Wiki Offline

A fully offline general-knowledge fallback for OVOS - "who was Charlie Chaplin", "what is the Eiffel Tower", "tell me about tomato". An offline alternative to ovos-skill-wikipedia/ovos-skill-ddg/ ovos-skill-wolfie for the topics covered by each language's own Wikipedia "Level 4 Vital Articles" equivalent - the subjects that language's Wikipedia community itself has curated as the most essential encyclopedia entries. No internet connection needed at runtime for the natively-supported languages. English, Spanish, and French are natively supported; any other language works too via on-demand translation - see Language support below.

• Usage
• What this is (and isn't)
• Language support
• A Common Query + Fallback skill, not a fixed intent
• Data sourcing and licensing
• Known limitations
• Install
• Development

Usage

"who was Charlie Chaplin"
"what is the Eiffel Tower"
"tell me about tomato"
"what do you know about photosynthesis"
"where is Mount Everest"
"quién fue Charlie Chaplin"       (Spanish)
"cuéntame sobre el Tomate"        (Spanish)
"qui était Charlie Chaplin"       (French)
"parle-moi de la Tomate"          (French)
"hvem var Charlie Chaplin"        (Danish, via translation - see below)
"wer war Charlie Chaplin"         (German, via translation - see below)

What this is (and isn't)

An encyclopedia lookup, not a reasoning engine. It resolves a single named entity from the question and speaks that entity's own short summary. It can answer "what is a tomato" but not "are tomatoes and potatoes related" - the second is a comparison across two entities, and there's no structured data being compared here, only free-text extracts (some relational questions get answered by luck, when one entity's own summary happens to mention the other - not by design). See DEVELOPMENT.md for the reasoning.

Language support

The natively-supported languages need nothing beyond pip install ovos-skill-wiki-offline - the data ships with the skill. For every other language (German and Danish included), this skill translates the question and answer on the fly, using whatever OVOS translation plugin is configured on the device - it doesn't ship or require one itself. Recommended: ovos-translate-plugin-nllb, which runs a local NLLB-200 model (no internet connection needed once the model is downloaded, unlike ovos-translate-plugin-server, which calls a remote server by default). See Install below for the setup steps. Translated answers take a few extra seconds (typically 2-5s) compared to a native-language answer - see DEVELOPMENT.md "Ad-hoc translation for unsupported languages" for measured performance and the full design reasoning.

A Common Query + Fallback skill, not a fixed intent

Every sibling skill in this project family (geography, geometry, convert, calculator) uses fixed Padatious intents as the primary path. This one doesn't - ~10,000 arbitrary proper nouns don't fit a bounded intent-slot vocabulary the way 194 countries or 24 geometry terms do. Instead this skill competes via Common Query (alongside Wikipedia/DDG/Wolfram, when the platform routes a question there) and catches whatever's left via a Fallback handler - the same dual pattern ovos-skill-wolfie already uses. See DEVELOPMENT.md for the full reasoning, including why this needed extending FallbackSkill specifically (not just adding the @fallback_handler decorator to a plain skill).

Data sourcing and licensing

Article titles and summaries both come directly from Wikipedia's own official APIs - the Level 4 Vital Articles category structure for the title list, and the REST summary API for the short spoken extracts. Wikipedia content is CC BY-SA 4.0; this skill bundles it offline the same way ovos-skill-wikipedia reads it aloud live. Full attribution and licensing details, plus why Level 4 was chosen over Kiwix/ZIM, DBpedia, or Wikipedia's own larger vital-article levels: CREDITS.md and DEVELOPMENT.md.

Known limitations

• Non-native languages need a translation plugin configured - see Language support and Install.
• The Spanish dataset is smaller (~6,200 topics vs ~10,000 for English/French) - 3 of Spanish Wikipedia's own 11 topic categories (biology/health, physics, social sciences) don't actually exist as pages despite the source list claiming full completion. See DEVELOPMENT.md "The Spanish gap" for the full story - this means Spanish can't currently answer biology, physics, or social-science questions.
• Single-entity lookup, not relational reasoning - see above and DEVELOPMENT.md.
• A snapshot, not a live mirror - each data/summaries_<lang>.json reflects Wikipedia as of whenever data/build_data.py was last run for that language, not the current live article.
• Scope is Level 4 (~10,000 topics for en-us/fr-fr) - Level 5 (50,000 articles) exists as a natural future extension (Level 4 is a strict subset), not built in v1.

Install

pip install ovos-skill-wiki-offline

That's enough for English, Spanish, and French. For any other language (German, Danish, or anything else), also install and configure a translation plugin - see Language support above for why this skill doesn't bundle one itself. Recommended, fully-local option:

pip install ovos-translate-plugin-nllb

Then add this to ~/.config/mycroft/mycroft.conf (or wherever your OVOS config lives):

{
  "language": {
    "translation_module": "ovos-translate-plugin-nllb",
    "ovos-translate-plugin-nllb": {
      "model": "nllb-200_600M_int8"
    }
  }
}

Restart ovos-core after installing/configuring. The model (~600MB) downloads automatically on first use and takes roughly 40 seconds to load the first time a translated answer is requested after each restart - after that it stays loaded for the rest of the session. Full plugin details, other model size options, and GPU configuration: ovos-translate-plugin-nllb on GitHub.

Any OTHER OVOS translation plugin works too, as long as language.translation_module points to it - this skill uses whatever's configured, generically, not specifically NLLB.

Development

See DEVELOPMENT.md.

Category

Information

Tags

#wikipedia #offline #encyclopedia #reference #commonquery #fallback