SerialKicked/Lethe-AI-Repo

Two models required by the Lethe-AI C# Middleware Library

gte-large.Q6_K.gguf

A Q6_K quantized version of General Text Embeddings (GTE) model under MIT License. Used for all things RAG in the library.

emotion-bert-classifier.gguf

A quantized version of Emotions Analyzer under MIT License, a fine-tuned BERT-base-uncased on GoEmotions for multi-label classification (28 emotions). Used for sentiment analysis tasks.

---

Lethe AI - A C# Middleware LLM Library

Powerful, object-oriented, and highly configurable, general purpose library used to connect a local back-end running a Large Language Model (LLM) to a front-end. This library offers many tools and features for those who want to code their own C# front-end or LLM-powered tools without having to do all the heavy lifting. It can happily connect itself to the most popular backends (the program loading the LLM proper) and allows your code to "speak" with the LLM is a few function calls. It includes easy-to-use (and easy to build upon) systems to handle most of the operations you'd wish to do with a LLM, alongside many advanced features like RAG, agentic systems, web search, text to speech, semantic similarity testing, and prompt manipulation.

🧩 Compatible Backends

• Kobold API: Used by KoboldCpp. This is the recommended backend with the most features.
• OpenAI API: Used by LM Studio, Text Generation WebUI, and others. Less features.

Lethe AI technically supports remote backends but this hasn't been tested, this library is mostly designed for local (or local network) LLM inference.

⭐ Main Features

• Easy to use classes for bot personas, system prompts, instruction formats, and inference settings
• Session-based chatlog with automated summaries for past sessions
• Streamed (or not) inference / reroll / and impersonate functions
• Support CoT / "thinking" models out of the box
• GBNF grammar generation directly from a class's structure for structured output
• Basic support for VLM (visual language models) depending on the back-end
• Tools for reliable Web Search (DuckDuckGo and Brave)
• Text To Speech support (through the Kobold API only)
• Many useful tools to manipulate text, count tokens, and more

📝 Long Term Memory System

• Keyword-triggered text insertions (also known as "world info" in many frontends)
• Customizable RAG System using the Small World implementation
• Automatic (optional) and configurable insertion of relevant past chat sessions into the context

🧠 Agentic and Brain Module for personas

• Background agent system (bot can run tasks in the background)
• Analyze past chat sessions, run relevant web searches and mention results in next session
• Mood tracking + drift system (personality coloring over time)
• Goal‑driven behaviors (long‑term projects, self‑seeding topics of interest)

🛠️ Advanced Features (Work in progress / experimental)

• Group chat functionalities (one user and multiple AI characters)
• Sentiment analysis

👀 See it in action

To demonstrate how powerful Lethe AI can be, check out Lethe AI Chat. This is a powerful AI chat program for Windows that uses most of the features present in the library. It comes with its own integrated editors, extended agentic tasks, and extensive settings. It can rival with most of the dedicated AI chat programs currently available.

The library's source code is available here: Lethe-AI Official Github