RAG Chatbot

• Table of contents
• Live demo
• What you get
• Requirements
• Installation
• Configuration
  • 1) Provider credentials
  • 2) Vector database (pgvector recommended)
  • 3) Widget signing (recommended)
• Create your first bot
• Embed the widget
• Operations
  • Readiness check
  • Queue and retries
  • Rate limiting
• Security & privacy
• Multi-tenancy
• Support & policies

Build, manage, and embed production-ready RAG chatbots directly from your Filament panel.

[!IMPORTANT] PHP 8.4+ is required (this plugin depends on laravel/ai, which requires PHP ^8.4).

#Table of contents

• Live demo
• What you get
• Requirements
• Installation
• Configuration
• Create your first bot
• Embed the widget
• Operations
• Security & privacy
• Multi-tenancy
• Support & policies

#Live demo

• Public landing
• Admin login

#What you get

• Filament resources for Bots, Sources, and Conversations
• Queue-based ingestion (text, file, URL) with status visibility
• pgvector (recommended) or ChromaDB vector backends
• Streaming chat (SSE) with citations/sources
• Embeddable widget (script tag) + optional NPM loader
• Privacy workflows: export and delete chat history per session
• Operational readiness via php artisan filament-rag:doctor

#Requirements

• PHP 8.4+
• Laravel 12+
• Filament 5.2+
• Postgres + pgvector (recommended) or ChromaDB (optional)
• An AI provider key compatible with laravel/ai (Gemini/OpenAI/Anthropic/xAI)

#Installation

1. Purchase the plugin on Anystack and configure Composer auth (Anystack handles the private package access).
2. Install the package:

composer require heiner/filament-rag
php artisan vendor:publish --tag=filament-rag-config
php artisan migrate

3. Register the plugin in your panel provider:

use Heiner\FilamentRag\FilamentRagPlugin;

// …
->plugins([
    FilamentRagPlugin::make(),
])

4. Start your queue worker (ingestion runs on queues):

php artisan queue:work

#Configuration

#1) Provider credentials

Set the default chat + embedding providers/models via env (override per bot is supported inside Filament).

# Defaults (can be overridden per bot)
RAG_CHAT_PROVIDER=gemini
RAG_CHAT_MODEL=gemini-2.5-flash-lite
RAG_EMBEDDING_PROVIDER=gemini
RAG_EMBEDDING_MODEL=gemini-embedding-001

# Provider keys (examples)
GEMINI_API_KEY=your-key-here
OPENAI_API_KEY=your-key-here
ANTHROPIC_API_KEY=your-key-here
XAI_API_KEY=your-key-here

#2) Vector database (pgvector recommended)

This plugin can use a separate Postgres connection for RAG storage, even if your main app uses MySQL/SQLite.

RAG_VECTOR_BACKEND=pgvector
RAG_DB_CONNECTION=rag_pgsql
RAG_DB_HOST=127.0.0.1
RAG_DB_PORT=5432
RAG_DB_DATABASE=filament_rag
RAG_DB_USERNAME=postgres
RAG_DB_PASSWORD=secret

# Required for pgvector column sizing / indexing decisions
RAG_VECTOR_DIMENSIONS=1536

[!NOTE] For higher-dimensional embeddings (> 2000 dims), the plugin skips pgvector ANN index creation to avoid migration failures.

#3) Widget signing (recommended)

RAG_WIDGET_SIGNING_ENABLED=true
RAG_WIDGET_SIGNING_KEY=your-long-random-secret
RAG_WIDGET_SIGNING_TTL_MINUTES=43200

#Create your first bot

1. Open your Filament panel
2. Create a bot in RAG Bots
3. Add one or more sources in RAG Sources
4. Wait until ingestion is completed
5. Test retrieval/chat from the bot page

#Embed the widget

Use the Embed Snippet action on the bot edit page to generate the exact script tag for your bot:

<script
  src="https://your-app.com/filament-rag/widget.js"
  data-bot="YOUR_BOT_PUBLIC_ID"
  data-token="SIGNED_WIDGET_TOKEN"
  defer
></script>

#Operations

#Readiness check

Run:

php artisan filament-rag:doctor

#Queue and retries

• Ingestion is queue-driven; you must run a worker in production.
• You can isolate ingestion to its own queue via env:

RAG_INGESTION_QUEUE_CONNECTION=database
RAG_INGESTION_QUEUE=rag-ingestion

#Rate limiting

Defaults are configurable via env:

RAG_MAX_REQUESTS_PER_MINUTE=40
RAG_MAX_REQUESTS_PER_MINUTE_PER_IP=120

#Security & privacy

Built-in controls:

• Domain allowlist per bot
• Optional signed embed tokens (recommended)
• URL ingestion blocks localhost/private networks by default (SSRF protection)
• Export/delete endpoints for session history

Endpoints:

• Export: GET /api/filament-rag/chat/{botPublicId}/history/export?session_id=...
• Delete: DELETE /api/filament-rag/chat/{botPublicId}/history with JSON { "session_id": "..." }

#Multi-tenancy

Out of the box, this plugin stores all data in shared rag_* tables.

If you need tenant isolation, you have two realistic options:

1. Per-tenant database/schema for the RAG connection (advanced): set RAG_DB_* to a tenant-specific database/schema at runtime.
2. Tenant-scoped models/tables (advanced): fork/customize migrations and models to include tenant_id and enforce global scopes.

[!IMPORTANT] Tenancy requirements vary wildly between apps; if this is critical for you, validate the approach in a spike before production rollout.

#Support & policies

• Support: webdevislife2021@gmail.com
• Release notes (latest): https://github.com/heinergiehl/rag-filament-docs/blob/main/RELEASE_NOTES_v1.0.1.md
• Previous release notes: https://github.com/heinergiehl/rag-filament-docs/blob/main/RELEASE_NOTES_v1.0.0.md
• Support policy: https://github.com/heinergiehl/rag-filament-docs/blob/main/SUPPORT_POLICY.md
• Refund/license terms: https://github.com/heinergiehl/rag-filament-docs/blob/main/REFUND_AND_LICENSE.md
• Security & privacy: https://github.com/heinergiehl/rag-filament-docs/blob/main/SECURITY_AND_PRIVACY.md
• Data retention: https://github.com/heinergiehl/rag-filament-docs/blob/main/DATA_RETENTION_POLICY.md