v2.7.0

New feature: background transactions

The /api/transactions/ endpoint, which is used e.g. by the Gramps Web Sync addon, now has an optional background boolean parameter. If true, the transaction is processed in the background rather than within the request. This allows a transaction to take much longer rather than the usual 1-2 minutes until a HTTP request times out. It will enable much better experience with the Sync addon when processing a large number of changes. (#593)

Other improvements

• The openai module is no longer imported at the top level, which caused an error when not installing the ai extra dependencies (#594)
• An issue with local prefixed media file directories (#515) was solved
• The codebase has been overhauled, type hints improved and a static type checker added to the cuntinuous integration pipeline (#567)

v2.6.0

Improvements to development server

@dsblank has implement various improvements to the development server, including improved logging and optionally opening the browser, among others. See the documentation for all the options.

Caching for vector embedding model

The fix released in v2.5.3 temporarily slowed down vector search. This release reintroduces caching for the vector embedding model to remedy this.

v2.5.3

This is a bug fix release that solves a regression introduced in v2.5.0, which breaks the search feature in multi-tree setups.

Also, this release introduces a new simple /ready endpoint that can be used as a container readiness probe.

v2.5.2

Improvements

This minor release brings the following improvements & fixes:

• A performance bottleneck has been removed for the metadata endpoint in multi-tree deployments (#568, #569)
• A new environment variable GUNICORN_TIMEOUT has been added to the default docker image to customize gunicorn's timeout (which defaults to 180 s if not specified) #570 by @Illia-M
• A regression with the face detection feature has been fixed in the default docker image. Face detection is not supported on the armv7 platform due to missing library support, but works on arm64 (eg. 64-bit Raspberry Pi)

Python 3.8 support removed

Since Python 3.8 has reach end of life, you will need Python 3.9 or higher to run Gramps Web API starting with this release.

v2.5.1

This release fixes a minor issue with some chat models.

v2.5.0

Semantic search

This release adds semantic search powered by vector embeddings, in addition to the existing full-text search. This leads to the following API changes:

• A new configuration option VECTOR_EMBEDDING_MODEL
• The search CLI command gets an optional --semantic switch to manage the semantic search index (which is separate from the full-text search index, but stored in the same database)
• The /api/search endpoint gets an optional boolean parameter ?semantic
• The /api/metadata endpoint reports whether semantic search is enabled via the server.semantic_search key and the number of objects in the semantic search index via the search.sifts.count_semantic key

AI chat endpoint

Powered by the semantic search, this release adds an AI chat endpoint using the retrieval-augmented generation (RAG) technique (#539).

For setup and usage documentation, see

• https://www.grampsweb.org/install_setup/chat/
• https://www.grampsweb.org/user-guide/chat/

Note that semantic search and AI chat are not supported on the armv7 platform (e.g. Raspberry Pi 32-bit) since they depend on PyTorch. Without these features enabled, Gramps Web continues to support armv7 and we continue to provide images for this platform. The arm64 platform (e.g. Raspberry Pi 64-bit) is fully supported.

Object Query Language

A Python-like query language has been added as an alternative to the "Gramps Query Language" (GQL) by @dsblank #553. See https://github.com/dsblank/object-ql for detais.

Other changes

• Generate app/secret only if SECRET_KEY is unset by @pwalkr in #538
• silence warning gtk version by @JohnRDOrazio in #542
• Fix issue #545: migrate to pyproject.toml by @JohnRDOrazio in #547
• Replace Haar Cascade by YuNet for face detection by @DavidMStraub in #546
• Added Object Query Language by @dsblank in #553
• Provide more insights in emails to admin and self-registered user by @hgohel in #557
• python.linting settings are deprecated by @JohnRDOrazio in #541

New Contributors

• @pwalkr made their first contribution in #538
• @JohnRDOrazio made their first contribution in #542
• @dsblank made their first contribution in #553
• @hgohel made their first contribution in #557

Thank you to all contributors and translators! 🎉

Full Changelog: v2.4.2...v2.5.0

v2.4.3

This is a patch release to resolve issues with database connections not being properly closed in some circumstances.

v2.4.2

This is another quick maintenance release to update the dependency on the new search backend, increasing stability.

v2.4.1

This maintenance release brings one major performance improvement: importing genealogical data should now be much faster, as the incremental search reindex it triggers has been accelerated (#530). The dependence on the new search backend sifts has also been updated to profit from bug fixes there.

v2.4.0

New Search Backend

This release replaces the search backend (based on the unmaintained whoosh library) with a new, taylored and SQL-based backend. This will allow further improvements to the search function in future releases. This is a breaking change, and details are given below.

Configuration changes

The old SEARCH_INDEX_DIR configuration parameter is replaced by SEARCH_INDEX_DB_URI, which can be a SQLite or PostgreSQL database URL. If the new parameter is not set but SEARCH_INDEX_DIR is present, Gramps Web API will default to storing the search index in an SQLite database inside this directory and issue a deprecation warning. This should ensure that all Gramps Web API instances will continue working without configuration changes for now.

API changes

The search endpoint has new parameters: type to filter the search results by the object type and change to filter it by the modification date.

Search syntax changes

The search still supports AND/OR operators, and they are now case insensitive (also and/or works). The wildcard * is supported at the end of a search term. Filtering in the search query e.g. by type: is no longer supported (but this was poorly document anyway and is now supported by the new API parameters).

Search improvements

Indepently from the new search backend, the search indexer has been improved to enhance the search experience:

• Searching for a name now also returns families where a parent's name matches the search term.
• Searching for a name now also returns events where a primary particpant's name, or a participant family's parent's name, matches the search term.