RAG API that keeps your documents private

Retrieval-Augmented Generation (RAG) is how production AI systems ground a model in your own data. Most RAG stacks are assembled from 4-6 vendors (vector DB, embedder, retriever, reranker, LLM, observability). Plugsky's RAG API collapses that into a single managed service.

What is RAG-as-a-service?

Plugsky RAG API exposes three endpoints that work together:

• POST /v1/embeddings — turn text into vectors (Plugsky or OpenAI ada-style)
• POST /v1/rag/collections — create and manage vector collections
• POST /v1/rag/query — query a collection, get back ranked chunks with citations

You can also use the embeddings endpoint standalone for similarity, clustering, or hybrid search in your own pipeline.

Embeddings API

from openai import OpenAI
client = OpenAI(base_url="https://api.plugsky.com/v1", api_key="sk-live-...")

emb = client.embeddings.create(
    model="plugsky-embed-v1",
    input="Plugsky is an OpenAI-compatible AI platform."
)
print(len(emb.data[0].embedding), "dimensions")

Plugsky ships plugsky-embed-v1 (1536 dimensions, OpenAI ada-compatible) and plugsky-embed-large (3072 dimensions, higher quality).

Collections, documents, queries

# 1. Create a collection
collection = client.rag.collections.create(name="acme-handbook")

# 2. Upload documents
client.rag.documents.create(
    collection_id=collection.id,
    file=open("handbook.pdf", "rb"),
    metadata={"department": "engineering", "year": 2026}
)

# 3. Query
result = client.rag.query.create(
    collection_id=collection.id,
    query="What is our vacation policy?",
    top_k=5,
)
for chunk in result.chunks:
    print(f"[{chunk.score:.2f}] {chunk.text[:100]}...")
    print(f"   source: {chunk.source}")

Documents are chunked, embedded, and indexed automatically. Retrieval supports keyword, vector, and hybrid search with optional cross-encoder reranking.

Citations and grounding

Every RAG query returns ranked chunks with source attribution. You can pass the chunks to any Plugsky chat model and get an answer with inline citations — Plugsky's RAG + chat composition handles prompt assembly, citation formatting, and refusal when the corpus doesn't have an answer.

Private RAG and on-prem

On self-serve plans, your documents are stored in Plugsky's managed vector store with per-collection encryption at rest. They never train a model.

On Enterprise, you can deploy RAG inside your own VPC with customer-managed keys (BYOK), private endpoints, and audit logs. Use cases include legal contract review, internal knowledge bases, financial filings, and medical records.

Frequently asked questions