Core Concepts
This page explains the main building blocks of Filament RAG and shows how they connect.
The Core Flow
Filament RAG gives you a Filament-native control plane for running RAG chatbots inside a Laravel app.
The normal lifecycle looks like this:
- Create a bot
- Add RAG sources
- Let the system ingest those sources into searchable chunks
- Retrieve the most relevant chunks at question time
- Answer through the chat widget or your own frontend
- Review conversations, quality, and operating health inside Filament
Concept Map
| Concept | What It Means | Learn More |
|---|---|---|
| Bot | A configured assistant with its own prompt, model, retrieval settings, access rules, and widget branding | Bots |
| RAG Source | A piece of knowledge you want the bot to use, such as text, a file, or a URL | RAG Sources |
| Document | The normalized stored version of a source after extraction | Ingestion and Retrieval |
| Chunk | A smaller searchable section of a document used for retrieval and citations | Ingestion and Retrieval |
| Ingestion | The pipeline that extracts, normalizes, chunks, embeds, and stores source content | Ingestion and Retrieval |
| Retrieval | The step where relevant chunks are selected for a user question | Ingestion and Retrieval |
| Widget | The embeddable chat UI for websites or product frontends | Chat Widget |
| Context Area | The access scope for a bot, such as public, member, or admin | Context Areas |
| Conversation | A stored chat session for one bot and one session identifier | Conversations and Messages |
| Message | An individual user or assistant entry inside a conversation | Conversations and Messages |
How The Pieces Fit Together
Bot
The bot is the central runtime definition. It decides:
- which model and provider answer questions
- which sources belong to it
- how strict retrieval should be
- which users or areas can access it
- how the widget looks and behaves
Sources, Documents, and Chunks
A source is the input. During ingestion, that source becomes a normalized document, and that document is split into chunks. Retrieval searches those chunks, not entire files or pages.
That matters because chunk quality directly affects answer quality, citations, and relevance.
Retrieval and Answers
When a user asks a question:
- the system embeds the query
- finds relevant chunks
- filters and formats those chunks as context
- sends that context to the model
- returns an answer, often with citations
Conversations and Widget Runtime
The widget is only the interface layer. The bot, sources, retrieval, access rules, and conversations live in your Laravel app and are managed from Filament.
A Good Mental Model
- The bot defines behavior.
- The sources define knowledge.
- Ingestion prepares that knowledge.
- Retrieval picks the right context.
- The widget delivers the user experience.
- Conversations give you history and auditability.