Stina is a lightweight AI chat client specifically designed for environments like government, healthcare, and large corporate settings, where IT policies prohibit the use of web servers, installations of new software, or access to online AI chat bots. Chat with with LLMs with your own API keys or a local Ollama instance after downloading the repo and opening the chat.html file from the file system.

Stina keeps all conversation history and data in localstorage on the browser, which is great when your manager asks you about data security considerations of AI chatbots.

Stina is a client-side solution that requires no server, no installations, and no build tools. Written in vanilla javascript with all dependencies bundled. This enables editing the code in environments where there is no access to Node.js, NPM or other JS-build tools.

Right now it supports Azure AI Foundry deployments, OpenAIs API, Anthropic and Ollama.

Stina can be hosted on a SharePoint server, run directly from the filesystem, or accessed through a local web server. It supports multiple language models, custom instructions, and dynamic system prompts.

I built this for my own needs at work where we have access to API keys to an Azure AI Foundry sandbox on the company tenant. By renaming chat.html to chat.aspx and putting the repo in a folder on SharePoint, I can run this as a chatbot internally.

If you want to contribute, you are very welcome to do so. I would love to extend this with the ability to chat to files, and perhaps even add an in-memory clientside vector database support to allow doing lightweight clientside RAG on documents accessible with the filesystem.