YetiSearch Pro

v2.1.2

5 days ago

1. • Plugins that change documents through onYetisearchBuildDocument now see pages saved in the admin too. Before, the event fired only during a full reindex, so a page saved in the admin was stored as built until the next reindex

v2.1.1

2 weeks ago

1. • Noise calibration. Semantic search now works out for itself where nonsense stops and real searches start on your content. After each full embedding run it embeds a set of meaningless test phrases, measures how closely they match your pages, and sets the cutoff just above that. A small site and a large one need very different cutoffs, and one fixed number either let nonsense through on the large site or turned real searches away on the small one. Set Noise Calibration to Off to keep the 2.1.0 behavior
   • Calibration Strictness setting (default 2.0) moves the cutoff up or down without re-embedding. Lower it if good matches are missing, raise it if nonsense gets through
   • The YetiSearch Pro page in Admin Next shows each index's noise cutoff, whether it was measured or is the default, and any test phrases that were skipped because they match your own content. bin/plugin yetisearch-pro embed --status shows the cutoff too
2. • Requires yetidevworks/yetisearch 2.5.1
   • bin/plugin yetisearch-pro embed measures the cutoff even when every page is already embedded, so an upgraded site gets calibrated without re-embedding anything

v2.1.0

2 weeks ago

1. • Semantic search. Search can now find pages by meaning as well as keywords: "can't log in" finds the password reset page, "hire an advisor" finds the consulting page. Turn it on under Semantic Search in the plugin settings and point it at an embedding API: OpenAI, OpenRouter, a self-hosted Ollama server, or anything else that speaks the OpenAI embeddings API. Keyword and meaning rankings are merged, so pages that match on both rank first. Without it enabled nothing changes
   • Pages are embedded after each save once the response has been sent, after a reindex, by a new Semantic Search Embedding scheduler job (every 15 minutes, on by default, runs only while semantic search is enabled), or with the new bin/plugin yetisearch-pro embed command (--status, --test, --reset). Only new or changed pages are sent to the provider, so reindexing unchanged content costs nothing. bin/plugin yetisearch-pro index embeds afterwards unless given --no-embed, and bin/plugin yetisearch-pro schedule --task embed runs the job by hand
   • The YetiSearch Pro page in Admin Next shows embedding coverage with a progress bar, and has Embed, Test connection and Re-embed all buttons
   • The search modals show keyword results while the visitor types and replace them with results ranked by meaning once typing pauses, so no keystroke waits on the embedding provider. The /ys endpoint accepts semantic=0 for keyword ranking only and reports semantic in its response
   • Settings for provider, API base URL, API key (or env:NAME to read it from the environment), model, dimensions, the balance between meaning and keywords, minimum similarity, minimum query length, search timeout and query/document prefixes, plus a Semantic Search toggle per index. If the provider is slow or unreachable, searches fall back to keywords
2. • Requires yetidevworks/yetisearch 2.4, under its correct package name (the plugin asked for yetisearch/yetisearch, a name that does not exist on Packagist)
3. • Pages saved in Admin Next are indexed again right away. Admin Next saves pages through the API plugin, which fires onAdminAfterSave rather than onFlexAfterSave, so realtime indexing never ran for those saves
   • On a multi-language site, a page saved with realtime indexing on no longer ends up in the index twice. The realtime update built the document id without the page's language, so it added a second copy next to the one a full index had written instead of replacing it

v2.0.3

3 weeks ago

1. • A page reached through version fallback no longer has its Twig processed twice, which could abort the whole index run. A page that documents Twig inside {% verbatim %} threw "Unexpected end of template" on the second pass, and because that happened outside the guard around normal rendering, one such page stopped the run instead of falling back to raw markdown

v2.0.2

4 weeks ago

1. • An API key limited to specific permissions can no longer use YetiSearch Pro's tools beyond what it was given, even when the account that created it is a super admin. A key limited to YetiSearch Pro keeps working at the level it was given

v2.0.1

5 months ago

1. • Custom environment names are now supported when set explicitly via the Environment Mode field, which is now a free-form text input. Names like localhost, staging-eu, or abc.wydn.com are accepted and used as-is in the index database prefix. Auto mode keeps the original production / staging / development behavior so existing index databases on default installs are not renamed by the upgrade.
2. • Fixed --env switch being ignored by the indexer CLI when a GRAV_ENVIRONMENT shell variable was already exported (common on MAMP and similar local setups). The plugin now prefers Grav's resolved system.environment over the raw shell variable, so bin/plugin yetisearch-pro index --env=<name> correctly picks up the per-environment config and writes to the right database.

v2.0.0

5 months ago

1. • Added Grav 2.0, API, and Admin 2.0 support

v1.1.2

7 months ago

1. • SQLite backward compatibility: The RETURNING clause (requires SQLite 3.35.0+) is now optional. On older SQLite versions (3.24.0+), a fallback SELECT query is used to retrieve doc_id after upsert operations. This lowers the minimum SQLite requirement from 3.35.0 to 3.24.0 while preserving optimal performance on newer versions.

v1.1.1

7 months ago

1. • Fixed admin reindex failing when other plugins access $grav['admin']: The admin reindex handler removed the admin service from the DI container to avoid interference, but this caused plugins like revisions-pro to crash with "Identifier 'admin' is not defined" when Pages::enablePages() fired onPageProcessed events during page tree rebuilds, resulting in 0 documents indexed.

v1.1.0

7 months ago

1. • Added "Flush & Reindex" button to admin dashboard: New action button between "Reindex All" and "Clear Index" that deletes all existing indexes and rebuilds them from scratch, with a confirmation dialog to prevent accidental use.
2. • Fixed single-strategy index flushing in multilang sites: When using strategy: single with multiple languages, each language iteration was dropping and recreating the same physical index, causing only the last language's content to survive a --flush reindex. Flush now runs only once per physical index.
   • Fixed orphan detection deleting cross-language documents: With strategy: single, orphan detection ran per-language iteration, incorrectly treating documents from other languages as orphans and deleting them. Orphan detection now accumulates across all language iterations and runs once at the end.
   • Fixed smart indexing snapshot for single-strategy multilang: The snapshot used for fingerprint comparison was being overwritten by each language iteration, breaking skip-unchanged detection for all but the last language. The snapshot is now loaded once and shared across iterations.
   • Fixed admin dashboard showing inflated index count: With strategy: single and 12 languages, the dashboard reported "12 TOTAL INDEXES" and "12 ACTIVE INDEXES" even though there was only 1 physical index. Index counting now deduplicates by resolved index name.
   • Fixed language filter not working in admin Index & Search: The language dropdown in the admin browser/search had no effect for strategy: single indexes because the selected language was discarded during index resolution. The language is now preserved and passed as a query filter.
   • Fixed admin browser catalog fetching duplicate data: The getBrowserCatalog endpoint fetched index info, breakdown, and stats 12 times for the same physical index in a multilang single-strategy setup. It now fetches once per physical index.
   • Fixed hyphenated language codes breaking per_language strategy: Language codes containing hyphens (e.g. ca-valencia, zh-cn, pt-br) failed SQLite table name validation when used in index names. Hyphens are now sanitized to underscores in resolved index names.
   • Fixed Index & Search table layout: Columns were squished to the right when document IDs or routes were long. Table now uses fixed column widths so all columns remain visible.