Skip to main content
Describe what you want and Reagent picks the tool: one set works on items in your REAPER project, another on audio files on disk.

​
What You Can Ask For

Reagent shows these tool names in the chat while it works. The file tools also work on a Wwise project’s source files.
For items, loudness, silence and transient levels account for the item volume, the take volume and the average level of the take volume envelope. Muted items count as silent. Levels don’t include fades, track volume, or effects on the track or take. Analyze spectrum and Classify items read the source audio as it is.

​
Silence

  • Items: Reagent reports which items are effectively silent (99% or more of the item below -60 dB), not where the silence sits inside them.
  • Files: Reagent lists each silent stretch with its start and end time, plus the total. By default, silence means below -60 dB for at least 100 ms.
To find the gaps inside an item, ask Reagent to check its source file, as in the second example. Times then count from the file’s start, not project time, and levels ignore item and take volume. To cut silence out of items, see REAPER Tools. That tool shows as Remove item silence in the chat and Trim item silence on the approval card.

​
Transients and Peak Markers

Reagent reports the time and level of each hit, attack or other onset.
  • Items: By default, each item’s threshold sits about 20 dB below its average (RMS) level, and hits must be at least 50 ms apart.
  • Files: By default, hits must be at or above -40 dBFS and at least 600 ms apart. That gap skips fast repeats such as footsteps or a roll, so ask for a shorter one when hits come close together.
Ask for other values, such as “ignore anything below -30 dB” or “hits can be as close as 20 ms apart”. Mark peaks on item (Mark peaks on the approval card) runs its own detection, separate from transient analysis. It splits each item at quiet gaps and marks each sound once with a “Peak” take marker, at a hit’s attack or a swell’s loudest point. Sounds with a steady level get no marker. Like other project changes, it asks for approval and can be undone in REAPER. It doesn’t mark every transient, so to find every hit, ask for transients.

​
Loudness (RMS)

Reagent measures the average RMS level, in dB, of an item, a file or a section of a file. For several items, it lists them quietest first. RMS isn’t LUFS, the perceived-loudness measure used in delivery specs. To bring takes to a LUFS or peak dB target, use Volume Leveling.

​
Frequency Spectrum

REAPER items only. Reagent reports each band’s energy relative to the loudest band, which band dominates, and whether the item is empty.
  • Reagent analyzes a 16 kHz copy of the audio, so nothing above 8 kHz, such as air or hiss, is measured.
  • Items shorter than about a quarter of a second always read as empty. For very short sounds, such as clicks or UI blips, ask for RMS instead.

​
UCS Classification

Classify items suggests UCS (Universal Category System) category codes, called CatIDs, such as VEHCar (Vehicles, Car), for items in your REAPER project. On a Mac, classification needs Apple silicon and macOS 14 (Sonoma) or later. On an Intel Mac, or on macOS 12 or 13, Classify items, the UCS Renamer skill and the track content checks below fail. Windows and the other tools on this page aren’t affected. See System Requirements.
  • Runs on your computer with a custom-trained model that ships with Reagent. Your audio isn’t uploaded.
  • REAPER items only. To classify files, add them to a REAPER project first.
  • Listens to up to 10 seconds of each item, trying to skip leading silence on longer items.
  • Suggests the top 5 CatIDs per item, with confidence scores. Ask for up to 20, or for confident ones only.
  • Also reads track content. Reagent uses classification to tell what a track holds, such as dialogue, effects, ambience or music, instead of guessing from file names.
Suggestions stay in this chat’s working memory, not in your project. In the same chat, the UCS Renamer skill can give the takes UCS names (not the files on disk).
Classification returns lowercase codes, such as vehcar, so ask Reagent to look them up for the standard spelling (VEHCar), category and subcategory. The model knows 660 of the 753 CatIDs in the UCS list that ships with Reagent. If the right one isn’t suggested, describe the sound and ask Reagent to use the UCS Category Lookup skill, which searches the full list.

​
Reading Metadata

Extract metadata reads WAV, MP3, FLAC, Ogg Vorbis and AIFF files. It returns technical details, such as duration, sample rate, bit depth, channels and codec, plus tags in these standards:

​
Writing Metadata

Reagent can’t write metadata into audio files yet. If it offers to, the write fails and your files stay as they were. To embed UCS or other tags, use a separate metadata editor.

​
Where the Analysis Runs

The analysis, classification and metadata tools on this page run on your computer and don’t upload your audio. Their results go to the AI model as text with your conversation, including levels, times, band values, waveform outlines, category suggestions, the metadata they read and file paths. See Privacy & Security. Dialogue transcription is different: it uploads audio. See Dialogue.