# Why bitmark

bitmark is the open knowledge format. The people who own knowledge represent
it once, as typed, addressable bits, and every reader uses the same bits: a
person, a search system or an AI agent.

## Who owns the chunks?

Today most knowledge reaches AI as chunks that someone else cut: a crawler or
a search index slices a document by length and loses its structure, its
source and its rights on the way. bitmark turns this around. The bits are the
document's own parts, each with a type, a permanent address and its place in
the source, so the chunks belong to the people who wrote the content.

bitmark is how content is represented. Turning PDFs, XML or scans into
bitmark is a service that companies offer on top of the standard.

## RAG that cites

Every bit can carry a permanent ID, the customer's own ID from their source
system and its exact place in the original. An answer built from bits can
therefore point to the bit, the page and the paragraph it came from, and show
it in context.

```bitmark
[.smart-standard-article-normative]
[@customerId: p_4_1 ]
[customer▼ p_4_1 ]
The rated values are given in (+customer ref:tab_1)Table 1(-customer ref).
```

This bit is a normative provision with the customer's ID `p_4_1` and a link
to the customer's table `tab_1`. Both are kept through every conversion the parser makes; see
[The Rosetta Map](https://docs.bitmark-association.org/concepts/rosetta-map.md).

## Updates

A bit keeps its ID from one edition to the next, so a reader can match the
bits of a new edition with the ones it already holds. The parser's semantic
diff (`bitmark diff old.bitmark new.bitmark`) shows what changed as bitmark
an author would write. A content hash and a last-modified date on every bit,
so that readers fetch only what changed, are a proposal; see
[Versions and updates](https://docs.bitmark-association.org/concepts/versions-and-updates.md).

## Agents

Bits arrive in document order with their chapter structure. An agent can read
the way a researcher does: the table of contents first, then a summary, then
the section it needs, then the references to other documents. Delivered as
JSONL, one bit per line, a document streams and can be read in part.

## People

The same bits serve people: in apps, in print and through screen readers.
Images carry alternative text or a decorative mark, tables have real header
cells, and a bit in another language says so. Nothing in a bit describes
how it looks; each app chooses its own design.

## Seen from outside

The Swiss Data Alliance white paper
[*AI-ready Data*](https://iris.unil.ch/handle/iris/285961) (version 1.0,
26 October 2026), written for the AI Action Plan for Switzerland, calls
bitmark "the Swiss open-source standard" and recommends it for turning
unstructured or paper-only content into AI-ready data (p. 37). Its Swiss Life
Digital Heritage case represents about 10,000 books going back to 1850 as
bitmark chunks for a retrieval system (pp. 25–26); at the time of writing the
system is in testing, with full go-live planned for the end of 2026.

Next: [Getting started](https://docs.bitmark-association.org/concepts/getting-started.md).