Mintu v0.2.3: Building Desktop Automation, Apple-Style Spring Physics, and Safe Evaluators with Tauri & Rust

Introduction

With the release of Mintu v0.2.3, our open-source Dynamic Island desktop companion advances from a passive interface into an active, automated desktop assistant.

Built with Tauri v2, Rust, and TypeScript, Mintu runs locally with offline models (via llama.cpp) or connects to modern cloud providers. This release tackles core user experience pain points, desktop automation, and runtime security.

What We Built in v0.2.3

1. Real Desktop Automation

Users can now delegate OS-level tasks directly to the model:

• VS Code Code Generation (create_code_file): Instruct the pet to write scripts in Python, TypeScript, Rust, or HTML and immediately launch them in VS Code.
• Media Integration (play_spotify & search_youtube): Direct playback commands for songs, albums, and video tutorials.
• Desktop Guidance (desktop_guide): Screen guidance and cursor positioning via native Win32 APIs (SetCursorPos).

2. Solving the Streaming Auto-Scroll Problem

In real-time token streaming interfaces, unconditionally setting scrollTop = scrollHeight traps the user at the bottom of the viewport.

In v0.2.3, we introduced user scroll intent detection:

• The scroll area tracks upward displacement (distanceFromBottom > 50).
• If the user scrolls up to inspect previous code snippets or responses, token streaming updates the DOM without dragging the viewport down.
• When the user scrolls back to the bottom or submits a new message, automatic following resumes.

3. Apple-Inspired Spring Morph on Music Events

Whenever media playback starts or a track changes, the island applies spring physics velocity impulses:

• springWidth and springHeight expand fluidly before settling into equilibrium.
• A radial background glow samples the dominant tones of the active track's album cover art.

4. Expanded Model Catalog

• Local: Qwen 2.5 Coder 1.5B for local code authoring and DeepSeek-R1 Distill Qwen 1.5B for step-by-step reasoning.
• Cloud: Anthropic Claude 3.7 Sonnet, Google Gemini 2.0 Flash, DeepSeek-R1 Reasoner, and OpenAI o3-mini.
• Every catalog entry now indicates tool-calling compatibility so users know which models can execute desktop actions.

5. Comprehensive Security Hardening

• Replaced all legacy new Function math execution with a pure, recursive-descent math parser supporting operators (+, -, *, /, %, ^, **), unary signs, and division-by-zero protection.
• Hardened file sandboxing to strictly use Rust internal directories.
• Enabled production Content Security Policy (CSP) blocking external eval.

Open Source & Community

Mintu is free and open-source under the MIT license.

• GitHub: github.com/HuseynDeveloper/Mintu
• v0.2.3 Release: github.com/HuseynDeveloper/Mintu/releases/tag/v0.2.3