Remote ArcREMOTE ARC
Turn recordings into text and subtitles
Transcribe meetings, interviews, lectures and video audio into readable text and timed SRT/VTT files on your computer.
With Arc
Transcribe the recording I choose on my connected computer. Read the transcription skill, automatically prepare local whisper.cpp, FFmpeg and the base multilingual model if missing, and keep the audio on this device. Deliver a readable TXT/Markdown transcript and SRT/VTT subtitles. Review recognition samples and flag unclear speech; create meeting notes only if requested. Check current connections, permissions and dependencies first. Use real tools and continue from observations; claim only results actually observed or verified. Inspect uncertain external effects before any repeat. Report specific missing capabilities.
Arc requirements
The source recording/video, authorized file access and enough space for the local Whisper base model.
Arc boundaries
Recognition can make mistakes. Do not invent dialogue, speaker names or decisions; larger recognition models need size-aware download consent.
With Connect: example request
Transcribe the recording I choose on my connected computer. Read the transcription skill, automatically prepare local whisper.cpp, FFmpeg and the base multilingual model if missing, and keep the audio on this device. Deliver a readable TXT/Markdown transcript and SRT/VTT subtitles. Review recognition samples and flag unclear speech; create meeting notes only if requested.
With Connect: how the work proceeds
Inspect the audio track, language and length; extract suitable local audio. Transcribe, review samples and validate subtitle times before delivery.
Connect tools and permissions
read_skill · read_file · write_file · start_process. The source recording/video, authorized file access and enough space for the local Whisper base model.
What useful evidence looks like
Readable transcript and actual subtitle files, model used, source duration and uncertain passages.
Know the boundary
Recognition can make mistakes. Do not invent dialogue, speaker names or decisions; larger recognition models need size-aware download consent.