🐈 miyapad

Miyapad is a user-friendly, browser-based interface for interacting with language models. It is a fork of mikupad by lmg-anon with a major architectural overhaul, modern tooling and expanded features.

Features

• Multiple Backends: Supports llama.cpp, koboldcpp, AI Horde, DeepSeek and any OpenAI Compatible API.
• Connection Manager: Save named connection presets (endpoint, API type, API key, model) and switch between them per-session. Each session remembers its selected connection. Presets include per-API-type settings, a model browser, and CRUD operations (create, clone, delete, enable/disable).
• Session Persistence: Your prompt is automatically saved and restored across multiple sessions. Import and export sessions for sharing or backups. The dedicated Sessions modal provides search, sort by name/created/modified, and a table layout for managing sessions.
• Optional Server: Can be hosted on a local Node.js server for remote or LAN access. Features sqlite-zstd transparent Zstandard compression with auto-vacuum and background dictionary training, and optional server-side tokenization — drop a tokenizer.json into server/tokenizers/<name>/ and enable via Preferences → Server.
• Persistent Context:
  • Memory: Seamlessly inject text at the beginning of the context.
  • Author's Note: Seamlessly inject text at the end of the context, with adjustable depth.
  • World Info: Dynamically include extra information triggered by specific keywords.
• Prediction Undo/Redo: Easily experiment and refine generated text.
• Token Probability: Hover over any token to reveal the top 10 most probable tokens at that point. Click on a probability to regenerate text from that specific token. Token probability gradient endpoints and erase highlight are fully customizable via CSS variables instead of hardcoded. Optional server-side tokenization replaces client-side counting with accurate server-side counts.
  • If you're using oobabooga, make sure to use an _HF sampler for this feature to function properly.
  • If you're using koboldcpp, token probabilities are only available with Token Streaming disabled.
• Logit Bias: Fine-tune generation by adjusting the likelihood bias of specific tokens on-the-fly.
• Completion/Chat Modes:
  • Completion: Have the model directly continue your prompt.
  • Chat: Automatically adds the right delimiters based on your selected template, structuring prompts into messages compatible with the Chat Completions API.
• Screenshot Capture: Select text in the editor, click the camera icon, and render a styled quote PNG with AI vs User color coding. Customize via the gear icon — fonts, colors, background image, avatar, and metadata toggles.
• Themes: Customize your environment with a variety of themes.
• ... and more!

Architecture

This refactored fork moves from a single monolithic HTML file to a modular Parcel 2 project (~60+ files across src/). The monolithic styles.css is split into 20 component-specific partials under src/css/, and global state is managed via React Context API (SettingsContext, GenerationContext) instead of inline global state.

Getting Started

Open miyapad on GitHub Pages or download the pre-compiled miyapad.html from Releases and open it in your browser.

For the full server + frontend, download the standalone distribution archive from Releases for your platform, extract, and run the launch script.

Optional Server

Standalone Distribution (unzip and run)

1. Download miyapad-<platform>-x64.{tar.gz,zip} from Releases
2. Extract the archive
3. Run ./miyapad.sh (Linux/macOS) or miyapad.bat (Windows)

No install needed — Node.js and native addons are pre-packaged.

From Source

From the server/ directory:

npm install
npm start

See Backend Server for CLI options and environment variables.

Building

npm install && npm run build        # Frontend only → dist/miyapad.html
npm run build:dist                  # Frontend + standalone distribution → server/miyapad-dist/

Security

Network security concerns associated with hosting Miyapad on a public-facing server are out of scope of this project. It is the user's responsibility to secure their instance (e.g., with a reverse proxy, authentication, firewall rules, etc.) when exposing it to untrusted networks. Pull requests addressing such concerns are welcome and will be reviewed, as long as they don't break any localhost functionality.

Contributing

Contributions are welcome. To contribute:

1. Fork the repository.
2. Create a new branch: git checkout -b feature/your-feature-name
3. Make your changes and commit: git commit -m 'Add your feature'
4. Push to your fork: git push origin feature/your-feature-name
5. Open a pull request.

License

This project is licensed under the GNU Affero General Public License v3.0 — see the LICENSE file for details.

Miya