Release v1.1.0: The Autonomous Loop 🚀

Release v1.1.0: The Autonomous Loop 🚀

v1.1.0 is a major milestone: The Gemini Native Agent is now an officially working C++ autonomous coding agent. We are no longer just generating code. From a single prompt, the agent can now natively write, build, execute, and visually verify its own applications directly on the host OS.

✨ What's New

• The Autonomous Coding Loop: You can now issue a single command (e.g., "Write a C++ GUI that says 'Hello World' in red, compile it, run it, and confirm the text is red"). The agent autonomously handles the file creation, invokes the g++ compiler via the shell tool, launches the .exe as a background process, and captures a screenshot of the result to verify its own success.
• Recursive Process Tree Vision: The take_screenshot tool was completely rewritten to use WinAPI's Toolhelp32Snapshot. Instead of grabbing the whole desktop, it maps child-process trees (e.g., cmd.exe -> hello_world.exe) to guarantee perfectly cropped bounding boxes of the generated GUI windows.
• Execution Barrier (Anti-Stutter): Rebuilt the internal data pipeline to perfectly interleave LLM Tool Call states with the SQLite API memory. A C++ execution barrier intercepts and discards duplicate tool calls from the model, forcing sequential, verified reasoning and permanently preventing infinite API loops.
• Modular Architecture: The codebase has been completely refactored into domain-specific subdirectories (/agent, /tools, /data, /settings) for cleaner scaling.

🛠️ Toolset Capabilities

• File System Operations: write_file, read_file, and list_directory allow the agent to read and manipulate the workspace.
• Sub-Process Execution: execute_shell_command spawns asynchronous, memory-safe QProcess instances to compile code, run scripts, or manage background servers.
• Computer Vision: take_screenshot captures cropped screenshots of active GUI windows for visual self-verification.
• Network & Deployment: upload_ftp for remote deployments, fetch_webpage for reading external DOMs/docs, and git_manager for batched version control workflows.

📦 Installation

1. Download the GeminiNativeAgent_v1.1.0_Setup.exe asset attached below.
2. Run the installer (Zero-setup bundled runtimes for MinGW, Python, and Node are included so the agent can compile code out-of-the-box without relying on your system PATH).
3. On first launch, enter your Google Gemini API Key in the Settings menu.

🚀 Gemini Native Agent v1.0.0 (Initial Release

🚀 Gemini Native Agent v1.0.0 (Initial Release)

Welcome to the official v1.0.0 release of the Gemini Native Agent.

This release marks the transition of the project from a development build into a fully standalone, enterprise-grade End User Computing (EUC) application. Built entirely in native C++ and Qt6, this agent communicates directly with Google's Gemini REST API to provide a secure, stateful, and autonomous local coding assistant.

✨ Key Features in this Release

🧠 Native Function Calling & Autonomy

• Direct REST API Integration: Bypasses Node.js and Python wrappers for a lightning-fast, zero-dependency QNetworkAccessManager pipeline.
• Autonomous "Hands": The agent can natively write files, compile C++, run Python scripts, and execute background shell processes using QProcess.
• Silent "Eyes": The agent can autonomously read local files, list directories, and fetch webpages without interrupting your workflow.
• Computer Vision: Integrates with the Windows API to capture specific application windows, allowing the agent to visually debug its own UI code.

🛡️ Zero-Trust Security Sandbox

• Human-in-the-Loop Intercepts: All destructive actions (writing files, executing shell commands) are halted by a thread-blocking QMessageBox. No code executes on your machine without explicit physical approval (Enter/Esc).
• Path Traversal Protection: Operations are strictly locked to the designated workspace directory.

💾 Stateful Architecture

• SQLite Session Persistence: Conversations and contextual memory are safely stored in a local .db file using QSqlDatabase, allowing you to seamlessly hot-swap between multiple development projects.
• Optimized Token Usage: Utilizes Google's previous_interaction_id to offload conversation history to the server, dramatically reducing input token costs during long debugging sessions.

🔒 Secure Credential Management

• Encrypted Vault: Your Gemini API Key and environmental variables (like GitHub PATs) are stored securely in the native OS registry via QSettings, eliminating hardcoded vulnerabilities.

---

📦 Installation Instructions

This release includes a standalone, compiled Windows installer. No external frameworks (like Node.js or Python) are required to run the application.

1. Download GeminiAgent_Setup_v1.0.0.exe from the Assets section below.
2. Run the installer and follow the standard Windows setup wizard.
3. Launch the application from your Start Menu or Desktop shortcut.
4. On the first launch, you will be prompted to enter your Google Gemini API Key.
5. Create a new session and start issuing commands!

---

If you encounter any issues or bugs, please open an Issue on this repository. Enjoy building with your new autonomous local agent!