This project provides a simple chatbot webapp that is intended to help educate folks on how LLM-based interactions occur. This chatbot displays all messages going back and forth from the LLM, including system messages and tool execution requests and responses.
Additional features include:
- System prompt adjusting/rebasing - see how a conversation would change if you changed the initial system prompt
- Dynamic tool creation - add custom tools in the middle of a conversation to give the LLM new abilities
- MCP support - add tools by starting MCP servers
- AI-generated tools - give the LLM the ability to create a tool, which it then can execute
Run the following command to give it a try!
docker run -dp 3000:3000 -v /var/run/docker.sock:/var/run/docker.sock mikesir87/visual-chatbotNOTE: The Docker socket is mounted to allow the tool to launch containerized MCP servers
Once the container has started, you can open the app at http://localhost:3000/.
The application obviously needs to have an LLM to operate against. When it starts, a configuration modal will pop up, where you will have the ability to choose between the following LLM backends:
- OpenAI (an API key will be required)
- Ollama - will auto-detect whether it connects using localhost or
host.docker.internal(if running in a container) - Docker Model Runner - will auto-detect if it needs to connect via localhost or
model-runner.docker.internal
Once the backend is selected, a collection of models are presented. For Ollama and the Docker Model Runner, the list consists of those currently downloaded.
If you wish to setup the development environment, do the following:
-
Run the following in each of the
apiandclientdirectories:npm install npm run dev
-
Open your browser to http://localhost:5173/.
And don't worry... a containerized setup will be coming soon ;)
This project is mostly a for-fun training aid, so is likely fairly close to being "done." But, feel free to open an issue if you'd like and start a discussion.