alexbenko/ChatMancer

✍️ Contextual Conversations – remembers your chat history across messages

🤖 Model Switching – change models (e.g. GPT-3.5 ↔ GPT-4) on the fly without losing context

🖼️ Image Generation – describe what you want and get visuals via DALL·E

📄 PDF Q&A – upload a PDF and ask natural questions about its contents

🔐 Secure Auth – session-based access with password + token + CSRF protection

🧠 Code Display – beautifully rendered code with syntax highlighting and language labels

ChatMancer is a full-stack app powered by large language models (LLMs) that delivers an advanced, ChatGPT-like experience with added flexibility, control, and lower cost potential. Built on OpenAI's API, it supports contextual conversations, image generation, and document analysis — with full model-switching support.

• Custom memory system preserves message history across requests using session IDs.
• Tracks user/AI messages, model used, and content type (text or image).
• Supports future features like analytics, , or persistent logging.

• Easily switch between models (e.g., GPT-4, Claude) at any time.
• Session memory remains intact regardless of which model is used.
• Chat history is tracked separately from the LLM logic for full flexibility.

• Users can generate images using natural prompts like "draw a dragon in a coffee cup" — no command prefix needed.
• Powered by DALL·E.
• Image messages are stored with content_type: "image" for structured frontend rendering.

• Upload a PDF and ask questions about it directly.
• RAG (retrieval augmented generation) pipeline extracts and answers using the document.
• Automatically detects relevant context without requiring a command prefix.

• Uses custom invoke_with_metadata() to manage LLM execution, memory storage, and message tagging in one place.
• Code blocks are rendered with fenced Markdown (```lang) and copy buttons for readability.
• Flexible memory architecture supports per-message metadata, like model and content type.

1. Text file upload and QA support (RAG)
2. Add support for pulling content from URLs
3. GPT Vision / image Q&A support
4. Persistent model selection with historical tracking
5. Response .

This project uses OpenAI's APIs, which are not free. Costs depend on:

• Number of requests
• Models used (e.g., GPT-4 is more expensive than GPT-3.5)
• Image generation and embeddings

Most casual users will spend <$10/month, but you can set limits via your OpenAI billing dasard (e.g., $20 cap).

In your .env file under /api:

Your OpenAI API key. Required for GPT and DALL-E access.

Password protection for the app interface (basic access control). Passwords need to be encrypted using /api/lib/pass.py

git clone https://.com/alexbenko/ChatMancer.git
cd ChatMancer

1. Add .env to /api with required variables.
2. Build Docker image:

docker build -t chatmancer .

3. Run it:

docker run -p 8000:8000 chatmancer

4. Open the printed URL, enter the password, and chat away.

git clone https://.com/alexbenko/ChatMancer.git
cd ChatMancer

1. Install Python server deps:

pip install -r requirements.txt

2. Add your .env file in /api:

OPENAI_API_KEY=your_key
PASSWORD=your_password

3. Start backend:

cd api
python main.py

4. Start frontend:

cd ../web
npm install
npm run dev

5. Visit the printed URL and authenticate.

• Vite — Lightning-fast frontend tooling
• React — UI library
• TypeScript — Static typing
• Python 3.11 — Backend language
• FastAPI — Async web framework
• LangChain — LLM orchestration and memory
• OpenAI API — GPT + DALL-E access

Open to PRs, ideas, and feedback!

see LICENSE for full details.