---
title: "Retrieval architecture"
manual: "AI Assistant"
version: "main"
source: "Developers/Architecture/RetrievalArchitecture.rst"
rendered: "2026-10-03T14:54:45+00:00"
---

# Retrieval architecture {#retrieval-architecture}

Retrieval links the indexing and assistant domains.

## Data flow {#data-flow}

```mermaid
flowchart LR
    Q[Current query] --> R[RetrieverProcessor]
    R --> Emb[Embedding request]
    Emb --> AI[AiConnector]
    AI --> Search[VectorSearchRequest]
    Search --> VS[VectorStoreConnector]
    VS --> Docs[RetrievedDocument list]
    Docs --> State[RetrievalState]
    State --> Builder[AnswerContextBuilder]
    Builder --> Prompt[Retrieved Context]
```

## Retrieved documents {#retrieved-documents}

A retrieved document should carry:

-   text;
-   score;
-   title;
-   URL;
-   source type;
-   source identifier;
-   original payload metadata.

## Answer context builder {#answer-context-builder}

The answer context builder formats selected metadata and text into a prompt
section. It applies step limits and metadata field selection.