AI File Sorter 1.7.3 is now available

More reliable categorization across languages

One recurring issue with local models is not understanding, but consistency. The same concept can be phrased slightly differently depending on context or language. In 1.7.3, categorization is now done in a canonical internal form (English), and only then translated into the selected language.

This leads to:

• more stable categories across runs
• less drift when switching languages
• fewer near-duplicates caused by wording differences

Local AI that stays on task

LLMs sometimes return outputs that are too verbose, oddly formatted or "creatively interpreted." This was already handled in most cases, but certain edge cases could still affect results. In 1.7.3, the categorization pipeline has been hardened:

• improved prompt budgeting
• stricter output parsing
• more reliable category/subcategory extraction

The result is not "smarter AI," just more predictable behavior.

More resilient long runs

File systems aren't always perfectly "clean": some folders may be unreadable, some files may be locked, some paths may behave inconsistently. In 1.7.3, recursive scans take this into account, so that:

• problematic subfolders are skipped
• so the rest of the run continues

So a single edge case no longer interrupts a full sorting job.

OS-dependent and automated updates

The update system has been refined:

• separate update handling for Windows, macOS, and Linux
• better alignment with platform-specific packaging
• on Windows, installer downloads can be verified before launch

Stability improvements (the quiet kind)

Several smaller improvements reduce friction over time:

• safer handling of cached category labels
• more robust macOS runtime handling for local AI backends
• packaging and dependency improvements across platforms

None of these are, strictly speaking, vividly visible features, but they make the system behave more consistently.