Kladde.Verlauf stores the text sections of the tenders in the container knowledge, each section with its text and a vector of 1,536 dimensions (F-03, F-06). Two kinds of questions hit it. "Who services the elevator?" means "maintenance of the lift systems" without a single word in common: a case for vector search. "CPV 50750000" means exactly that number: a case for full-text search. This lesson builds both into the container, the next one combines them.

The three index types with their dimension limits and VectorDistance are in "K3, Lesson 3.6.4" and are not repeated. The examples run with Microsoft.Azure.Cosmos 3.64.0 and Newtonsoft.Json 13.0.4, which the SDK requires. Not run against Azure: everything is compiled, and what runs without an account was run locally; no command went to a Cosmos DB service. This lesson changes none of the facts F-01 to F-08.

Account and order

Vector and full-text search are features of the account. According to Learn, you turn on vector search in the portal under "Features" or with the capability EnableNoSQLVectorSearch (up to about 15 minutes of waiting). For full-text search, Learn requires signing up for the feature "Full Text Search"; its name in Bicep or the CLI is not on the pages that were read (open). That belongs in a script with -WhatIf.

You set the policies when you create the container. According to Learn, the vector policy and the vector index are immutable: paths can be added or dropped, but their settings do not change. Vector search does not work on accounts with shared throughput.