Wikidata Embedding Project
The Wikidata Embedding Project is a hosted semantic-search service over Wikidata, the free and collaborative knowledge graph maintained by the Wikimedia movement. It is led by Wikimedia Deutschland and built with Jina.AI, which supplies a multilingual embedding model covering more than 100 languages with an 8192 token context, and DataStax, which provides the vector database that stores and serves the results. Nearly 120 million Wikidata items have been embedded, so each entity and its relationships can be retrieved by meaning rather than by exact label or SPARQL query.
You do not download this one. Access is through a hosted vector database and REST API at wd-vectordb.wmcloud.org, with interactive documentation at the /docs path, and there is a native Model Context Protocol endpoint so an LLM or agent can query it directly as a tool. A request returns the most semantically similar items along with structured context such as labels, translations and a cleared image, and the service offers reranking to sharpen the ordering of candidates.
For RAG and AI work this is a drop-in grounding layer on verified, openly licensed facts: you retrieve relevant Wikidata entities for a question and feed them to your model, or combine vector matches with the underlying graph for GraphRAG style reasoning. It is best for fact checking, entity disambiguation, classification and reference linking rather than long-form passage retrieval.
On licensing, Wikidata's structured data is released under CC0, placing it in the public domain, while textual content such as certain descriptions carries CC BY-SA, which obliges attribution and share-alike. Treat outputs accordingly and check Wikidata's licensing page before redistributing. The main caveat is operational rather than legal: this is a hosted service on Wikimedia Cloud, not a bulk file, so you depend on its uptime and its query-time model, unlike the self-hosted Cohere Wikipedia embedding sets we also list, where you load vectors into your own index.
Related sources
Cohere Wikipedia 22-12 Embeddings (per-language)
Wikipedia encoded with the Cohere multilingual-22-12 embedding model, with embeddings (numeric vectors that capture meaning) computed on each article's title plus text. Published per language, covering many including Arabic, German, English, Spanish, French, Hindi, Italian, Japanese, Korean, Simple English, and Chinese. You can stream it rather than download it in full, which matters given the size.
Cohere Wikipedia Multilingual Embeddings (2023-11)
The November 2023 Wikipedia dump across 300+ languages, split into passages and turned into embeddings (numeric vectors that capture meaning) with the Cohere Embed V3 multilingual model. Close to 250 million paragraph embeddings in total. Drop it into a vector database for semantic search over all of Wikipedia, or use it as a knowledge source for multilingual RAG, including cross-lingual search where a query in one language returns relevant results from another.
Google Satellite Embedding (AlphaEarth Foundations)
Global annual satellite embeddings produced by Google and Google DeepMind's AlphaEarth Foundations model. Every 10 metre pixel holds a 64 dimensional vector summarising a year of multi-sensor Earth observation, available yearly from 2017.
LAION-400M with CLIP Embeddings
400 million image-text pairs filtered from Common Crawl using CLIP, a model that scores how well an image matches a caption, and released alongside their pre-computed CLIP embeddings and kNN indices for fast similarity search. The ready-made embeddings make it unusually easy to work with. Note that it ships image URLs, not the images themselves, so some links have decayed.