Update README.md

README.md

@@ -1,20 +1,26 @@
 # mikupad
 
-**mikupad** is a lightweight and efficient language model front-end powered by ReactJS, all packed into a single HTML file. This project is inspired by the likes of NovelAI and provides a simple yet powerful interface for generating text with the help of various backends.
+**mikupad** is a user-friendly, browser-based interface for interacting with language models. It's built with ReactJS and supports various text generation backends, all within a single HTML file.
+
+![image](https://github.com/user-attachments/assets/4c5fa8ff-5926-4a4b-807b-34e4f36a032c)
 
 ## Features
 
-* **Multiple Backends**: Multiple backends are supported, namely **llama.cpp**, **koboldcpp**, and any other backend with **OpenAI compatible** APIs. You can seamlessly switch between these backends to get different text generation experiences.
-* **Session Persistence**: Your text generation sessions are automatically saved and restored. This means you can work on your text in multiple sittings and continue right where you left off. Import and export your sessions to share your creative work or switch devices effortlessly.
+* **Multiple Backends**: Supports **llama.cpp**, **koboldcpp**, **AI Horde**, and any **OpenAI Compatible** API.
+* **Session Persistence**: Your prompt is automatically saved and restored, allowing you to continue seamlessly across multiple sessions. Import and export sessions for sharing or maintaining backups.
+* **Optional Server**: Can be hosted on a local Node.js server, enabling database access remotely or across your local network.
 * **Persistent Context**:
   * **Memory**: Seamlessly inject a text of your choice at the beginning of the context.
   * **Author's Note**: Seamlessly inject a text of your choice at the end of the context, with adjustable depth.
   * **World Info**: Dynamically include extra information in the context, triggered by specific keywords.
-* **Prediction Undo/Redo**: It's possible to undo and redo predictions, making it easy to experiment and fine-tune your generated text until it's just right.
-* **Token Probability** *(llamacpp/openai backend)*: When you hover over a token in the generated text, a list will show at most 10 tokens with their probabilities. This information can be a valuable aid in refining your text. Moreover, you can click on another token's probability to restart text generation from that point.
-  * If you're using oobabooga, make sure to use an _HF sampler for this feature to function properly.
-* **Logit Bias** *(llamacpp/koboldcpp backend)*: Adjust the likelihood of certain tokens being generated on the fly.
-* **Completion/Chat**: Let your model directly continue your prompt in Completion Mode, or toggle on Chat Mode to have your input be wrapped in user-defined instruct formats.
+* **Prediction Undo/Redo**: Easily experiment and refine your generated text with the ability to undo and redo predictions.
+* **Token Probability**: Hover over any token to reveal the top 10 most probable tokens at that point. Click on a probability to regenerate the text from that specific token.
+  * 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 the generation process by adjusting the likelihood bias of specific tokens on-the-fly.
+* **Completion/Chat Modes**:
+  * **Completion**: Have the language model directly continue your prompt.
+  * **Chat**: Mikupad simplifies using instruct models. It automatically adds the right delimiters when you start or stop generating, based on your selected template. This also structures your prompt into messages, making it compatible with the Chat Completions API (for OpenAI-compatible backends).
 * **Themes**: Customize your environment by choosing from a variety of themes.
 
 ## Getting Started