ChatGPT-like streaming chat bot powered by Azure OpenAI and GPT-4
NOTE: project and documentation in progress
This is a simple chat app that streams messages to and from an Azure OpenAI Services instance using a GPT-4 model deployment
- Node.js >= 20.x (with npm >= v10.x)
- Azure Subscription
- Azure OpenAI access enabled
- Azure OpenAI Services deployed along with a GPT-4 model
- API key for deployed service
- Clone repo
git clone https://github.com/cu-cit-cloud-team/azure-openai-gpt4-chat.git
- Enter directory
cd azure-openai-gpt4-chat
- Copy
.env.local.example
to.env.local
and fill in values - If using locally installed Node.js:
- Install dependencies
npm install
- Run project locally
npm run dev
- Install dependencies
- If using Docker:
- Run
docker-compose up dev
- Run
- Visit
http://localhost:3000
in your browser ctrl
+c
to stop- If using Docker, optionally run
docker-compose down --rmi 'all'
to clean up
- If using Docker, optionally run
- Chat with an Azure OpenAI GPT-4 model
- Chat responses stream in real-time for a ChatGPT-like experience
- Uses
textarea
for input for multi-line support- Focus
textarea
on page load -
textarea
auto-resizes as you type -
enter
key to submit message -
shift
+enter
for new line
- Focus
- Loading indicator next to chat response until it's finished
- Formats code blocks and other markdown inside of chat messages nicely
- Current chat history is stored in browser's local storage
- View/edit system message for chat session (e.g. "You are a helpful AI assistant")
- View/edit chat session parameters (e.g. temperature, top p, etc)
- Live token count (used/remaining) displayed for message and system message
- Change Azure OpenAI model (GPT-4, GPT-3.5, etc)
- Download current chat history as JSON
- Clear current chat history
- Keyboard shortcut (
cmd
+esc
) whentextarea
has focus
- Keyboard shortcut (
- Copy full messages and responses to clipboard
- Copy individual code snippets (within responses) to clipboard
- Delete individual messages and responses from chat history
- Stop response that's currently streaming in (useful for long responses that aren't necessary)
- Regenerate last response
- Hosted via Azure App Service
- Built/deployed with GitHub Actions
- Azure App Service App locked down via AD to just our team
- Tries to identify user from login session and personalize chat session
- Uses relative time for chat message timestamps
- Relative time has tooltip with full timestamp
- Tooltip shows time for same day and full date/time if older than "today"
- Theme changer with several light and dark themes
- Defaults to
light
ordark
based on system preference
- Defaults to
- Responsive layout
- Dockerfile and docker-compose for local development
- no new features currently under development
- Import chat session from JSON
- Multiple chat sessions
- Sidebar with scrollable and filterable list of chat sessions
- Stateful via local storage
- Delete individual chat sessions
- Export individual chat sessions to JSON
- Import chat sessions from JSON
- React
- JavaScript framework
- Next.js
- React framework
- Vercel AI SDK
- JS Library for AI-powered UIs
- LangChain.js
- JS Library for interacting with LLMs (e.g. GPT-4)
- Tailwind CSS
- CSS framework
- daisyUI
- Component library for Tailwind CSS
- Azure OpenAI Service
- API and model deployments (GPT-4 and GPT-3.5)
- Azure App Service
- Application hosting and user authentication
- GitHub Actions/Workflows
- Build/publish to Azure App Service
- Build notifications
- Create GH Release
- Dependency updates