LocalAI Release 0.0.4

LocalAI Release Notes

Version 2.0 - The Intelligence Update

Release Date: January 20, 2026

LocalAI 2.0 represents a major architectural leap, introducing a complete Retrieval-Augmented Generation (RAG) pipeline. This release transforms the tool from a simple connector into a context-aware coding assistant that understands your specific project structure.

🌟 New Features

🧠 Advanced RAG Pipeline

• Semantic Search: Queries are now processed for intent and meaning, not just keyword matching. The system finds relevant code even if you use different terminology.
• Dependency Awareness: The search engine now builds a graph of your code's structure. It "reads" your code to understand inheritance, method calls, and references, surfacing deeply related files.
• Conversation Memory: LocalAI now remembers your previous questions in the session. You can ask follow-up questions like "How can I optimize that?" or "Refactor this class" without restating context.

⚡ Performance & Experience

• Real-Time Streaming: Responses from Gemini now stream in word-by-word. No more staring at a loading spinner waiting for the logic to generate.
• Background Indexing: Indexing large projects happens asynchronously. You can continue working in the Unity Editor while LocalAI learns your codebase in the background.
• Sub-100ms Search: New vector clustering optimizations ensure instant answers, even in projects with 10,000+ scripts.

🎯 Accuracy & Feedback

• Relevance Feedback Loop: A new rating system allowing you to mark results as helpful or not. LocalAI learns from this feedback to rank future search results better for your specific needs.
• Quality Test Suite: Comprehensive automated testing ensures the retrieval engine maintains high accuracy and precision.

🛠️ Improvements & Fixes

• UI Overhaul: Cleaned up the Editor Window interface with better displaying of code blocks and search results.
• Stability: Fixed compilation errors in the service layer and improved error handling for network timeouts.
• Gemini Integration: Enhanced system prompts for better Unity-specific code generation (Unity 2021.3+ best practices).

📦 Upgrade Instructions

1. Import the new LocalAI.unitypackage.
2. Navigate to Tools > LocalAI > Settings.
3. Click Rebuild Index to generate the new semantic vector store for your project.

Release 0.0.3– Local AI Unity Tool (Experimental)

Context-Aware Scene Intelligence & Stability Pass

This release focuses on deeper Unity Editor awareness, stricter safety limits, and improved AI reliability.
The assistant now behaves more like a senior Unity developer working directly inside the Editor.

New Features

Live Unity Editor Awareness

• Automatically reads selected C# scripts, including full file contents
• Analyzes selected GameObjects
• Attached Components
• Tags and Layers
• Hierarchy position
• Inspects Prefabs, including root components and visual structure

Smart Context System

• Recursively reads Components, Serialized Fields, and Script content
• Dynamically builds the most relevant context while staying within safe limits

Scene Analysis Report

• Flags performance risks (polycount, draw calls)
• Identifies cleanup opportunities (empty objects, missing scripts)
• Reports platform-specific warnings for Mobile and VR

Accuracy Improvements

• AI tuned to behave as a Senior Unity Expert
• Prioritizes standard Unity patterns and best practices
• Avoids expensive APIs such as Find calls in Update
• Emphasizes performance, safety, and maintainability

Safety & Stability

• Dynamic Context Budgeting based on configured Context Size
• Prevents editor freezes and crashes from oversized selections
• Visual context usage feedback with real-time progress bar
• Warnings when files or selections are truncated
• Automatically disables Ask and Generate actions when limits are exceeded

Performance & Maintenance

• Removed debug spam caused by logging every token to the console
• Centralized all AI prompts into AIPrompts.cs
• Improved editor responsiveness and stability

Release v0.0.2 – Local AI Unity Tool (Experimental)

Release – Local AI Unity Tool (Experimental)

This release focuses on improving usability, response quality, and flexibility between local and cloud-based AI models. The tool continues to evolve as an experimental Unity developer assistant.

What's New

Ask Button (NEW)

• Introduced a dedicated Q&A mode for asking general Unity-related questions without needing code context.

Local AI 2.0

Improved Mistral 7B local inference with:

• Expert-level system prompting
• Repetition penalties to reduce looping and low-quality responses
• More consistent and relevant Unity-focused answers

Local + Cloud AI Support

Users can now choose between:

• Offline local inference (Mistral 7B)
• Cloud APIs (Gemini, OpenAI, Claude)

This allows flexibility based on hardware, privacy, and quality needs.

Selectable Responses

• AI-generated responses are now fully selectable, enabling partial text copying for easier reuse in code or documentation.

UI Improvements

• Cleaner layout, better spacing, and smoother interaction for a more focused developer experience.

Notes

• This project is still experimental.
• Local model performance depends on system resources.
• Feedback and contributions are welcome.

Local AI Unity Tool 0.0.1

🚀 v0.0.1 - Initial Release

Welcome to the first official release of the Local AI Assistant for Unity! This tool brings the power of Large Language Models directly into your Unity Editor workflow, with a focus on privacy and performance.

---

✨ Key Features

• 🔒 100% Private & Offline: All processing happens locally. No data ever leaves your machine. All inference is performed locally using your CPU or GPU.
• ⚡ Native Performance: Integrated with llama.cpp for high-performance inference, supporting Apple Silicon (Metal) and Windows (AVX2).
• 🛠️ Unity-Native UI: Built entirely with Unity's UI Toolkit, ensuring a seamless look and feel that matches the Editor.
• 🎯 C#-Focused Generation: Fine-tuned logic ensures the AI prioritizes C# and Unity-specific code, avoiding irrelevant language outputs.
• 📦 Automated Setup: One-click installation for native binaries and an easy-to-use model downloader.
• ⚙️ Highly Configurable: Adjust context window sizes (up to 8K) and response lengths to fit your hardware and requirements.

---

🏗️ Technical Stack

• Model: Mistral 7B Instruct (v0.1) in GGUF format (Q4_K_M).
• Backend: llama.cpp ARM64/x64 native libraries.
• Integration: Safe P/Invoke marshalling in C#.
• UI Framework: Unity UI Toolkit (UXML/USS).

---

🚀 Quick Start

1. Open Tools → Local AI → Install Native Libraries to setup binaries.
2. Open Tools → Local AI Assistant → Settings to download the Mistral model.
3. Start chatting with your AI assistant using the main window.

---

Built with ❤️ for the Unity Community.