Added ISCC_WEB_SEMANTIC_DEFAULT to set the backend default for the semantic query parameter
on POST /api/v1/iscc when it is omitted (off by default), decoupled from the iscc-sdk
ISCC_SDK_EXPERIMENTAL default
Added ISCC_WEB_UI_SEMANTIC_DEFAULT to control the initial state of the Semantic Code toggle in
the bundled demo frontend (on by default)
The demo frontend now reads runtime configuration from the backend, so the Semantic Code toggle
default and the file auto-deletion notice follow the server settings (ISCC_WEB_UI_SEMANTIC_DEFAULT
and ISCC_WEB_STORAGE_EXPIRY) instead of hardcoded values
Added an opt-in GPU Docker image variant published under -gpu tags (:0.3.1-gpu,
:0.3-gpu, :main-gpu) on a CUDA base that runs the experimental semantic codes on an
NVIDIA GPU; the default image stays the lean CPU python:3.13-slim build
Tightened client-side ISCC code input validation in the demo frontend decoder
Added optional semantic ISCC-UNITs for text and image content via the semantic query
parameter on POST /api/v1/iscc (experimental 5-unit ISCC-CODE, off by default)
Added granular content simprints to ISCC results, configurable per request via the
granular query parameter (on by default)
Added POST /api/v1/simprint endpoint for generating content simprints from plain text
Added ISCC-UNIT listing (units field) to ISCC results
Added wide ISCC-SUM fallback for unsupported media types instead of an error response
Added configurable CORS support via ISCC_WEB_CORS_ORIGINS
Redesigned the demo frontend as an adaptive decoder canvas with media file, plain text and
ISCC code inputs, per-unit bit visualizations and side-by-side ISCC comparison
Unified the interactive API docs at /docs on Stoplight Elements
Fixed metadata embedding when media storage and temp directory are on different drives
Updated core stack to Python 3.13, BlackSheep 2.6, iscc-sdk 0.9.3 and pydantic v2
(breaking: responses now carry iscc-schema 0.8.0 $schema/@context URLs and a
generator field; running from source requires Python >=3.13)
Updated frontend toolchain to Node 24, pnpm, Vite 8, Vue 3.5, TypeScript and Uppy 5
Rebuilt the Docker image on python:3.13-slim with pre-downloaded semantic models; images
are published to ghcr.io with main and semver tags (no latest tag)