place for a serious collection.

Language study

Turn a reviewed lexicon or reading collection into large, calm cards with multilingual equivalents.

Libraries and exhibits

Explore a bounded collection through an ambient display without exposing the source books to a cloud service.

Private research

Build a focused terminal around material that should remain on local storage, with visible evidence boundaries.

Founding collection-fit sprint · $250

Bring one collection and one real use case.

A deliberately small first engagement: one private collection, one language goal, and one existing machine. After a free fit check, the paid sprint produces a written data and privacy map, a small browser proof from a representative sample when the source is usable, and a clear go/no-go recommendation for a larger deployment.

  • Customer-provided material only; you must have the right to use it.
  • Hardware, shipping, custom OCR, and a production deployment are not included.
  • Payment is requested by Stripe only after the fit and scope are accepted.