> ## 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.

# Multi-Dataset Queries

> Query across multiple datasets with different embedding models

## Querying Multiple Datasets

Pass multiple dataset IDs to search across them in a single call:

```python theme={null}
response = await client.query(
    text="transformer architectures",
    dataset_ids=["papers-ds", "patents-ds", "news-ds"],
    top_k=20,
)
```

Datagate fans out to all datasets in parallel and merges the results.

## Single-Model vs Multi-Model

### Same embedding model

When all datasets use the same model (e.g., `text-embedding-3-small`):

* Query text is embedded **once**
* Results are ranked by **raw similarity score** (0–1, higher = more relevant)
* Scores are directly comparable across datasets

### Different embedding models

When datasets use different models (e.g., one uses `text-embedding-3-small`, another uses `embed-v4.0`):

* Query text is embedded **once per unique model** (in parallel)
* Results are ranked by **Reciprocal Rank Fusion (RRF)**
* RRF uses `1/(k + rank)` where rank is the result's position within its dataset
* This is fair across models because it only considers rank order, not raw scores

You don't need to handle this — it's automatic. The `embedding_model` field appears on results only in multi-model queries so you can see which model produced each result.

## Partial Failures

If one dataset's vector DB is unreachable, the other datasets still return results. Check `response.warnings` for details:

```python theme={null}
response = await client.query(
    text="search query",
    dataset_ids=["ds-1", "ds-2", "ds-3"],
)

if response.warnings:
    for warning in response.warnings:
        print(f"Warning: {warning}")

# Results from successful datasets are still returned
for result in response.results:
    print(f"[{result.score:.4f}] {result.dataset_id}: {result.metadata}")
```

## Pre-Computed Vectors

When using `vector` instead of `text`:

* **All datasets must use the same embedding model** — the server can't split a single vector across different models (returns 400 if they differ)
* Vector dimension must match the model's expected dimension

```python theme={null}
# Only works if both datasets use the same model
response = await client.query(
    vector=[0.012, -0.034, ...],
    dataset_ids=["ds-1", "ds-2"],
)
```