Agentic AI Chatbot Builder

• Live Demo
• Screenshots
  • Bot management
  • Bot customization
  • Knowledge sources
  • Conversation history
  • Chat widget
  • Visual workflow builder
  • AI-assisted drafting
  • Run history and live tracing
  • Versions and releases
  • API connectors
• Features
  • Source-grounded knowledge
  • Visual agentic workflows
  • API connectors
  • Chat widget
  • Channel integrations
  • Production tooling
• Requirements
• Installation
• Configuration
• Usage
  • Fastest path to a working bot
  • Navigation customization
• Embedding the Chat Widget
  • Option 1 — Blade component (Laravel views)
  • Option 2 — Script tag embed (any site)
  • Option 3 — NPM package
• Workflow Node Types
• Importable Example Workflows
• Documentation
• Support
• License

Build, manage, and embed production-ready AI assistants directly in your Filament panel — from a simple grounded Q&A bot to guided agentic workflows with branching logic, AI nodes, backend actions, external API calls, and package-owned Telegram/Slack channels.

#Live Demo

Try every feature before you buy:

filament-agentic-chatbot.heinerdevelops.tech

Click Enter Demo on the login page. Pre-configured bots, ingested documentation sources, sample workflows, and a live chat widget are all ready.

#Screenshots

#Bot management

Manage multiple assistants from a Filament-native control plane. Each bot has its own model, prompt, retrieval settings, and analytics.

#Bot customization

Tune widget style, copy, quick prompts, area overrides, and live preview without leaving the bot editor.

#Knowledge sources

Add URL, file, raw-text, or API-fed JSON sources. Track ingestion status and refresh stale sources directly from the panel.

#Conversation history

Inspect conversation transcripts without leaving Filament. Review what users asked, what was retrieved, and which citations backed each answer.

#Chat widget

Show the live embedded chat experience up close, including the branded header, structured replies, and quick prompts.

#Visual workflow builder

Use the plugin feedback collector as a concrete example: the node library stays on the left, the canvas sits in the middle, and inline settings remain visible on the right while you build.

#AI-assisted drafting

Describe the flow you need in natural language, generate a draft workflow, then refine and publish.

#Run history and live tracing

Inspect every execution with completed-path highlighting, step traces, variables, halt reasons, and timestamps.

#Versions and releases

Publish versioned releases with notes and roll back to any prior version in seconds.

#API connectors

Define reusable external API connection profiles and reference them across workflow nodes or API-fed knowledge sources.

#Features

#Source-grounded knowledge

• Multi-bot management — unlimited bots, each with its own identity, model, prompt, retrieval config, and access controls
• Source types — URL page, PDF / file upload, plain-text snippet, API-fed JSON records
• Queue-based ingestion — async chunking, embedding, and upsert with per-source status tracking
• API knowledge sources — sync records from authenticated GET JSON endpoints through reusable API Connectors
• Vector backends — PostgreSQL + pgvector (recommended) or ChromaDB; configurable dimensions
• Provider support — Gemini, OpenAI, Anthropic, xAI, OpenRouter, DeepSeek, Groq, Mistral, Ollama, and Azure OpenAI for chat; Gemini, OpenAI, OpenRouter, Mistral, Ollama, Azure OpenAI, Cohere, Jina AI, and Voyage AI for embeddings
• Streaming chat with optional grounded retrieval, inline source citations, and citation coverage metrics
• Conversation history — full transcripts with per-message retrieval context and citation links
• Per-bot analytics — conversation count, message volume, source health score, citation coverage chart

#Visual agentic workflows

• Broad node catalog: Trigger, Send Message, Collect Input, AI Agent, Knowledge Base, Answer, Switch/Router, Condition, Action, HTTP Request, API Connector, Memory Read/Write, Delay, Loop, Sub-Workflow, and more
• Multi-branch canvas powered by Vue Flow — drag, connect, and organize nodes freely
• AI Agent node — configurable Provider, Model, and System Prompt per node within the same workflow
• Knowledge Base node — inline source retrieval mid-flow, configurable result count
• Switch/Router node — N branches + default; route by intent, field value, or AI classification
• HTTP Request node — call external APIs with variable interpolation in body and headers
• Scoped workflow memory — keep normal chat state in the conversation scope and one-run scratch state in workflow-run scope
• Variable interpolation — {{variable_name}} syntax in system prompts, messages, and node configs
• AI Draft — generate a complete workflow graph from a natural-language description in seconds
• Run history with step-level traces, current node, variable snapshot, halt reason, and timestamps
• Versioned releases with release notes; rollback to any prior version without re-editing
• Import / Export — share workflows as JSON between environments

#API connectors

• Reusable connection profiles — URL, auth method (Bearer, API Key, Basic Auth, Custom Header), default headers, SSL, and timeout
• Referenced by name in workflow API nodes and API knowledge sources; update credentials in one place

#Chat widget

• Blade component for Laravel views — authenticated or guest access
• JavaScript loader — single <script> tag embed for any external site (no npm required)
• NPM package for React, Vue, or any bundled frontend
• Signed widget tokens for access-controlled deployments (RAG_WIDGET_SIGNING_ENABLED)
• Mobile-polished — responsive layout, keyboard-safe positioning, iOS safe-area support
• Customizable — colors, placeholder text, greeting, launcher position, and z-index

#Channel integrations

• Telegram and Slack drivers — provider webhooks run through the same bot, workflow, conversation, usage, and budget runtime as the widget
• Channel admin resource — manage credentials, webhook URLs, linked Bot Access Tokens, diagnostics, test sends, and Telegram setup actions from Filament
• Text-first rich rendering — buttons, cards, sources, and images degrade cleanly to channel-safe text, with optional native Telegram buttons and Slack Block Kit
• Production delivery safeguards — webhook verification, inbound event deduplication, retry-aware outbound delivery, long-message splitting, raw payload redaction, and queue-based processing

#Production tooling

• php artisan filament-agentic-chatbot:doctor — pre-flight checks for config, DB, vector store, queue worker, and AI provider
• Data retention policy — scheduled pruning of old conversations with configurable age thresholds
• GDPR / privacy endpoints — export and delete conversation data per user on demand
• Queue isolation — ingestion jobs run on a configurable queue separate from your default queue
• Scheduled API source sync — periodically re-ingest API-backed knowledge sources with safety limits

#Requirements

• PHP 8.3+
• Laravel 12 or 13
• Filament 5.2+
• PostgreSQL with pgvector extension (recommended) or a running ChromaDB instance
• AI provider API key such as GEMINI_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY, XAI_API_KEY, OPENROUTER_API_KEY, DEEPSEEK_API_KEY, GROQ_API_KEY, MISTRAL_API_KEY, or AZURE_OPENAI_API_KEY
• A Laravel queue worker

#Installation

composer require heiner/filament-agentic-chatbot
php artisan vendor:publish --tag=filament-agentic-chatbot-config
php artisan migrate
php artisan queue:work

For the current public release line, install v0.12.0 or newer:

composer require heiner/filament-agentic-chatbot:^0.12.0

If you are testing the post-v0.12.0 AgentGraph SDK refactor branch before the SDK is published to your Composer repository, add the SDK repository to the host app root composer config before requiring the plugin:

composer config repositories.agent-graph vcs https://github.com/heinergiehl/agent-graph.git
composer config repositories.filament-agentic-chatbot vcs https://github.com/heinergiehl/filament-agentic-chatbot.git
composer require heiner/filament-agentic-chatbot:'*@dev'

Register the plugin in your panel provider:

use Heiner\FilamentAgenticChatbot\FilamentAgenticChatbotPlugin;

public function panel(Panel $panel): Panel
{
    return $panel
        ->plugins([
            FilamentAgenticChatbotPlugin::make(),
        ]);
}

Include the plugin views in your custom Filament theme:

/* resources/css/filament/admin/theme.css */
@source '../../../../vendor/heiner/filament-agentic-chatbot/resources';

npm run build && php artisan filament:upgrade

#Configuration

.env reference — all values have sensible defaults, override only what you need:

# Vector backend
RAG_VECTOR_BACKEND=pgvector          # pgvector | chromadb

# Dedicated DB connection for vector data (pgvector only)
RAG_DB_CONNECTION=rag_pgsql
RAG_DB_HOST=127.0.0.1
RAG_DB_PORT=5432
RAG_DB_DATABASE=filament_agentic_chatbot
RAG_DB_USERNAME=postgres
RAG_DB_PASSWORD=secret

# AI provider and models
RAG_CHAT_PROVIDER=gemini             # gemini | openai | anthropic | xai | openrouter | deepseek | groq | mistral | ollama | azure
RAG_CHAT_MODEL=gemini-2.5-flash-lite
RAG_EMBEDDING_PROVIDER=gemini
RAG_EMBEDDING_MODEL=gemini-embedding-001
RAG_VECTOR_DIMENSIONS=1536           # must match embedding model output

# Widget authentication
RAG_WIDGET_SIGNING_ENABLED=true
RAG_WIDGET_SIGNING_KEY=replace-with-a-long-random-secret

# Provider API key
GEMINI_API_KEY=your-key-here

Verify your configuration at any time:

php artisan filament-agentic-chatbot:doctor

#Usage

#Fastest path to a working bot

1. Create a bot in Chatbots → Bots
2. Add knowledge sources in Knowledge → Sources using text, file, URL, or API source types
3. Wait for ingestion to complete
4. Test from the bot edit screen using the Test Widget button
5. Embed the widget (see below)
6. Add workflows only when you need guided multi-step behavior

#Navigation customization

FilamentAgenticChatbotPlugin::make()
    ->navigationGroup('AI')   // move all resources under a custom nav group
    ->navigationSort(80)      // control position in the sidebar

#Embedding the Chat Widget

#Option 1 — Blade component (Laravel views)

<x-filament-agentic-chatbot::widget
    bot-id="{{ $bot->id }}"
    :token="$widgetToken"
/>

Generate a signed token (when RAG_WIDGET_SIGNING_ENABLED=true):

use Heiner\FilamentAgenticChatbot\Support\WidgetToken;

$token = WidgetToken::for(botId: $bot->id, userId: auth()->id());

#Option 2 — Script tag embed (any site)

<script
  src="https://your-app.com/filament-agentic-chatbot/widget"
  data-bot="YOUR_BOT_PUBLIC_ID"
  data-token="YOUR_SIGNED_TOKEN"
  defer
></script>

The legacy /filament-agentic-chatbot/widget.js script URL remains supported for existing snippets.

#Option 3 — NPM package

npm install @heiner/filament-agentic-chatbot-widget

import { ChatWidget } from "@heiner/filament-agentic-chatbot-widget";

<ChatWidget botId="YOUR_BOT_ID" token="YOUR_SIGNED_TOKEN" />;

#Workflow Node Types

#Importable Example Workflows

Fifteen ready-to-import examples covering common agentic patterns:

Browse workflow examples

Import any JSON file via the workflow editor's Import button.

#Documentation

• Product overview
• Quick start guide
• Core concepts
• Agent runtime architecture
• Agentic workflows — all node types explained
• Sources and ingestion
• Bots
• Database and breaking changes
• API connectors
• Channel integrations
• API source roadmap
• Chat widget embedding
• Conversations and messages
• Workflow JSON schema reference
• Workflow prompt templates
• Operations and production readiness
• Security and privacy
• Data retention policy
• How it differs from Filament RAG
• Release notes v0.12.0
• AgentGraph SDK refactor notes
• Known limitations

#Support

Open an issue in the public issue tracker. Response target: 2 business days.

See the full Support Policy.

#License

This is a commercial plugin. A license is required for production use. Source code distribution is not permitted.

See the full Refund and License terms.