gezk
VERSION 0.5 · PRELIMINARY
An open format for knowledge catalogs. A .gezk file is a
portable, read-only body of reference documents with full-text and vector indexes,
shipped as one file that any reader can search, cite and browse offline. It is built
from parts that already exist everywhere — ZIP, SQLite, JSON, brotli, Ed25519
— so a reader needs no vector-database extension, and no tie to any one product.
JSON Schemas
These URLs are the $id of each schema. They are generated from the
reference implementation, so they follow the specification rather than the other
way round. Refinements JSON Schema cannot express — NFC normalization of
document ids, the reserved-name rule on versions — are enforced by conforming
readers on top of these files. The path carries the format version, so a later
line never overwrites what catalogs published under 0.5 point at.
| Schema | Validates |
|---|---|
| catalog-manifest | manifest.json inside a .gezk archive |
| registry-index | A publisher’s registry index.json |
| source-notices | LICENSES/source-notices.json |
| embedding-profile | The embedding block of a manifest |
| chunking-profile | The chunking block of a manifest |
| catalog-document | One normalized document fed to a compiler |
The specification, the conformance kit every implementation is held to, and a reference reader in Python live in the gezk repository. The specification is licensed CC BY 4.0; the code, schemas and conformance kit are MIT.
gezk on GitHub