Asryx

v1.1.0 2026-05-25

Mostly architectural changes.

The biggest win is ripping out the external CLI piping logic I used as a initial PoC, and linking whisper.cpp directly as a library, running the transcription entirely in-process.

On the runtime side, the state handling is a lot more robust now. The recorder termination sequence properly escalates from SIGINT down to SIGTERM and SIGKILL if the process gets stubborn, which prevents orphaned recording processes.

It also handles empty audio gracefully with a direct no output fallback instead of silently failing, and dumps actual error logs if the clipboard or transcription fails.

A new language configuration option was added as well, to explicitly set a target language to speed things up instead of being forced into Whisper's auto-detect.

Also, The installation scripts got a major overhaul.

Changes since v1.0.0

• a8b30e2 fix runtime state handling (#3)
• e2d2813 use one source of truth for consts (#2)
• aac366a use direct bindings (#1)
• 647d4d0 show "no output" for empty transcripts

Pull requests

• #1: use direct inference
• #2: use one source of truth for consts
• #3: fix runtime state handling

Authors

• rccyx (4 commits)

Compare

https://github.com/rccyx/asryx/compare/v1.0.0...HEAD