WebAI Chat

AI-powered chatbot that answers website questions using a local SLM (Small Language Model). Deploy on your customer's server with full privacy.

Features

• Setup Wizard: Browser-based GUI installer (webaichat-wizard) with step-by-step configuration
• Local SLM: Runs Qwen2.5-3B via Ollama (4GB VRAM)
• Cloud AI: Support for OpenAI, Anthropic, Azure, Google, Groq, and Together via LiteLLM
• Hybrid Mode: Local model first, cloud fallback
• RAG: Answers from your website content, not general knowledge
• Web Crawler: On-demand website crawling with configurable depth
• Document Upload: Upload PDF, TXT, or MD files to add to the knowledge base
• Batch Upload: Upload multiple text files at once
• CLI Ingest: Batch ingest from a folder via webaichat ingest
• Widget: Floating chat widget embeddable on any site with customizable appearance
• Admin Panel: Manage crawls, uploads, conversations, and system health
• Secure Login: Cookie-based session authentication for admin panel
• Config Hot-Reload: Settings update without server restart
• Streaming Responses: Server-sent events (SSE) for real-time chat
• Source Citations: Show source documents in responses
• Deployment Files: Auto-generate Docker, systemd, Heroku, Railway, Render configs
• Website Snippets: Generate embed code for any platform (HTML, WordPress, React, Vue)
• 100% Open Source: MIT licensed, no cloud dependencies
• Privacy-First: All data stays on customer's server

Screenshots

Chat Widget & Interface

Chat Widget	Chat Streaming	Chat Response

Chat with Citations	Login Page	Admin Panel

Admin Features

Conversations	Crawl Page	Settings

Widget Configuration

Quick Start

Prerequisites

• Python 3.10+
• Ollama installed: https://ollama.ai
• 4GB VRAM GPU (optional, for local SLM)

Installation

cd WebAI-Chat
python -m venv venv
venv\Scripts\activate          # Windows
source venv/bin/activate       # Linux/macOS
pip install -r requirements.txt
pip install -e .

Setup Ollama

# Install Ollama from https://ollama.ai
# Pull the language model
ollama pull qwen2.5:3b
# Pull the embedding model (uses SentenceTransformer, not Ollama)
# Embedding models are downloaded automatically on first use

Configure

cp config.yaml.example config.yaml
# Edit config.yaml with your settings

Run

# Start the server
webaichat serve

# Or with custom host/port
webaichat serve --host 0.0.0.0 --port 9000

Crawl a Website

# Via CLI
webaichat crawl https://example.com

# Via API
curl -X POST http://localhost:9000/api/crawl \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com"}'

Upload Documents

# Via API (single file)
curl -X POST http://localhost:9000/api/upload \
  -H "Content-Type: multipart/form-data" \
  -F "file=@document.pdf"

# Via CLI (batch from directory)
webaichat ingest path/to/folder --source "my-docs"

Embed the Widget

Add to any website:

<script>
  window.WEBAI_CHAT_CONFIG = { apiBase: "https://webaichat.yourserver.com/" };
</script>
<script src="https://webaichat.yourserver.com/static/widget.js"></script>

Admin Panel Authentication

The admin panel supports secure cookie-based session authentication for managing the chatbot. Enable it by setting credentials in config.yaml:

server:
  host: "0.0.0.0"
  port: 9000
  username: "admin"
  password: "your-secure-password"

How It Works

• When username and password are set, visiting /admin redirects to /login
• After successful login, a secure HTTP-only session cookie is set (8-hour timeout)
• A logout button is available in the admin panel sidebar
• When credentials are empty, admin access remains open (no authentication)
• Configuration changes take effect immediately without server restart

Authentication Endpoints

Method	Endpoint	Description
GET	/login	Login page
POST	/api/admin/login	Authenticate and create session
POST	/api/admin/logout	Clear session and logout
GET	/api/admin/check-auth	Check current auth status

Security Notes

• Passwords are hashed using bcrypt directly (not passlib, for compatibility)
• Session tokens are generated using secrets.token_urlsafe(32)
• Sessions expire after 8 hours automatically
• Cookies are HTTP-only to prevent XSS access
• The chat API (/api/chat) is separate from admin authentication

CLI Commands

Setup Wizard (GUI Installer)

webaichat-wizard
# or
python -m wizard

Opens a browser-based setup wizard at http://localhost:8080 that guides you through:

1. System check (Python, Ollama, GPU detection)
2. Deployment mode selection (local/cloud/hybrid)
3. Server configuration (host, port, authentication)
4. AI model setup (Ollama model detection, cloud API key input)
5. Chat settings (system prompt, tokens, retrieval)
6. Crawler settings
7. Widget customization
8. Deployment target selection (Docker, systemd, Heroku, Railway, Render)
9. Website injection snippet generation
10. One-click install (dependencies, config.yaml, deployment files)

Serve

webaichat serve
webaichat serve --host 0.0.0.0 --port 9000

Crawl

webaichat crawl https://example.com
webaichat crawl https://example.com --skip /login /admin /contact

Ingest (Batch Upload)

webaichat ingest path/to/folder --source "my-knowledge-base"

Model Management

webaichat model pull
webaichat model check
webaichat model info
webaichat model pull --model llama3.2:3b

API Endpoints

Chat

Method	Endpoint	Description
POST	/api/chat	Send a chat message (returns full response)
POST	/api/chat/stream	Streaming chat response (SSE)
GET	/api/sessions/{id}/messages	Get conversation history

Crawl & Upload

Method	Endpoint	Description
POST	/api/crawl	Crawl a website
POST	/api/upload	Upload a document (.pdf, .txt, .md)

Admin Panel (requires auth)

Method	Endpoint	Description
GET	/admin	Admin panel
GET	/login	Login page
POST	/api/admin/login	Authenticate and create session
POST	/api/admin/logout	Clear session and logout
GET	/api/admin/check-auth	Check current auth status
GET	/api/admin/health	System health
GET	/api/admin/stats	Conversation stats
GET	/api/admin/conversations	List conversations
GET	/api/admin/conversations/{id}	Get conversation details
DELETE	/api/admin/conversations/{id}	Delete conversation
GET	/api/admin/export	Export conversations (JSON/CSV)
POST	/api/admin/clear-crawl	Clear all crawl and upload data
GET	/api/admin/documents	List documents
GET	/api/admin/open-folder	Open data folder in file explorer
GET	/api/admin/model-info	List Ollama models
GET	/api/admin/mode	Get deployment mode
POST	/api/admin/mode	Update deployment mode
GET	/api/admin/settings	Get settings
POST	/api/admin/settings	Update settings
POST	/api/admin/settings/reset	Reset to defaults
GET	/api/admin/config/raw	Get raw config YAML
POST	/api/admin/config/raw	Save raw config YAML
GET	/api/admin/source-stats	Indexed chunk counts by source

Widget

Method	Endpoint	Description
GET	/api/widget/config	Widget configuration
POST	/api/widget/config	Update widget settings
GET	/api/widget/health	Widget health check
GET	/config.yaml	Serve config file

Configuration

See config.yaml.example for all settings:

• server: Host, port, admin username/password
• chat: Deployment mode (local/cloud/hybrid), system prompt, max tokens, top-k retrieval
• ollama: Ollama URL, language model name, embedding model
• cloud: Cloud provider (openai/anthropic/azure/google/groq/together), model, API key, base URL
• crawler: Rate limit, URL skip patterns
• widget: Position, color, logo, greeting, theme, quick replies, emoji picker, source citations, and more
• logging: Log level

Data Storage

• Crawled files: data/crawled/
• Uploaded documents: data/docs/
• Vector store: In-memory (ChromaDB) — cleared on server restart, re-indexed from data/docs/
• Conversations: data/webaichat.db (SQLite)
• Logs: data/logs/

Note: The vector store uses an in-memory ChromaDB client to avoid Windows file locking issues. Data is re-indexed on each server restart from files in data/docs/. Embedding models are cached in .cache/huggingface/.

Deployment

Bare Metal

pip install -r requirements.txt
pip install -e .
cp config.yaml.example config.yaml
# Edit config.yaml
webaichat serve

With systemd (Linux)

sudo tee /etc/systemd/system/webaichat.service << EOF
[Unit]
Description=WebAI Chat Server
After=network.target

[Service]
Type=simple
User=www-data
WorkingDirectory=/opt/WebAI-Chat
ExecStart=/opt/WebAI-Chat/venv/bin/python -m webaichat serve
Restart=always

[Install]
WantedBy=multi-user.target
EOF

sudo systemctl enable webaichat
sudo systemctl start webaichat

Troubleshooting

bcrypt Error

If you see passlib bcrypt errors, this project uses direct bcrypt library instead. Make sure you have bcrypt installed:

pip install bcrypt

Ollama Connection Refused

# Check if Ollama is running
ollama list

# Start Ollama if not running
ollama serve

Model Not Found

# Pull the required model
ollama pull qwen2.5:3b

# Verify it's installed
ollama list

Port Already in Use

# Find what's using the port (Windows)
netstat -ano | findstr :9000

# Change the port in config.yaml
# server.port = 8080

Hardware Requirements

• CPU: Any modern CPU
• RAM: 8GB minimum
• GPU: 4GB VRAM (optional, for local SLM)
• Storage: 2GB per model + data

Architecture

┌─────────────────────────────────────────────┐
│           Customer Web Server                 │
│                                              │
│  ┌──────────────┐    ┌────────────────────┐  │
│  │   Web GUI     │    │   Backend API       │  │
│  │  (Widget)     │◄──►│  (FastAPI)          │  │
│  └──────────────┘    └──────────┬─────────┘  │
│                                 │              │
│              ┌──────────────────┼──────────┐  │
│              ▼                  ▼           ▼  │
│       ┌──────────┐    ┌────────────┐ ┌───────┐│
│       │ ChromaDB │    │   Ollama   │ │Crawler││
│       │(in-memory)│   │ (Qwen2.5)  │ │       ││
│       └──────────┘    └────────────┘ └───────┘│
│                                 │              │
│                          ┌──────┴──────┐       │
│                          │  SQLite     │       │
│                          │(conversations)      │
│                          └─────────────┘       │
└─────────────────────────────────────────────┘

License

MIT License - See LICENSE file