34.1 Text: embeddings and completions

// a completion with the configured model, or with a model from CompletionModelsByKey
string summary = await db.Services.AI.GetCompletionAsync("Summarise in one sentence: " + article.Body);
string careful = await db.Services.AI.GetCompletionAsync(prompt, modelKey: "strong");

// one vector per paragraph, in order
List<float[]> vectors = await db.Services.AI.GetEmbeddingsAsync(["first paragraph", "second paragraph"]);

To use the hosted Relatude AI service, set AISettings for the database (§12.1 lists every key and provider):

"AISettings": {
  "TypeName": "RelatudeServices",
  "CompletionModel": "fast",
  "CompletionModelsByKey": { "strong": "balanced" }
}

Model names. The hosted service has its own model keys: embedding-small and embedding-large for EmbeddingModel, fast and balanced for CompletionModel. Leave them empty for the service's default. CompletionModelsByKey maps your own keys onto them.
Semantic search computes embeddings by itself, so most applications never call GetEmbeddingsAsync.
Embeddings are cached on the server by text and model (AISettings.CacheType), so each text is paid for once. Completions are not cached.