Skip to content

Configuration

~/.config/rex/config.toml (rex config path). Settings in the app write through to this file and the app reloads it when edited by hand. Every key has a default; rex config show prints the effective configuration. schemaVersion = 1.

The Rex settings window

schemaVersion = 1

[general]
launchAtLogin = false
appVisibility = "menubar"        # menubar | dock | both
meetingsRoot = "~/Meetings"
logLevel = "info"                # debug | info | warning | error

[audio]
micDevice = "default"            # device name, or "default"
micPriority = ["default"]        # tried in order when micDevice is absent
micDelayMs = 150                 # mic delay before AEC
aec = true
keepAlive = true                 # silent keep-alive output so the tap never idles

[recording]
format = "m4a"
bitrateKbps = 96                 # per track
fragmentIntervalSec = 5

[transcription]
model = "parakeet-tdt-0.6b-v2"   # see Transcription › Models
beamRetry = true                 # second pass on empty voiced segments
vad = "fluidaudio"
diarization = true
maxSpeakers = 8
speakerThreshold = 70            # 40–90, % cosine distance; lower = more speakers
mode = "post"                    # "live" reserved

[models]
unloadPolicy = "idle"            # keep | idle | immediate
idleMinutes = 3
preloadOnMeetingStart = false
cacheDir = "~/Library/Application Support/Rex/Models"

[detection]
enabled = true
endGraceSec = 20                 # mic released this long = meeting ended; Rex offers to stop, never auto-stops
useWindowTitles = true           # needs Accessibility
autoDismissSec = 0               # 0 = the prompt waits for an answer

[[detection.apps]]
bundleId = "com.microsoft.teams2"
name = "Microsoft Teams"
rule = "ask"                     # ask | ignore (auto reserved)

[[detection.apps]]
bundleId = "Cisco-Systems.Spark"
name = "Webex"
rule = "ask"

[[detection.apps]]
bundleId = "us.zoom.xos"
name = "Zoom"
rule = "ask"

[ui]
playerTimeline = "levels"        # levels | speakers - player timeline rendering
showRecordingBar = true          # top-of-screen bar while recording
lowConfidencePercent = 60        # 30-90 - segments below this get the low-confidence caption

[calendar]
provider = "none"                # none | outlookSnapshot

[attendance]
provider = "none"                # none | teamsSnapshot

[speakers]
you = "You"
unknownLabel = "Speaker {n}"

[export]
auto = false
formats = ["md", "srt"]          # md | txt | json | srt | vtt
includeAudio = false
markdownTimestamps = false
markdownType = "meeting-transcript"
markdownTags = ["meeting"]

[notifications]
transcriptReady = true
sound = true
soundName = "Glass"              # any system sound name

[retention]
deleteAudioAfterDays = 0         # 0 = never

[updates]
check = true                     # check for a newer version every six hours
feedURL = "https://rex.pjtlabs.com/releases/latest.json"

[[workflow]]                     # see Workflows
name = "Summarize"
command = "~/bin/summarize.sh"
args = []
triggers = ["transcript-ready"]
enabled = true
timeoutSec = 600
env = {}

Reference

Every key, its label in the app's Settings window (- when it has no UI), what it does, the values it accepts, and the default.

[general] - Settings › General

Key In the app What it does Values Default
launchAtLogin Launch at login Registers Rex as a login item so it starts with your session. true / false false
appVisibility Show Rex in Where the app lives: menu-bar icon, Dock icon, or both. menubar, dock, both menubar
meetingsRoot Meetings folder Folder that holds one subfolder per meeting. any path, ~ allowed ~/Meetings
logLevel Log level How much detail rex.log records; Debug helps when investigating a problem. debug, info, warning, error info

[ui] - Settings › General

Key In the app What it does Values Default
showRecordingBar Show recording bar The floating bar at the top of the screen while a recording is running. true / false true
playerTimeline - What the player timeline draws: audio level bars or speaker turns. levels, speakers levels
lowConfidencePercent Low-confidence flag (Settings › Transcription) Segments below this confidence get the Low confidence caption; words a tenth below it get the dotted underline. 30–90 60

[audio] - Settings › Audio

Key In the app What it does Values Default
micDevice Microphone Input device to record; default follows the system default. device name or default default
micPriority Microphone priority Tried in order when the selected mic is unavailable; default means the system default. list of device names ["default"]
micDelayMs Mic delay Delays the mic track so it lines up with the system audio before echo cancellation. 0–500 ms 150
aec Cancel system audio echo (AEC) WebRTC echo cancellation of the meeting audio out of the mic track. true / false true
keepAlive Keep output device alive while recording Plays silence to the output device so the system-audio tap never goes idle. true / false true
referenceDevice - Ignored (kept for config compatibility): the AEC reference always follows the global system tap. - default

[recording] - Settings › Advanced

Key In the app What it does Values Default
format - Container for recordings. m4a m4a
bitrateKbps AAC bitrate per track Recording quality per track; 96 kbps is plenty for speech. 32–256, steps of 16 96
fragmentIntervalSec Fragment interval The file is written in fragments of this length, so a crash costs at most this many seconds. 1–60 s 5

[transcription] - Settings › Transcription

Key In the app What it does Values Default
model Model Speech-recognition model (download under Transcription › Models). installed ASR model id parakeet-tdt-0.6b-v2
beamRetry Retry empty segments with beam search Voiced segments that decode to nothing are retried with a slower beam-search pass (Parakeet models). true / false true
vad Voice activity detection Finds the stretches of speech in each track before transcription. installed VAD id fluidaudio
diarization Identify speakers on the system track (diarization) Splits the system track into distinct speakers. true / false true
maxSpeakers Maximum speakers Upper bound on how many distinct speakers diarization may find. 1–20 8
speakerThreshold Speaker sensitivity Cosine-distance cutoff for "same speaker" in percent; lower splits voices more readily. 40–90 70
mode - Transcription mode; live is reserved. post, live post

[models] - Settings › Transcription and Advanced

Key In the app What it does Values Default
unloadPolicy After transcribing What happens to the loaded models after a transcription: stay, unload after idle, or unload now. keep, idle, immediate idle
idleMinutes Idle minutes Idle time before models unload (with unloadPolicy = "idle"). 1–120 min 3
preloadOnMeetingStart Preload models when a meeting starts Loads the models during the meeting so transcription starts without a loading delay; holds a few GB of memory while recording. true / false false
cacheDir Model cache Where downloaded models are stored. any path ~/Library/Application Support/Rex/Models

[detection] - Settings › Detection

Key In the app What it does Values Default
enabled Detect meetings from watched apps Watches the apps below for microphone use and offers to record. true / false true
endGraceSec End meeting after mic released for The meeting counts as over once the app has released the mic this long; Rex offers to stop, never stops on its own. 5–300 s 20
useWindowTitles Name meetings from window titles Names the recording after the meeting app's window title (needs Accessibility). true / false true
autoDismissSec Auto-dismiss prompt after An unanswered Record / Ignore prompt disappears after this long; 0 means it waits. 0–300 s 0 (never)

[[detection.apps]] - Settings › Detection › Watched apps

Key In the app What it does Values Default
bundleId Bundle ID The watched app's bundle identifier. bundle id Teams / Webex / Zoom rows
name Name Display name used in prompts and meeting titles. text per row
rule Rule Ask shows the Record / Ignore bar; Ignore never prompts for that app. ask, ignore (auto reserved) ask

[calendar] - Settings › Advanced

Key In the app What it does Values Default
provider Meeting titles from Names recordings from the meeting on your calendar right now; outlookSnapshot reads New Outlook's local widget cache (read-only). none, outlookSnapshot none

[attendance] - Settings › Detection › Attendees

Key In the app What it does Values Default
provider Organizer & invited from After a recording, attaches organizer + invited emails (an attendance block in meeting.json) read read-only from the local Microsoft Teams app. Requires Full Disk Access. Best-effort — never affects the recording or its title. none, teamsSnapshot none

[speakers] - Settings › Speakers

Key In the app What it does Values Default
you Your name Labels the microphone track. text You
unknownLabel Unknown speaker label Pattern for unnamed speakers on the system track; {n} becomes 1, 2, 3… text with {n} Speaker {n}

[export] - Settings › Export

Key In the app What it does Values Default
auto Export automatically when a transcript is ready Writes the chosen formats into the meeting's exports/ folder after each transcription. true / false false
formats Formats Which formats to write. any of md, txt, json, srt, vtt ["md", "srt"]
includeAudio Include a copy of the audio Copies the recording into exports/ too. true / false false
markdownTimestamps Include timestamps in Markdown Prefixes each Markdown segment with [HH:MM:SS] (applies to transcript.md and .md exports; JSON/SRT/VTT always carry timing). true / false false
markdownType — type value written into the Markdown YAML frontmatter. Empty string omits the field. string "meeting-transcript"
markdownTags — tags list written into the Markdown YAML frontmatter (transcript.md and .md exports). Empty list omits the field. list of strings ["meeting"]

[notifications] - Settings › Notifications

Key In the app What it does Values Default
transcriptReady Notify when a transcript is ready Posts a notification when a transcription finishes. true / false true
sound Play a sound Plays a sound with the notification. true / false true
soundName Sound Which system sound. any system sound name Glass

[updates] - Settings › General

Key In the app What it does Values Default
check Check for updates Fetches the latest version number every six hours; a newer one offers an Install button in the menu bar and Settings. true / false true
feedURL - Where the version manifest lives; change it only to test a mirror. URL https://rex.pjtlabs.com/releases/latest.json

[retention] - Settings › Advanced

Key In the app What it does Values Default
deleteAudioAfterDays Delete audio older than Moves old recordings' audio to the Trash (see Retention); 0 keeps audio forever. 0–3650 days 0 (never)

[[workflow]] - Settings › Workflows

Key In the app What it does Values Default
name Name Display name of the workflow. text -
command Command Executable to run with the meeting folder as its last argument. path -
args Arguments Extra arguments, before the meeting folder. list of strings []
triggers Triggers When it runs; error fires when another workflow fails, manual adds a button to every meeting. recording-started, recording-stopped, transcript-ready, error, manual ["transcript-ready"]
enabled Enabled Disabled workflows never run. true / false true
timeoutSec Timeout A run still going after this long is killed (SIGTERM, then SIGKILL). 5–7200 s 600
env Environment Extra environment variables for the run. KEY = "value" table {}

Retention

With deleteAudioAfterDays > 0, Rex moves recording.m4a (and any audio copy in exports/) of meetings older than that to the Trash - never a hard delete - on launch, every six hours, and after the setting changes. Only meetings with a complete recording and a ready transcript qualify; the meeting being recorded, transcribed, queued, or running a workflow is skipped. meeting.json, the transcript and exports stay, so the meeting remains searchable; restore the file from the Trash and playback works again. The job refuses to touch a meetings folder that is the disk root, your home folder, a symlink or a folder without Rex meetings.