AI Functions

embed(text)

Generate an embedding vector for a string — the write-side counterpart to Model.similar, which embeds the query but not the documents you store.

Parameters

Returns

class Article < Model
  vector_index "embedding", dimension: 1536, metric: "cosine"

  before_save fn() {
    this.embedding = embed(this.title + "\n" + this.body)
  }
end

embed_batch(texts)

Embed many texts in a single request, returned in input order. Use it to back-fill embeddings over an existing collection instead of one call per row.

Parameters

Returns

articles = Article.where({ "embedding": null }).all
vectors  = embed_batch(articles.map(fn(a) a.title))
articles.each_with_index(fn(article, i) {
  article.embedding = vectors[i]
  article.save()
})

llm_generate(system, user)

Chat completion via an OpenAI-compatible chat/completions endpoint.

Parameters

Returns

summary = llm_generate(
  "You summarize support tickets in one sentence.",
  ticket.body
)

Streaming: inside an sse block, out.llm_stream(system, user) streams the completion token-by-token to the browser and returns the full answer. For an answer grounded in your own data, retrieve context first with Model.rag (embed → ANN → generate, returning { answer, sources }).

rerank(query, rows[, { field:, limit: }])

Reorder an array of already-retrieved records by how many query tokens each one's text contains — most relevant first. Pure and offline (no LLM, no server round-trip), so it's a cheap second pass after similar / hybrid / graph_rag to bias the order toward a phrase.

Parameters

Returns

rows = Article.similar("vector databases", "embedding", 20)
top5 = rerank("hnsw index tuning", rows, { "field": "content", "limit": 5 })

For LLM-based reranking or a stored retrieve→rerank pipeline, drop to raw SDBQL: RERANK(query, docs, { mode: "llm" }) and RAG_PIPELINE(name, @vec) via db.query.