# Neural Search

The Neural Search configuration section controls how the neural search pipeline retrieves, ranks, and filters knowledge base content across your tenant. These settings apply to all bots and cannot be overridden at the bot or data source level. Access these settings at **Settings > Configuration > Neural Search**.

## Retrieval

### Neural Inference Per Retriever Results Limit

| **Type**    | Integer |
| ----------- | ------- |
| **Default** | 8       |

Controls how many candidate documents each retriever fetches during neural search before neural search merges and re-ranks them. Raise this value when relevant documents are being missed in search results; lower it to reduce latency when retrieval speed is a priority.

See also: [Neural Search Result Limit](#neural-search-result-limit)

### Neural Search Result Limit

| **Type**    | Integer |
| ----------- | ------- |
| **Default** | 10      |

No description available.

### Return top section of KB

| **Type**    | Checkbox |
| ----------- | -------- |
| **Default** | Disabled |

No description available.

## Entity filtering

### Neural Inference Retriever Intersection Filter List

| **Type**    | Text field     |
| ----------- | -------------- |
| **Default** | Not configured |

Specifies which retrievers apply intersection-based entity filtering during neural search. This field accepts a comma-separated list of internal retriever names. Contact Aisera support to configure this setting.

### Neural Inference Retriever Union Filter List

| **Type**    | Text field                    |
| ----------- | ----------------------------- |
| **Default** | `BMRetriever,EMRetriever_new` |

Specifies which retrievers apply union-based entity filtering during neural search. This field accepts a comma-separated list of internal retriever names. Contact Aisera support to configure this setting.

## Reranking and relevancy

### Enable Cross-Encoder

| **Type**    | Checkbox |
| ----------- | -------- |
| **Default** | Disabled |

Activates a reranking step that scores each retrieved result by comparing the full query against the document's title and subject, replacing the initial retrieval ranking. Enable this when embedding-based ranking produces results that are semantically related but not precisely ordered by relevance to the exact query phrasing.

{% hint style="info" %}
 The cross-encoder evaluates document title and subject only, not full document content. 
{% endhint %}

{% hint style="warning" %}
Enabling this adds inference latency proportional to the number of retrieved results. It also changes how relevance scores are scaled in the response, which may affect downstream systems that apply score thresholds.
{% endhint %}

### Tenant Relevancy Confidence Cutoff Score (Normalized 0-100)

| **Type**    | Decimal |
| ----------- | ------- |
| **Default** | 0       |

No description available.

### Per App Relevancy Confidence Cutoff Score (Normalized 0-100)

| **Type**    | Text field |
| ----------- | ---------- |
| **Default** | `[{}]`     |

No description available.

## Form intercept

### Filter by Product Information in Form Intercept

| **Type**    | Checkbox |
| ----------- | -------- |
| **Default** | Disabled |

When a **Form Intercept** channel submits a query that includes a product identifier, this setting restricts neural search results to knowledge base content tagged with that identifier. Use this when your tenant supports multiple products with separate knowledge base content and you want to prevent cross-product results.

{% hint style="info" %}
Tag your knowledge base documents with product information metadata matching the identifiers your **Form Intercept** channel sends.
{% endhint %}

See also: [Form Intercept Delimiter](/aisera-platform/channels/form-intercept.md)

### Form Intercept Delimiter

| **Type**    | Text field |
| ----------- | ---------- |
| **Default** | `\|\|`     |

Specifies the delimiter neural search uses to separate the search query from the product identifier in **Form Intercept** queries. This value must match the delimiter your **Form Intercept** channel uses when constructing queries that include product context.

{% hint style="info" %}
Neural search treats the delimiter as a literal string, not a regular expression.
{% endhint %}

See also: [Filter by Product Information in Form Intercept](#filter-by-product-information-in-form-intercept)


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.aisera.com/aisera-platform/tenant-setup/aisera-platform-configuration/tenant-configuration-settings/neural-search.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.