Model catalogue · OpenAI
OpenAI Embedding
OpenAI's embedding model turns text into numerical vectors so that a search works on meaning rather than word matching - the substructure of file search: it produces no text and makes no assessment, but decides which passage of a file comes close to a question. Via Azure the chain under section 203 of the German Criminal Code is closed: processing stays inside the EU, storage at rest in the chosen Azure geography, and the confidentiality line signed there covers the models sold by Azure. What has to be regulated separately is the vector store: it is derived from client data and belongs to file-keeping.
The confidentiality agreement under section 203 of the German Criminal Code is signed. We set the route up without storage at the provider (zero retention) once the provider has committed to it - at Microsoft through the exception from abuse monitoring, which usually takes months. Until then, storage and review by the provider are not ruled out.
What it is good for
- Foundation for search and question-answering over your own file corpus, without a language model having to see the whole corpus
- Three tiers, two of them additionally evidenced as a regional EU deployment
- Same contractual line as the other models sold by Azure
Where it stops
- Produces no answers - without a downstream language model it stops at a list of hits
- The vector store is derived from client data and belongs to file-keeping: location, access protection and deletion have to be regulated separately
- Switching tiers forces a recomputation of the entire corpus - vectors from different tiers are not comparable
Status through us
The same model, procured differently, is a different legal position. The status above refers to procurement through us; processing takes place in a third country.
Which route carries it, with which instrument and under which condition, we set out in full to the firm before it commissions us - section 203(4) requires that the chain can be checked, not that it is published. The full matrix, with the sources behind every route, can be requested in the German catalogue after confirming your e-mail address, and we go through it in the conversation.
Is OpenAI Embedding the right fit?
We go through your tasks and the route that carries them - and say plainly where it does not.
Arrange a conversation