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

# Find the information that matters

> Retrieval connects a user's current request with relevant information held in memory and configured knowledge sources. Its purpose is to help your application find useful evidence before generating a  response.

## From query to results

## Retrieval strategies

> **Semantic search**
>
> Find information based on similarity in meaning, rather than requiring exact keyword matches. Helps when a query uses different words from the original stored information.

> **Keyword search**
>
> Match explicit terms, identifiers, names, or phrases. Valuable for exact facts and technical references.

> **Hybrid retrieval**
>
> Combine semantic and lexical signals. Balances conceptual similarity with exact-term matching.

> **Reranking**
>
> Reorder an initial candidate set using a more focused relevance model. Can improve result ordering but adds computation and may increase latency.

## Retrieval controls

* Search scope and authorization.
* Candidate limits and result count.
* Similarity thresholds, where supported.
* Metadata filters.
* Ranking and tie-breaking behavior.
* Freshness and update behavior.
* Timeouts, retries, and error handling.

## Evaluate retrieval quality

| Metric | What it measures |
| :- | :- |
| Recall | How often the relevant information appears in the retrieved results. |
| Precision | How much of the retrieved information is relevant. |
| Ranking quality | Whether the most useful results appear near the top. |
| Latency | How long retrieval takes end-to-end. |
| Coverage | Whether important sources and data types are searchable. |
| Failure rate | How often requests fail or return no useful results. |

## Troubleshooting


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.