Interly: tất cả phiên bản và phiên bản cũ

Interly 0.6.0 — Deep Windows Integration
Interly 0.6.0 is a major expansion of what the agent can actually do on a Windows computer.
Version 0.5.1 established the standalone distribution path, the canonical interly command, bounded automatic approvals with set-free, and the resilient update mechanism. 0.6.0 moves substantially deeper into Windows itself.
Interly is no longer limited to reasoning about Windows through commands, processes, files and an isolated browser. It can now inspect the visible desktop, understand what is on screen, identify native interface controls, manipulate application windows, use the mouse and keyboard, interact with the clipboard, and coordinate these capabilities through a stronger permission and planning system.
This release is an important step towards Interly becoming a genuine Windows computer agent rather than simply an AI exposed to a collection of Windows tools.

Windows becomes part of the agent environment
The largest change in 0.6.0 is that Interly can now perceive and interact with the ordinary Windows desktop.
Interly can enumerate visible top-level windows and receive exact information about them, including their native handles, process IDs, titles and screen rectangles. Once a window has been identified, Interly can focus it, minimise it, maximise it, restore it, move it or resize it.
Window operations are revalidated before execution rather than blindly trusting stale window information.
This means Interly can begin working with applications as visible pieces of the desktop environment instead of treating every program as nothing more than a process or executable.

Full desktop and application screenshots
Interly can now capture the Windows desktop directly.
It supports:
- full virtual-desktop screenshots, including multi-monitor desktop space;
- screenshots of an exact selected window;
- PNG output for later inspection or OCR;
- bounded screenshot use within an individual request.
Screenshots become an important bridge between Interly's existing system access and the graphical Windows environment.
Instead of assuming what a program should currently display, Interly can inspect what the computer is actually showing.

Built-in OCR
0.6.0 introduces bundled optical character recognition.
Interly can analyse an approved local screenshot or image and extract visible text along with its approximate bounding boxes.
This gives the agent another way to understand software that does not expose useful structured information.
For example, Interly can conceptually move through a loop such as:
capture window → read visible text → locate relevant area → decide next action
The OCR engine is included in the standalone distribution rather than requiring the user to separately configure an OCR installation.

Windows UI Automation inspection
Interly now integrates with Windows UI Automation.
It can inspect the controls exposed by the foreground application and return information such as control names, control types, automation identifiers and screen rectangles.
Importantly, inspection does not automatically activate those controls.
This creates a distinction between perception and action:
inspect → identify → decide → request permission → interact
That distinction is central to the direction of Interly.
Rather than immediately giving an AI arbitrary mouse access and hoping it clicks correctly, the system now has mechanisms for identifying what exists on screen before attempting an interaction.

Guarded mouse control
Interly can now perform generic desktop mouse actions.
Supported interaction includes mouse movement, clicking, double-clicking and bounded wheel scrolling.
These actions remain permission-aware.
The intended desktop interaction model is also explicitly biased towards using known information—such as a UI Automation rectangle, OCR position or previously returned screen coordinate—rather than inventing arbitrary coordinates.
Generic mouse control therefore acts as a fallback interaction primi