> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getdatagate.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Core Concepts

> Datasets, connectors, pricing, and how queries work

## Datasets

A **dataset** is a collection of vector-indexed content published by a seller. Each dataset has:

* **Name and description** — what the data contains
* **Price per chunk** — cost per result returned (e.g., `$0.001`)
* **Visibility** — `public` (anyone can query) or `private` (requires seller-granted access)
* **Metadata schema** — describes what fields are available for filtering
* **Embedding model** — which model was used to vectorize the data (e.g., `text-embedding-3-small`)

## Connectors

Connectors are seller-side credentials linking a dataset to a vector database. Supported providers:

| Provider              | Auth                |
| --------------------- | ------------------- |
| Pinecone              | API key             |
| Qdrant                | API key             |
| Weaviate              | API key             |
| Milvus (Zilliz Cloud) | Username + password |

As a buyer, you never interact with connectors directly. The seller configures them, and Datagate routes queries to the right vector DB.

## How Queries Work

When you call `query()`:

1. **Access check** — Datagate verifies you can access the requested datasets (public = always, private = need access grant)
2. **Balance check** — estimates max cost (topK × price\_per\_chunk per dataset) and checks your balance
3. **Embedding** — converts your text query into vector(s) using each dataset's embedding model
4. **Fan-out** — searches each dataset's vector DB in parallel
5. **Merge & rank** — combines results across datasets, ranked by relevance score
6. **Billing** — charges per chunk actually returned (not estimated cost)

### Single-model vs multi-model

* **Same embedding model** across all datasets → embed once, merge by raw similarity score
* **Different embedding models** → embed per-model in parallel, merge using Reciprocal Rank Fusion (RRF) since raw scores from different models aren't comparable

You don't need to handle this — it's automatic. Just send text and dataset IDs.

## Pricing

* **Per-chunk billing**: you pay `price_per_chunk` for each result returned, plus a small platform fee
* **Deposits**: add funds via Stripe Checkout
* **Auto-deposit**: configure a threshold — when your balance drops below it, auto-refill from your saved payment method