> For the complete documentation index, see [llms.txt](https://docs.hexabot.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.hexabot.ai/retrieval-augmented-generation.md).

# Retrieval-Augmented Generation

Retrieval-Augmented Generation, or RAG, lets an AI workflow retrieve relevant information before it generates an answer. In Hexabot, the source information normally comes from **Content** entries managed in the administration interface.

Starting with Hexabot 3.4.x, RAG is helper-based. Hexabot no longer imposes one indexing framework or one retrieval strategy. You can use:

* `fulltext-search`, the built-in lexical helper for SQLite and PostgreSQL;
* [`hexabot-helper-sqlite-vector`](https://hexabot.ai/extensions/6a6daf6d147efb889b884892), the installable semantic-search helper based on `sqlite-vec`;
* [`hexabot-helper-pgvector`](https://hexabot.ai/extensions/6a6dafd8147efb889b884994), the installable semantic-search helper for PostgreSQL;
* a custom helper backed by Weaviate, Pinecone, Elasticsearch, a graph-retrieval pipeline, a hybrid ranker, or another system.

The npm package names identify the installable extensions. After installation, select their runtime helper names—`sqlite-vector` or `pgvector`—under **Default RAG helper**.

{% hint style="info" %}
A RAG helper performs **retrieval**. It does not automatically ground every LLM call. Your workflow must pass the retrieved text to an AI generation action, or mount `retrieve_rag_content` as a tool on an AI Agent.
{% endhint %}

### How RAG works in Hexabot

```mermaid
flowchart LR
    A[Content entries] --> B[Canonical search text]
    B --> C{Default RAG helper}
    C --> D[fulltext-search]
    C --> E[sqlite-vector]
    C --> F[pgvector]
    C --> G[Custom helper]
    D --> H[retrieve_rag_content]
    E --> H
    F --> H
    G --> H
    H --> I[AI action or AI Agent]
    I --> J[Grounded response]
```

1. You create structured entries under **Content**.
2. Hexabot builds canonical search text from each entry's title and string-valued properties.
3. The helper selected in `global_settings.default_rag_helper` retrieves the most relevant entries or chunks.
4. The `retrieve_rag_content` action returns ranked hits and a combined `text` value.
5. An AI action or AI Agent uses that context to produce the final response.

### Core concepts

| Concept               | Meaning                                                                                      |
| --------------------- | -------------------------------------------------------------------------------------------- |
| Content corpus        | The Content entries that provide source material for retrieval.                              |
| Canonical search text | A text representation containing the entry title and its string-valued properties.           |
| RAG helper            | The strategy used to retrieve relevant content. One helper is selected as the default.       |
| Retrieval action      | `retrieve_rag_content`, which calls the selected helper from a workflow or as an agent tool. |
| Generator             | The AI action or AI Agent that turns the question and retrieved context into a response.     |
| Reindex               | A reconciliation operation that rebuilds or refreshes a helper's retrieval structures.       |

### What changed in Hexabot 3.4.x

The former RAG implementation was coupled to LlamaIndex and maintained separate lexical and embedding indexes. Hexabot 3.4.x replaces it with a smaller core contract and installable helpers.

The practical changes are:

* There is no RAG enable/disable switch. Retrieval is available through the selected helper.
* `fulltext-search` is built in and selected by default.
* Vector retrieval is optional and installed per database.
* The helper, not a per-action “mode,” determines how retrieval works.
* Configuration and availability failures are exposed by `retrieve_rag_content` through its `warning` output.
* Custom helpers can own their settings, storage, indexing, and ranking strategy.

### What Hexabot retrieves

The default corpus is the CMS Content table. For each entry, Hexabot includes:

* the entry title;
* properties whose stored value is a string, typically Text and Text Area fields.

Numbers, booleans, relation objects, and media objects are not converted to descriptive text automatically.

{% hint style="warning" %}
Uploading a PDF, image, or other file to the Media Library does not automatically parse it into RAG content. Extract the text into Content entries or implement an ingestion/custom-helper pipeline for document retrieval.
{% endhint %}

### Start here

* Follow [Quickstart](/retrieval-augmented-generation/quickstart.md): build a grounded answer workflow to use the built-in helper without an embedding provider.
* Read [Choose a retrieval helper](/retrieval-augmented-generation/choose-a-rag-helper.md) to decide between lexical, vector, hybrid, and external retrieval.
* Use [Tune and troubleshoot](/retrieval-augmented-generation/tune-and-troubleshoot.md) retrieval when results are missing or irrelevant.
* Existing deployments should read [Migrate from v3.3.x](/retrieval-augmented-generation/migrate-from-v3.3.x.md).


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## 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, and the optional `goal` query parameter:

```
GET https://docs.hexabot.ai/retrieval-augmented-generation.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

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.