v2.2.0

Author: HardCodeDev777

.gitattributes

@@ -1 +1,5 @@
+* text=auto
 *.bat linguist-detectable=true
+
+# I almost didn't change code and I don't want whole Whisper.unity C# statistic in this repository and my profile stats
+src/Unity/Assets/UnityNeuroSpeech/Whisper/** linguist-vendored

.gitignore

@@ -1,4 +1,21 @@
 # MkDocs
 .venv/
 .idea/
-site/
+site/
+
+# CMake and result output
+src/Native/build/
+src/Native/result/
+
+# VS
+src/UnityExtensions/.vs
+src/UnityExtensions/UnityExtensions/bin
+src/UnityExtensions/UnityExtensions/obj
+
+# Unity 
+# It's not project files to download and open with unity. It contains only frameworks files, 
+# critical assets (TTS models, audio files) are excluded. Downloaded latest Release to get a full working version for your Unity project.
+/src/Unity/**
+!/src/Unity/Assets/
+!/src/Unity/Assets/UnityNeuroSpeech/
+!/src/Unity/Assets/UnityNeuroSpeech/**

CHANGELOG.md

@@ -1,6 +1,27 @@
 > [!IMPORTANT]
 > Only the latest version is officially supported.
 
+# v2.2.0 [21.01.2026]
+
+## Unity
+
+- **Fixed IL2CPP freeze when running TTS process.**
+- **Replaced deprecated `Microsoft.Extensions.AI` with OllamaSharp.**
+- Centralized all framework paths into `StaticData.cs`.
+- Simplified **UNS Manager Creation** — now only Whisper model name is required instead of full `StreamingAssets/` path.
+- Removed legacy and low-value features (IL2CPP logging, custom framework locations).
+- `CreateSettings.cs` no longer generates `CreateAgent.cs` from template.
+- Improved code comments, logs, and internal validation.
+
+---
+
+## ~Setup~
+
+- Setup script was removed and replaced with a fully documented manual installation process.
+
+---
+
+ 
 
 # v2.1.0 [23.11.2025]
 

LICENSES.md

@@ -18,19 +18,6 @@
 
 ---
 
-> [!IMPORTANT]
-> Full response time (from your speech to TTS generated voice) can sometimes take a minute or more right now.
-> This isn't a bug – it's the current reality of running powerful AI models locally and for free:
-> - Good STT (like Whisper) needs time to be accurate.
-> - Even small LLM (via Ollama) needs some time to think up a good response.
-> - Generating high-quality voice with TTS is also a complex and not fast task.
-> 
-> The key thing is: this framework is built on the most optimal and user-friendly local solutions available for each stage (STT, LLM, TTS). You have the freedom to choose and download your own models (like Whisper `.bin` for STT and any model for Ollama), to use any custom voice you want, to make it for different languages, to customize your agents and to find the perfect balance between speed and quality for your setup.
-> 
-> Also this project is actively maintained. With every update, I'm working on making it faster, more optimized, and easier to use!
-
----
-
 UnityNeuroSpeech is an open-source framework for creating **fully voice-interactive AI agents** inside Unity.  
 It connects:
 
@@ -76,7 +63,7 @@ No subscriptions, no accounts, no OpenAI API keys.
 
 ## 🧪 Built with:
 
-- 🧠 [Microsoft.Extensions.AI](https://learn.microsoft.com/en-us/dotnet/ai/) (Ollama)
+- 🧠 [OllamaSharp](https://github.com/awaescher/OllamaSharp) 
 - 🎤 [whisper.unity](https://github.com/Macoron/whisper.unity)
 - ⚡ [UniTask](https://github.com/Cysharp/UniTask)
 - 🧊 [Coqui XTTS](https://github.com/idiap/coqui-ai-TTS)

README.md

@@ -1,21 +1,88 @@
-# 🚀 Quick Start
+# 🚀 Getting Started
 
 ---
 
 ## 🛠 Installing Requirements
 
+UnityNeuroSpeech requires several things to be installed before using it in Unity. Here what you need to install:
+
 ---
 
-UnityNeuroSpeech requires several programs to be installed.  
-You can simply run `setup.bat` — it will download everything automatically.  
-Then just import the `.unitypackage` into your project.
+### 1. Ollama
+
+**Ollama** is a platform for running large language models (LLMs) locally. You can use models like DeepSeek, Gemma, Qwen, etc. 
+Note that small models might affect accuracy and context understanding, but big models response can take a long time.
+
+Install it from the [official website](https://ollama.com/download).
+
+Then you need to download LLM model with this command:
+```console
+ollama pull modelname
+```
+> For quick test I recommend to download **qwen2.5:3b** - it responds very fast.
 
 ---
 
-## 🎙️ Voice Files
+### 2. STT
+
+**Whisper** — Speech-To-Text model that transcribes and translates audio with high accuracy.
+
+You need to download Whisper model from [here](https://huggingface.co/ggerganov/whisper.cpp/tree/main).
+> For quick test I recommend to download `ggml-base.bin`.
+
+---
+
+### 3. TTS
+
+**UV** — a modern, ultra-fast Python package and environment manager. It replaces traditional tools like `pip`. 
+**Coqui TTS**(runs XTTS) uses **UV** to simplify installation and allows running the TTS command directly, without manual Python setup.
+
+**Coqui XTTS** — a Text-To-Speech model that can generate speech in any custom voice you want: Chester Bennington, Vito Corleone (The Godfather), Cyn (Murder Drones) or any other.
+
+ 
+
+Install **UV** with this command:
+
+```console
+powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
+```
+
+Then install **Coqui TTS**:
+
+```console
+uv tool install --python 3.11 "coqui-tts==0.27.2"
+```
+> You can try to install latest **Coqui TTS** version, but I can't guarantee that it will work.
 
 ---
 
+### 4. Unity deps
+
+UnityNeuroSpeech uses **UniTask** for... I think everyone knows what **UniTask** is. 
+You can install it from the [official repository](https://github.com/Cysharp/UniTask/releases).
+
+Or via **UPM**:
+
+```console
+https://github.com/Cysharp/UniTask.git?path=src/UniTask/Assets/Plugins/UniTask
+```
+
+---
+
+### 5. UnityNeuroSpeech
+
+Now you can finally import **UnityNeuroSpeech** to your Unity project. 
+
+Download `UnityNeuroSpeech.X.X.X.unitypackage` and `UNS_StreamingAssets.unitypackage` from the [official repository](https://github.com/HardCodeDev777/UnityNeuroSpeech/releases). Then you can import them in your project.
+
+They are splitted to avoid importing almost 2GB XTTS model in your project if you only need code/fixes.
+
+**Don't forget to put your downloaded Whisper model(`.bin`) in `Assets/StreamingAssets/UnityNeuroSpeech/Whisper/` - yes, it's important.**
+
+---
+
+## 🎙️ Voice Files
+
 Don’t forget that you need voice files for TTS speech.  
 Make sure your files meet the following requirements:
 
@@ -38,7 +105,5 @@ All voices must be placed in:
 
 ## 🖼️ Microphone Sprites
 
----
-
-You’ll need two sprites for the microphone state (enabled/disabled).  
-Yes — without them, it won’t work 🤠
+You’ll need two sprites for the microphone state (enabled/disabled).
+But for quick test you can use random default sprites from Unity.

docs/index.md

@@ -48,8 +48,6 @@ public class AlexBehaviour : AgentBehaviour
 
 ### 🔍 Methods Overview
 
----
-
 - **AfterTTS** — Called after the audio playback finishes.  
 - **BeforeTTS** — Called before sending text to the TTS model.  
 - **AfterSTT** — Called after the STT model finishes transcribing microphone input.  
@@ -70,8 +68,6 @@ public override void Awake()
 
 ### 💡 What Is `SetBehaviourToAgent()`?
 
----
-
 The `SetBehaviourToAgent()` method connects your `AgentBehaviour` to the agent’s internal event hooks:
 
 ```csharp

docs/unity/agent-api.md

@@ -9,40 +9,24 @@ You can find tooltips for each field directly in the Unity Editor.
 
 ## Step 1. 🧪 Settings
 
----
-
 Go to **UnityNeuroSpeech → Main → Create Settings** in the Unity toolbar.  
 Default settings are recommended.
 
 ---
 
 ## Step 2. 👀 UNS Manager
 
----
-
 **UnityNeuroSpeech Manager** is a GameObject in your scene that controls all non-agent scripts.  
 Without it, no agent (talkable AI) will work.
 
----
 
 Create a `Dropdown` in your scene.  
-Then go to **UnityNeuroSpeech → Main → Create UNS Manager**.  
-
-The important setting there is:
-
-- **Whisper model path in StreamingAssets** — path to your downloaded Whisper model (`.bin`) inside the `StreamingAssets` folder (without the `Assets` directory).  
-  Example:  
-  If the full path is  
-  `Assets/StreamingAssets/UnityNeuroSpeech/Whisper/ggml-medium.bin`  
-  then you should enter  
-  `UnityNeuroSpeech/Whisper/ggml-medium.bin`
+Then go to **UnityNeuroSpeech → Main → Create UNS Manager**.
 
 ---
 
 ## Step 3. 🧠 Agent
 
----
-
 An **Agent** in UnityNeuroSpeech is a GameObject that can listen, respond, and talk using LLM.  
 **Once you create your first agent, you’ll be able to talk with your AI!**
 
@@ -52,7 +36,7 @@ Add a `Button` and an `AudioSource` to your scene.
 Then go to **UnityNeuroSpeech → Main → Create Agent**.  
 Here are some important settings:
 
-- **Agent index** — the index mentioned in the QuickStart.  
+- **Agent index** — the index mentioned in the Getting Started.  
   It links an agent to its voice file.  
   ⚠️ Each agent must have a unique index!
 
@@ -95,7 +79,7 @@ Agent performance (“speed”) depends on:
 - Voice files length  
 - AI response size  
 
-Small models like **deepseek-r1:7b** or **ggml-tiny.bin** run fast but may ignore system prompts (emotions, actions, etc.).  
+Small models like **qwen2.5:3b** or **ggml-base.bin** run fast, but may ignore system prompts (emotions, actions, etc.).  
 Large models like **ggml-large.bin** usually work perfectly — but will be very slow 😐
 
 > On first load, TTS may respond slowly — it’s ok. It'll work faster next time.

docs/unity/steps-to-make-it-work.md

@@ -2,37 +2,35 @@
 
 ---
 
-UnityNeuroSpeech provides several Editor tools to make development more convenient.
+UnityNeuroSpeech provides several Editor tools to improve development experience.
 
 ---
 
 ## 🗒️ Prompts Test
 
----
 
 Let’s say you want to check how a selected LLM model responds to a specific prompt.  
-Normally, you would have to run the game, wait for Whisper to load, say something into the microphone (and risk transcription errors), then wait for the LLM and TTS to finish — quite the hardcore workflow, right?
+Normally, you would have to run the game, wait for Whisper to load, say something into the microphone, 
+then wait for the LLM and TTS to finish — quite the hardcore, right?
 
 This tool allows you to test prompts instantly.  
 You only wait for the **LLM** (as usual) to generate a response — and you can even see the **generation time in milliseconds**!
 
----
 
-To access it, go to **UnityNeuroSpeech → Tools → Prompts Test**.
+Go to **UnityNeuroSpeech → Tools → Prompts Test**.
 
 ---
 
 ## 🕵️‍♂️ Decode Encoded
 
----
 
 If you use AES encryption, your `.json` dialog history files will be encrypted.  
 But what if you want to view their contents?  
 This tool lets you decrypt and read them easily.
 
----
 
-To access it, go to **UnityNeuroSpeech → Tools → Decode Encoded**.
+Go to **UnityNeuroSpeech → Tools → Decode Encoded**.
+
+Important setting:
 
-Note about the **Key to encrypt** field:  
-You must use the same key you specified in your `AgentBehaviour` script.
+- **Key to encrypt**: You must use the same key you specified in your `AgentBehaviour` script.

docs/unity/useful-tools.md

@@ -14,7 +14,7 @@ theme:
     - search.suggest
     - header.autohide
 nav:
-  - Quickstart: index.md
+  - Getting started: index.md
   - Unity:
       - Steps to make it work: unity/steps-to-make-it-work.md
       - Agent API: unity/agent-api.md