chatvat 0.2.6

The Universal RAG Chatbot Factory. Zero-dependency AI deployments.

ChatVat (The ChatBot🤖 Factory🏭)

The Universal RAG Chatbot Factory

---

🌟 The Vision

ChatVat is not just another chatbot script. It is a Manufacturing Plant for self-contained AI systems.

It solves the "It works on my machine" problem by adhering to a strict "Zero-Dependency" philosophy. ChatVat takes your raw data sources—websites, secured APIs, and documents—and fuses them with a production-grade RAG engine into a sealed Docker container. This "capsule" contains everything needed to run: the code, the database, the browser, and the API server.

You can deploy a ChatVat bot anywhere: from a MacBook Air to an air-gapped server in Antarctica, with nothing but Docker installed.

Core Philosophy

• Split Architecture: A lightweight CLI (~15MB) for management, and a heavy-duty Docker container for the AI engine. No more installing 3GB of CUDA drivers on your laptop just to run a build tool.
• Universal Connectivity: Acts as a generic MCP (Model Context Protocol) connector. Can ingest data from any API using custom headers and auth keys.

Disclaimer : We advise users to ensure they have authorized access to secured APIs and to follow the provider's guidelines when extracting data from sensitive APIs. Read the full disclaimer below for more details.

• Self-Healing: Built-in deduplication (Content Hashing), crash recovery, and "Ghost Entry" prevention.
• Production Parity: The bot you test locally is bit-for-bit identical to the bot you deploy, thanks to baked-in browser binaries.

---

⚡ Quick Start

1. Installation

Install the lightweight ChatVat CLI. (It installs in seconds and won't bloat your system).

pip install chatvat

2. Initialize the Assembly Line

Create a clean directory for your new bot and run the configuration wizard.

mkdir my-crypto-bot
cd my-crypto-bot
chatvat init

The wizard will guide you through:

• Naming your bot
• Setting up AI Brain (Groq Llama-3 + HuggingFace Embeddings)
• Connecting Data Sources (URLs, Secured APIs, or Local Files)
• Defining Deployment Ports

3. Build the Capsule

Compile your configuration and the ChatVat engine into a Docker Image.

chatvat build

What happens here? The CLI performs Source Injection: it copies the core engine code into a build context, injects your chatvat.config.json, and triggers a multi-stage Docker build. It optimizes the image by installing specific browser binaries (Chromium only) and purging build tools, keeping the final image lean.

4. Deploy Anywhere

Run your bot using standard Docker commands. Note the use of --ipc=host to prevent browser crashes on memory-heavy sites.

# Example: Running on Port 8000
docker run -d \
  -p 8000:8000 \
  --env-file .env \
  --ipc=host \
  --restart always \
  --name crypto-bot \
  chatvat-bot

---

🧠 Architecture Deep Dive

ChatVat implements a modular RAG (Retrieval-Augmented Generation) pipeline designed for resilience.

The Components

Component	Role	Description
The Builder	CLI Manager	Runs on host. Lightweight (~15MB). Orchestrates the factory process and Docker builds.
The Cortex	Intelligence	Powered by Groq for ultra-fast inference and HuggingFace for embeddings. Runs inside Docker.
The Memory	Vector Store	A persistent, thread-safe ChromaDB instance. Uses MD5 hashing to silently drop duplicate data during ingestion.
The Eyes	Crawler	A headless Chromium browser (via Crawl4AI/Playwright) managed with --ipc=host stability to read dynamic JS websites.
The Connector	Universal MCP	A polymorphic ingestor that can authenticate with secured APIs using environment-variable masking (e.g., ${API_KEY}).
The API	Interface	A high-performance FastAPI server exposing REST endpoints.

The "Split Strategy" Workflow

Unlike traditional tools that force you to install heavy AI libraries locally:

1. Local (Host): You only have typer, rich, and requests. Fast and clean.
2. Container (Engine): The Docker build installs torch, langchain, playwright, and chromadb.
3. Result: You get the power of a heavy AI stack without polluting your local development environment.

---

🛠️ Configuration Guide

Your bot is defined by chatvat.config.json. You can edit this file manually after running init.

{
    "bot_name": "ChatVatBot",
    "port": 8000,
    "refresh_interval_minutes": 60,
    "system_prompt": "You are a helpful assistant for the .....",
    "llm_model": "llama-3.1-70b-versatile",
    "embedding_model": "all-MiniLM-L6-v2",
    "retriever_k": 5,
    "max_tokens": 400,
    "sources": [
        {
            "type": "static_url",
            "target": "[https://docs.stripe.com](https://docs.stripe.com)",
            "max_depth": 2,
            "recursion_scope": "restrictive"
        },
        {
            "type": "dynamic_json",
            "target": "[https://api.github.com/repos/my-org/my-repo/issues](https://api.github.com/repos/my-org/my-repo/issues)",
            "headers": {
                "Authorization": "Bearer ${GITHUB_TOKEN}",
                "Accept": "application/vnd.github.v3+json"
            }
        },
        {
            "type": "local_file",
            "target": "./policy_docs.pdf"
        }
    ]
}

Source Configuration

Field	Type	Default	Description
type	string	required	"static_url", "dynamic_json", or "local_file"
target	string	required	URL or local file path
headers	object	{}	Custom HTTP headers for API authentication
max_depth	int	1	Crawl depth. 1 = single page, 2+ = follow links recursively
recursion_scope	string	"restrictive"	"restrictive" = only follow links under the target path. "domain" = follow any link on the same domain

Global Configuration

Field	Type	Default	Description
bot_name	string	required	Name of your chatbot
port	int	8000	Deployment port for the FastAPI server
refresh_interval_minutes	int	0	Auto-update interval in minutes (0 = disabled)
system_prompt	string	(default assistant prompt)	Custom persona/system prompt for the LLM
llm_model	string	"llama-3.3-70b-versatile"	Groq LLM model name
embedding_model	string	"all-MiniLM-L6-v2"	HuggingFace embedding model
retriever_k	int	5	Number of top relevant chunks to retrieve from vector DB (1-20)
max_tokens	int	400	Count of total tokens to be produced in response depending upon task

Field Details

• refresh_interval_minutes: Set to 0 to disable auto-updates.
• static_url: Uses Playwright to render JavaScript before scraping.
• max_depth: (static_url only) Controls how deep the crawler follows links. 1 = single page (default), 2+ = recursive BFS crawl.
• recursion_scope: (static_url only) Controls which links the crawler follows. "restrictive" (default) = only links under the target path, "domain" = any same-domain link.
• dynamic_json: Acts as a Universal Connector. Supports custom headers.
• headers: Securely inject secrets using ${VAR_NAME} syntax. The engine resolves these from the container's environment variables at runtime.
• llm_model: You can select your required Groq LLM model while initialising the ChatBot (e.g., llama-3.3-70b-versatile, mixtral-8x7b-32768).
• embedding_model: HuggingFace model for generating text embeddings (e.g., all-MiniLM-L6-v2, all-mpnet-base-v2).
• retriever_k: Controls how many relevant document chunks are retrieved from the vector database for each query. Higher values (e.g., 10-20) provide more context but may include less relevant information. Lower values (e.g., 2-3) are more focused but may miss contextual details. Default is 5.
• max_tokens: Maximum number of tokens the LLM may generate in a single response. Use this to limit output length and guard against token-exhaustion attacks or runaway outputs. Default is 400.

---

📚 API Reference

Once the container is running, interact with it via HTTP REST API.

1. Health Check

Used by cloud balancers (AWS/Render) to verify the bot is alive.

GET /health

Response:

{
  "status": "healthy",
  "version": "0.1.10"
}

2. Chat Interface

The main endpoint for sending queries.

POST /chat

Payload:

{
  "message": "What is the return policy for digital items?"
}

Response:

{
  "message": "According to the policy document, digital items are non-refundable once downloaded..."
}

---

⚠️ Disclaimer & Legal Notice

Author: Madhav Kapila Project: ChatVat - Conversational AI & Web Crawling Engine

This software is provided for educational and research purposes only.

1. No Liability: The author (Madhav Kapila) is not responsible for any damage caused by the use of this tool. This includes, but is not limited to:
   • IP bans or blacklisting of your device/server.
   • Legal consequences of crawling restricted or sensitive websites.
   • Data loss or corruption on the user's local machine or target infrastructure.
2. User Responsibility: You, the user, acknowledge that you are solely responsible for compliance with all applicable laws and regulations (such as GDPR, CFAA, or Terms of Service of target websites) when using this software.
3. "As Is" Warranty: This software is provided "as is", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, and non-infringement.

By downloading, installing, or using this software, you agree to these terms.

---

Built with ❤️ by Madhav Kapila.