Skip to content

aerugo/stina

BranchesTags

Repository files navigation

Stina AI Chat Client

Screenshot 2025-01-15 at 18 12 42

Stina is a lightweight AI chat client specifically designed for environments with strict security or IT restrictions. In government, healthcare, and large corporate settings, installing servers or external dependencies is often not allowed; internet access to AI services might be heavily controlled, and only certain APIs or local LLMs (like Ollama) can be used.

Stina requires no server, no software installation, and no build tools—just download the repository and open chat.html in a modern browser. You can still configure advanced setups via optional JavaScript config files or by editing settings in the UI. Stina supports multiple LLM providers (Azure, OpenAI, Anthropic, Ollama), custom instructions, dynamic system prompts, and SharePoint compatibility.

Table of Contents

Introduction

Stina is an AI chat client built with pure JavaScript, HTML, and CSS—no Node.js or build step required. Its key design goals:

  • Run from the local filesystem (open chat.html directly).
  • Zero server dependencies—everything is processed client-side, except for calls to your chosen LLM endpoints (Azure, OpenAI, etc.).
  • Strict-environment friendly—ideal for locked-down enterprise or government setups.
  • Ollama integration—use local LLMs if you have Ollama running on your machine.

Features

  • No Build Dependencies: Simply download and open chat.html. No server or package installations are needed.
  • API Key Support: Connect to Azure AI Foundry, OpenAI, Anthropic, or other providers using your own keys.
  • Ollama Integration: Run local language models if you have Ollama installed.
  • Dynamic System Prompts (Instructions): Edit or add new instructions to change the assistant’s behavior on the fly.
  • Multiple Models: Switch to different models mid-chat (e.g., from GPT-4 to GPT-3.5 or a local model).
  • Multi-language UI: English and Swedish translations are included by default.
  • SharePoint Compatibility: Rename chat.html to chat.aspx and upload to SharePoint if that’s your environment.
  • User-Friendly: A simple, clean interface with a sidebar of chats, an input area, and a customizable prompt toolbar.

Getting Started

Prerequisites

  • A modern web browser (Chrome, Firefox, Edge, or Safari).
  • (Optional) API keys for your desired LLM provider(s) (Azure, OpenAI, Anthropic).
  • (Optional) A local Ollama instance if you want to run local LLMs.

Installation

  1. Clone or Download this repository:

    git clone https://github.com/yourusername/stina.git

    or download the ZIP and extract it.

  2. (Optional) Add Keys/Configs: If you wish to predefine providers or instructions, see Using Configuration Files.

  3. Open chat.html:

    • Directly: Double-click chat.html or select “File > Open...” in your browser.
    • If your browser blocks local file scripts for security reasons, you can:
      • Change local security settings, or
      • Temporarily serve the folder via a minimal local server (e.g., python -m http.server 8080) and visit http://localhost:8080/chat.html.
    • If you are on a closed network that supports SharePoint, see Accessing Through SharePoint.

Configuration

Stina can pull configuration from optional JavaScript files or from local storage via the UI. This helps in strict environments: you can either edit the code or just do everything in-memory via a “Settings” modal.

Using Configuration Files

  1. Locate Template Files (if provided or in your custom fork):

    • providers.template.js
    • models.template.js
    • instructions.template.js

  2. Create Actual Configuration Files:

    • providers.template.js → providers.js
    • models.template.js → models.js
    • instructions.template.js → instructions.js

  3. Edit Configuration:

    • providers.js: Enable/disable providers and insert your API keys/endpoints: 
      window.providerConfigs = {
  azure: {
    enabled: true,
    endpoint: "https://YOUR_RESOURCE_NAME.openai.azure.com",
    apiKey: "YOUR_AZURE_API_KEY",
  },
  openai: {
    enabled: false,
    apiKey: "YOUR_OPENAI_API_KEY",
  },
  // ...
};
    • models.js: Add or override model definitions, including temperature, max tokens, etc.: 
      window.additionalModels = {
  "gpt-4-custom": {
    label: "GPT-4 Custom",
    deployment: "your-azure-deployment",
    provider: "azure",
    context_length: 8000,
    max_tokens: 2000,
    temperature: 0.7,
    system: true,
  },
};
    • instructions.js: Add additional instructions or system prompts: 
      window.additionalInstructions = [
  {
    id: "custom_instructor",
    order: 3,
    label: "Custom Instructor",
    content: "Provide guidance in a friendly, concise manner.",
  },
];

In-Memory Configuration

If you cannot create/edit local files or prefer not to:

  1. Open Stina—just open chat.html in your browser.
  2. Click “Settings” (top-right gear icon).
  3. Providers tab:
    • Enable providers and enter your keys/URLs.
  4. Language tab:
    • Switch the UI between English and Swedish (or your custom translations if added).
  5. Data tab:
    • Import/export chats, clear all local storage, etc.

These in-memory settings are stored in your browser’s IndexedDB (and partially local storage). They persist for that browser/profile on the same machine.

Configuring Azure AI Foundry with User-Provided API Keys

You can set up a common Azure endpoint but require each user to enter their own API key:

  1. Edit providers.js: 
    window.providerConfigs = {
  azure: {
    enabled: true,
    endpoint: "https://YOUR_RESOURCE_NAME.openai.azure.com",
    // Omit apiKey so each user enters their own
  },
  openai: { ... },
  // ...
};
  2. User Steps:
    • Open chat.html → Settings → Providers → Azure → input personal apiKey.
    • Save → The key is stored locally in that user’s browser.

Advantages:

  • Maintains security by not bundling any shared key.
  • Centralizes the endpoint while letting each user handle their own credentials.

Usage

Running Stina

  1. Open chat.html in your browser.
  2. Initial Setup:
    • If you do not have any config files, you may need to add your keys in Settings.

Using Features

  • New Chat:
    • Click New Chat in the navbar.
  • Sending Messages:
    • Type in the bottom input box; press Enter or click Send.
  • Switching Models:
    • Use the model selector in the toolbar beneath the input to switch LLMs.
  • Instructions (System Prompts):
    • Drop down the instructions menu. Choose an existing prompt or select Create new instruction....
  • Sidebar Management:
    • All chats are listed in the left sidebar. Click to switch; click × to delete.
  • Settings:
    • Configure providers, API keys, endpoints, language, etc.

Example: Collaborating with Mini-Agents

Screenshot 2025-01-15 at 18 11 49

Scenario: You want input from both a technical specialist and a legal specialist in the same conversation.

  1. Create Instructions

    • “Technical Expert” → a prompt for network security topics.
    • “Legal Expert” → a prompt for regulatory compliance.

  2. Select Technical Expert

    • Send questions on infrastructure best practices; get specialized answers.

  3. Switch to Legal Expert

    • In the instructions dropdown, pick “Legal Expert.” Ask about compliance or privacy regulations.

  4. Switch Models (optional)

    • If you have different models for different tasks, select them in the model dropdown.

  5. Continue

    • The chat retains context. You can jump back to the “Technical Expert” at any time.

  6. Export

    • In Settings > Data > Export Chat, you can save the conversation as JSON.

Benefit: You effectively have two different system prompts in the same conversation, letting you switch roles fluidly.

Customization

Adding Custom Instructions

  1. In the UI:
    • Create new instruction... from the instructions dropdown → name it, write content, save.
  2. Using instructions.js (filesystem approach): 
    window.additionalInstructions = [
  { 
    id: "friendly-assistant", 
    order: 4, 
    label: "Friendly Assistant", 
    content: "Provide kind, helpful responses."
  }
];

Adding Custom Models

  1. Using models.js: 
    window.additionalModels = {
  "custom-model": {
    label: "Custom Model",
    deployment: "gpt-3.5-mydeployment",
    provider: "openai",
    max_tokens: 1500,
    temperature: 0.5,
    system: true
  }
};

If you have providers.js set up, ensure that custom-model references a provider that’s enabled.

Data Privacy

  • Local Storage: Chat history, instructions, and configurations persist in your browser’s IndexedDB.
  • No Server: No messages or data are sent to any server outside of calls to the configured LLM endpoints.
  • Self-Control: You can clear all data in Settings > Data or by clearing your browser’s storage.

Note: On shared machines, other users with the same browser profile could access your stored data.

Accessing Through SharePoint

If you need to host Stina on SharePoint:

  1. Rename chat.html to chat.aspx.
  2. Upload chat.aspx and all src/ files to your SharePoint document library, preserving their structure.
  3. Access the .aspx page in SharePoint to run Stina.

Planned Features

  • Export/Import: More robust import/export (JSON or YAML).
  • Advanced File Upload: Parse PDFs, DOCX (partially supported now), images, or entire knowledge bases.
  • Vector Database: Possibly store large doc sets for advanced retrieval-augmented generation.
  • Additional Providers: Add more LLM endpoints as usage demands.

Contributing

We welcome community contributions! Feel free to:

  1. Fork the repo.
  2. Create a branch (feature/your-feature).
  3. Commit and push changes.
  4. Open a Pull Request, describing your feature or fix.

License

Stina is licensed under the MIT License. See the LICENSE file for details.

Bundled Libraries and Credits

Stina includes or references:

All trademarks are the property of their respective owners. Contributions and issues are welcome!

About

Chatbot client that runs without a server.

Resources

Readme

License

MIT license
Activity

Stars

2 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages