Settings reference
Generated from the Settings pages and the configuration schema of the current build (source revision e6fd2c7): 168 schema keys, 225 anchors.
v0.30.0-beta.2 and the stable v0.22.1 still have the older tabbed Settings (General, Modes, Commands and so on), and do not have the controls marked newer than v0.30.0-beta.2.Settings has six pages and a set of Tools windows. Every setting below has a stable link of the form /samsara/docs/settings/#config.key, which is what the question-mark help in the app points at.
How Settings works
- Each page shows its everyday rows first. Choose Advanced on a page to reveal the rest.
- Search matches labels, help text and configuration keys, and opens the advanced section it finds a match in.
- The question mark beside a row opens its explanation. One explanation is open at a time; Esc or Close help dismisses it.
- Edits take effect when you choose Apply. The Tools windows have their own rules: the Command library and Alarms save as you go; Music and the other tool preferences use Apply.
Results update as you type. Jump to a page:
No setting matches that filter.
General
Make Samsara comfortable for everyday use.
Interface size
Size text and controls for comfortable use.
Raise it if text and controls are hard to read or hit. Standard is 100%, Large 115%, Extra large 130%. It scales the whole app and applies after a restart.
Theme
Choose Samsara's appearance.
Dark is the default; light and system are available.
Microphone
Choose where Samsara listens.
Change it when Samsara listens to the wrong device. System default follows Windows; Refresh rescans after you plug something in. Refresh is refused while a recording is in progress.
Language
Choose the language you dictate in.
Set it to the language you speak so recognition uses the right model.
Contextual hints
Show guidance as you learn.
Turn hints off once you know your way around. Reset hints (immediate, no Apply) lets every hint fire again.
Start at sign-inAdvanced
Launch Samsara when you sign in to Windows.
Turn on if you want Samsara ready after every Windows sign-in.
Listening indicatorAdvanced
Show when Samsara is listening.
The small marker that shows when Samsara is listening. Turn it off if you find it distracting.
Indicator positionAdvanced
Choose the indicator's starting position.
Floating windowsAdvanced
Recover the indicator or dictation preview.
Live surfaceAdvanced
Try the experimental listening display; off by default.
Preview fades afterAdvanced
Delay before the idle dictation preview fades.
Faded preview visibilityAdvanced
Choose how visible the idle preview stays.
Streaming captureAdvanced
Use the existing dedicated CapsLock capture.
Shows live partial text while you hold the key, on a dedicated CapsLock capture. Off by default.
Dictation
Choose how you start speaking and deliver text.
Session button
Enable hands-free sessions from your shortcut.
Turn this on when you want to use hands-free sessions. You can then start a session with the session button you choose in Shortcuts.
The master switch for hands-free sessions. Also choose the button and its hold or toggle behaviour on the Shortcuts page; only toggle gives a session that stays open.
Wake activation
Start a session by saying your wake phrase.
Turn this on to begin a session by saying your chosen wake phrase. Turn it off when you want Samsara to listen only after a button press.
Off by default. Turn it on to start a session by voice; the Wake toggle shortcut (Ctrl+Alt+W by default) flips it from the keyboard.
Speech pace
Allow time for pauses in your speech.
Choose the pace that gives you enough room to pause naturally while speaking. A slower pace is useful when you need more time between words or phrases.
Choose Relaxed or Unhurried if you pause a lot. Relaxed multiplies the silence timings by 2.5, Unhurried by 4. Custom lets you set each timing in the config file.
Cleanup mode
Choose how dictated text is cleaned up.
Choose how much cleanup Samsara applies before it delivers your words. Pick the least processing that gives you text you are comfortable sending.
Clean removes fillers such as um and uh, fixes spacing and capitalises; Verbatim leaves the transcript exactly as heard.
Insertion / output
Paste completed text, or keep it on the clipboard.
Choose whether completed dictation is pasted into the active text box or left on the clipboard. Use the clipboard option when you want to review or place the text yourself.
Default on: finished text is pasted into the active window and your clipboard is restored. With it off, the build we read does not insert the text at all (it keeps the draft), despite the in-app wording about the clipboard.
Commands while dictating
Recognise commands in a dictation session.
Turn this on if you want spoken commands to work while you are dictating normally. Turn it off if words that sound like commands should always be entered as text.
Off by default. Turn it on to let spoken commands work during ordinary dictation. Turn it off if words that sound like commands should always be typed.
Wake phraseAdvanced
Choose the trained activation phrase.
Choose the phrase you will say to wake Samsara. Use a phrase you can say comfortably and that is unlikely to appear in ordinary conversation.
Choices are jarvis, hey jarvis, computer, hey computer, samsa, hey samsa. Only the Jarvis phrases have a dedicated detector model; the others rely on the speech recogniser and are less reliable.
Recognition modelAdvanced
Choose the speech recognition model.
Choose the speech-recognition model Samsara uses for your dictation. Larger models can be more accurate but need more computer resources and may respond more slowly.
Bigger models are more accurate and slower. Base is the recommended starting point; the model downloads the first time it is used.
Compute deviceAdvanced
Choose the hardware used for recognition.
Choose the hardware Samsara uses to recognise your speech. Leave the current choice in place unless you know another device is available and works better for you.
CPU always works. CUDA appears only when all ten GPU runtime files are present, and needs a restart. See the CUDA page.
Precision metadataAdvanced
Stored diagnostic precision; the loader chooses actual precision.
This shows diagnostic information about how recognition is configured. You do not need to change it for ordinary dictation.
PerformanceAdvanced
Balance recognition speed and accuracy.
Choose whether you prefer faster results or more time for recognition to refine them. Try a different setting only if the current speed or accuracy is not working for you.
Trailing spaceAdvanced
Add a space after inserted dictation.
Turn this on to add a space after Samsara inserts completed text. Turn it off if you usually dictate into places where you want to control spacing yourself.
CapitalisationAdvanced
Capitalise dictated text automatically.
Turn this on to start dictated sentences with a capital letter automatically. Turn it off if the program you dictate into handles capitalisation for you.
Number formattingAdvanced
Format spoken numbers as digits.
Turn this on to turn spoken numbers into digits when Samsara is confident about them. Turn it off if you prefer numbers to stay as words.
Timeout after wakeAdvanced
Wait for speech after wake activation.
Choose how long Samsara waits for you to begin speaking after it hears your wake phrase. Increase it if you need extra time before starting your request.
How long Samsara waits for your request after hearing the wake phrase. Default 5 seconds.
Command pauseAdvanced
Wait before completing a command phrase.
Choose how long a pause tells Samsara that you have finished a spoken command. A longer pause gives you more time to speak; a shorter pause makes commands run sooner.
How long a pause ends a spoken command. Raise it if commands are cut off.
Dictation pauseAdvanced
Wait before completing a dictation phrase.
Choose how long a pause closes one piece of hands-free dictation. A longer pause keeps more words together, while a shorter pause can split your speech into smaller pieces.
How long a pause closes one chunk of hands-free dictation. The schema default is 0.65 s; a freshly written config uses 2.0 s.
Quick silence timeoutAdvanced
Adjust recognition silence timing.
Choose how quickly Samsara treats a quiet gap as silence while recognising speech. Change this only if normal pauses are being cut off or recognition waits too long.
Wake thresholdAdvanced
Adjust wake detection sensitivity.
Choose how readily Samsara responds to your wake phrase. A lower threshold may hear the phrase more easily, while a higher threshold can reduce accidental wake-ups.
Wake sensitivity. Default 0.20. Lower it if the wake phrase is missed; raise it if it fires by accident.
Continuous presetAdvanced
Keep the existing continuous insertion trigger.
Choose how an existing continuous dictation session commits speech. This setting is kept for people who already use continuous mode and does not change the primary dictation key.
Silence commits the continuous capture automatically; key waits for the Continuous commit shortcut.
Continuous silenceAdvanced
Silence before a continuous phrase completes.
Choose how much silence ends a phrase in an existing continuous dictation session. Increase it if you need longer pauses before a phrase completes.
Minimum speechAdvanced
Ignore speech shorter than this duration.
Choose the shortest speech Samsara should accept as a phrase. Increase it if brief sounds are being treated as dictation, or decrease it if short words are missed.
Microphone calibrationAdvanced
Choose automatic or manual calibration.
Choose whether Samsara adjusts microphone sensitivity automatically or uses a setting you provide. Automatic calibration is the usual choice unless you are tuning a known microphone problem.
Calibration multiplierAdvanced
Adjust automatic microphone calibration.
Adjust how strongly automatic microphone calibration responds to your environment. Change it gradually and test ordinary dictation before making another adjustment.
Manual thresholdAdvanced
Set the speech threshold for manual calibration.
Set the microphone sensitivity yourself when you use manual calibration. Change it only if automatic calibration does not respond reliably in your usual environment.
Stop wordsAdvanced
Customise the session's stop words.
Add words you want to use to stop what Samsara is doing. Put one phrase on each line and keep the words easy to say in an emergency.
Emergency stop words. Default: halt, cease. Setting your own list replaces the built-in ones. A stop ends speech and pending work but keeps your draft and the session.
Exit phrasesAdvanced
Add phrases that leave a session.
Add extra phrases that should leave a hands-free session. Put one phrase on each line and avoid phrases you might say while dictating normally.
Extra phrases that leave a session, one per line. They add to the built-in ones (go to sleep, samsara sleep, sleep now, stop listening, exit hands free) and keep your draft.
Command debounceAdvanced
Avoid repeated activation from one press.
Choose how long Samsara ignores repeated button presses after one press. Increase it only if one press is accidentally starting more than one action.
Ignore repeat presses of the session button within this many milliseconds. Default 200.
Session timeoutAdvanced
End a session after inactivity.
Choose how long an unused hands-free session stays open before it ends. A longer timeout gives you more time between requests, but leaves the session available for longer.
A session ends after this many seconds of silence. Default 300.
Cancel windowAdvanced
Allow time to cancel commands heard during dictation.
Choose whether Samsara briefly waits before it runs a command heard while you are dictating. Use the extra time when you want a chance to say no or cancel.
A command heard inside dictation waits this long, showing a chip, so you can say no. Default 3 seconds, maximum 5, 0 turns it off. Commands that only read never wait.
Command miss limitAdvanced
Stop after repeated unrecognised commands.
Choose how many unrecognised requests a hands-free session accepts before it stops. Lower it if you want a session to end quickly after repeated misunderstandings.
A session ends after this many unrecognised requests in a row. Default 5.
Command wordAdvanced
Run a command by speaking this prefix first.
Choose the word you say before a spoken command. Pick a word that is easy to remember and unlikely to occur in the text you dictate.
Say this word first to force a phrase to be treated as a command.
Wake profilesAdvanced newer than v0.30.0-beta.2
Pair a wake phrase with an app, and choose its send word.
Each profile pairs a wake phrase with an app. Saying the phrase focuses that app and starts dictating into it; saying the profile's send word at the end of an utterance presses Enter there. Defaults: okay claude (claude.exe, send word over, on), activate hermes (send word send, on), activate warp (warp.exe, send word over, off).
Audio
Manage sounds, speech and captions in one place.
Output device
Choose where Samsara plays sound.
Where Samsara plays its sounds and speech. Leave on the system default unless you route sound elsewhere.
Feedback sounds
Hear signals for recording and actions.
Beeps for recording start, stop and actions.
Sound volume
Adjust feedback volume and test it.
Sound theme
Choose the feedback sound set.
Text-to-speech
Let Samsara speak responses aloud.
Off by default. Turn it on to have Samsara speak replies and notices.
Ava captions
Show Ava's responses as text.
Shows Ava's replies as text in a caption panel.
Mute Ava's voice
Silence Ava while keeping other speech available.
Silences Ava's voice but keeps captions; needs captions on. Sounds and command acknowledgements are unaffected.
Spoken noticesAdvanced
Choose which session notices are spoken.
Questions (the default) speaks only when Samsara is waiting for your answer; everything speaks every notice.
Speech engineAdvanced
Choose local Windows speech or a network voice.
Edge uses Microsoft's online voices and needs the internet; WinRT uses the offline Windows voices.
VoiceAdvanced
Choose a voice and hear a sample.
Speech speedAdvanced
Adjust spoken response speed.
Speech volumeAdvanced
Adjust speech independently of feedback sounds.
Speech during feedbackAdvanced
Remaining speech volume while feedback sounds play.
Session-end soundAdvanced
Play a sound when a session ends.
Plays a short sound when a hands-free session ends. Default on.
Spoken reply limitAdvanced
Limit the length of spoken acknowledgements.
Spoken acknowledgements are cut at this many characters. Default 50; 0 means no limit.
While recordingAdvanced
Choose how other apps' audio is handled.
Duck lowers other apps while you record (default); Pause stops media and resumes only what Samsara paused; Leave alone does nothing.
Ducked volumeAdvanced
Remaining other-app volume while recording.
Hands-free audioAdvanced
Reduce other apps during hands-free listening.
Hands-free capture volumeAdvanced
Remaining other-app volume while capturing speech.
Hands-free idle volumeAdvanced
Remaining other-app volume while listening.
Echo cancellationAdvanced
Experimental microphone echo cancellation.
Experimental and off by default. It reduces Samsara's own speech in the microphone by only a few percent.
Echo latencyAdvanced
Compensate for measured playback latency.
Ava
Choose how Ava helps, with permissions you control.
Ava runs newer than v0.30.0-beta.2
Choose Off, on this PC, or your cloud provider.
Off, On this PC (Ollama) or In the cloud. Cloud asks for your consent first, for that exact provider. Applying a route also turns off automatic corrections, web search and warm-up, so re-check them afterwards.
Effective route
Where Ava will handle your requests after Apply.
Memory retention newer than v0.30.0-beta.2
Choose whether Ava starts fresh, keeps the last conversation, or remembers facts you tell her.
Clear on close forgets everything; Keep last session keeps the last conversation; Ava remembers me keeps facts you tell her in a readable file. See the Ava page.
Editing
Allow Ava to edit what you just dictated.
Experimental. Say 'Ava, make that more formal'; Ava shows the change and you say apply to type it.
Ava interprets commands
After a normal command miss, Ava may choose one available command.
After a normal command miss Ava may pick one available command. Off by default and disabled while Ava is Off.
Ava consentAdvanced
Read how Ava handles local and cloud requests.
Cloud providerAdvanced
Choose the provider receiving cloud requests.
DeepSeek, OpenAI, Anthropic or OpenRouter. Changing it asks for consent again.
Cloud credentialAdvanced
Enter your provider credential and test the connection.
Your own key, stored on this computer and sent only to the provider you chose.
Cloud model overrideAdvanced
Use a particular provider model.
Leave blank to use the provider's built-in default model.
Web searchAdvanced
Allow supported providers to search the web.
Cloud with DeepSeek only. The question and recent conversation go to DeepSeek; answers are information only and never run commands.
Cloud timeoutAdvanced
Limit how long a cloud request can take.
How long a cloud request may take. Default 30 seconds.
Local modelAdvanced
Choose the installed model used locally.
The installed Ollama model used for command interpretation when Ava runs on this PC.
Keep model warmAdvanced
Keep the local model ready between requests.
Keeps the local model loaded between requests so answers come faster, using memory.
Warm at startupAdvanced
Send a small request at startup to prepare Ava.
Sends one small request at startup to load Ava. Default on for local, off for cloud because cloud warm-ups cost tokens.
Ava personalityAdvanced
Choose the style of Ava's answers.
Relaxed (default) or Strict answers.
Remembered turnsAdvanced newer than v0.30.0-beta.2
Limit the conversation retained as context.
How many turns of the current conversation Ava keeps as context, 5 to 500. Default 20.
Editing timeoutAdvanced
Limit how long an editing request can take.
How long an editing request may take. Default 12 seconds.
Local editing modelAdvanced
Choose the local model for editing.
Demo pacingAdvanced
Use slower editing animations for demonstrations.
Automatic correctionsAdvanced
Enable the separate automatic cleanup pass.
A separate cleanup pass over dictated text, using the route you chose. It also needs the editing permission, which the current build may not be able to grant from Settings (a known issue).
Correction modelAdvanced
Use this local model for automatic corrections.
Correct hold dictationAdvanced
Apply automatic corrections to held dictation.
Correct wake dictationAdvanced
Apply automatic corrections to wake dictation.
Correct streamingAdvanced
Apply automatic corrections to streaming dictation.
Repair disfluenciesAdvanced
Remove filler words and spoken self-corrections.
Shortcuts
Keep your essential actions within reach.
Primary dictation
Choose the dictation key and its behaviour.
Hold it while you speak. Default Ctrl+Shift. The mode beside it is hold, toggle or continuous.
Session button
Choose the hands-free button and its behaviour.
The hands-free session button. Default Right Ctrl; Mouse 4, Mouse 5, other modifier keys and F13 to F24 are available.
Ava held key
Hold a key while speaking to Ava.
Hold this key to talk to Ava. Default Right Alt. It does nothing unless Ava is on.
Cancel
Cancel the active recording.
Cancels the recording in progress. Default Escape.
Undo
Undo the most recent supported action.
Undoes the last supported action. Default Ctrl+Alt+Z.
Paste staged thought
Insert the thought waiting in your session.
Pastes the thought waiting in your hands-free session, the same as saying finish. Default Ctrl+Space.
Quick memo
Capture a short voice memo.
Starts a quick memo. Default Ctrl+Alt+M.
Corrections
Capture a correction; report from that window.
Opens correction capture; you can report a problem from that window. Default Ctrl+Alt+X.
Complete alarm
Mark the active alarm as completed.
Marks the sounding alarm as done. Default F7.
Dismiss alarm
Dismiss an alarm without completing it.
Dismisses the alarm without completing it. Default F8.
Command-only keyAdvanced
Capture one command without dictation output.
Hold it to say one command with no dictated text. Default Ctrl+Alt+C.
Wake toggleAdvanced
Start or stop wake listening.
Turns wake listening on and off. Default Ctrl+Alt+W.
Continuous toggleAdvanced
Start or stop the continuous preset.
Starts and stops the continuous preset. Default Ctrl+Alt+D.
Continuous commitAdvanced
Commit the current continuous capture.
Commits the continuous capture when the commit trigger is key. Default Ctrl+Space.
Correction report shortcutAdvanced
Keep the existing direct report shortcut.
The direct report shortcut. Default Ctrl+Alt+R.
Mouse suppressionAdvanced
Suppress browser-back when Mouse 4/5 is bound.
Stops the browser Back action firing when Mouse 4 or 5 is your session button. Default on.
Help
Set up, practise and recover when you need to.
Documentation
Find commands and usage guidance.
Setup / practice
Find tutorials and microphone or Ava guides.
Support / problem
Review diagnostics and get help with a problem.
Private backup / restore
Manage private configuration backups.
Updates
Check GitHub for updates; daily checks are optional.
Off by default. When on, Samsara asks GitHub for the latest stable release at most once a day. Nothing but the request is sent.
Stage timingAdvanced newer than v0.30.0-beta.2
Record how long each dictation step takes in a local file.
Turn this on to measure delays when tuning dictation performance. It saves timings locally without dictated text; leave it off unless you are debugging.
Records how long each dictation step takes, locally and without dictated text.
Diagnostic event logAdvanced newer than v0.30.0-beta.2
Save dictation results and troubleshooting details to a local file.
Turn this on when investigating missing or unexpected dictation results. The local file can include dictated text, so leave it off unless you are debugging.
Writes a local diagnostic log that can include dictated text. Leave it off unless you are debugging.
Benchmark recordingAdvanced
Save dictation audio and transcripts locally for review.
Saves dictation audio and transcripts locally for review. Off by default.
Tools windows
Open Tools from Settings or the tray. Each tool is its own window; the settings below are changed inside it.
Command library
Open it from Tools. Switch command packs on or off, manage profiles, and add, edit, test or delete your own voice commands. Pack switches need a restart of Samsara; your own commands apply immediately and are written to commands.json in the app folder.
command_packs.core
Core: Essential commands - undo, copy, paste, window snap, scroll. Always on, no switch.
command_packs.session
Unified Session: Toggle command mode recovery commands (e.g. retype that). Always on, no switch.
command_packs.experimental
Experimental: In-development features (workflow capture, etc.) - may be unstable. Off by default. Changing a pack needs a restart of Samsara.
command_packs.text-editing
Text Editing: Punctuation, formatting, bold/italic/underline, find, select. On by default. Changing a pack needs a restart of Samsara.
command_packs.window-management
Window Management: Move windows between monitors, cursor teleportation, movie mode. On by default. Changing a pack needs a restart of Samsara.
command_packs.window-cube-numbers
Window Cube Numbers: Bare one to nine switch windows while the cube is pinned. On by default. Changing a pack needs a restart of Samsara.
command_packs.browsers
Browsers: Open Chrome/Firefox/Edge, tab control, zoom, navigation. On by default. Changing a pack needs a restart of Samsara.
command_packs.word
Microsoft Word: Focused Word formatting, selection, and word count. On by default. Changing a pack needs a restart of Samsara.
command_packs.media
Media & Music: Spotify, play/pause, volume, next/previous track. On by default. Changing a pack needs a restart of Samsara.
command_packs.smart-home
Smart Home: Hyperion LED strip control, light effects and colors. Off by default. Changing a pack needs a restart of Samsara.
command_packs.3d-printing
3D Printing: FlashForge printer control - start, pause, status. Off by default. Changing a pack needs a restart of Samsara.
command_packs.stremio
Stremio: Stremio media player - pause, skip, fullscreen. Off by default. Changing a pack needs a restart of Samsara.
command_packs.screen-capture
Screen Capture: Screenshot, GIF recording, screen region capture. Off by default. Changing a pack needs a restart of Samsara.
command_packs.macros
Custom Macros: Going dark, focus mode, break time, morning routine. Off by default. Changing a pack needs a restart of Samsara.
command_packs.gaming
Gaming: Hold keys, jump, release all - game-specific controls. Off by default. Changing a pack needs a restart of Samsara.
command_packs.audio
Audio Devices: Switch between speakers, headset, earbuds. Off by default. Changing a pack needs a restart of Samsara.
command_packs.utilities
Utilities: Timer, web shortcuts, quick ask, calculator, notepad. Off by default. Changing a pack needs a restart of Samsara.
command_packs.smart-actions
Smart Actions: Brain dump, voice notes, AI agent routing. On by default. Changing a pack needs a restart of Samsara.
command_packs.tasks
Tasks: Add items to task lists by voice. On by default. Changing a pack needs a restart of Samsara.
command_packs.health
Health Tracking: Pain levels, medication logging, symptom tracking, health summaries. On by default. Changing a pack needs a restart of Samsara.
command_packs.alarms
Alarm Commands: Voice control for alarms - complete, dismiss, list, enable/disable. On by default. Changing a pack needs a restart of Samsara.
command_packs.accessibility
Accessibility: Show Numbers overlay, semantic clicking by voice. On by default. Changing a pack needs a restart of Samsara.
command_packs.discord
Discord: Needs a Discord webhook. Off by default. Changing a pack needs a restart of Samsara.
Alarms
Add, edit, toggle, test and delete recurring alarms. Alarm records save immediately; the two preferences below use Apply and Close.
alarms.enabled
Turns alarm reminders on or off.
alarms.nag_interval_seconds
How often a sounding alarm repeats until completed or dismissed, 15 to 300 seconds.
Health
A read-only view of the health log kept by the Health voice commands, plus CSV export and an optional medication dictionary. It changes no setting. Data stays in health_log.json in your Samsara folder.
Music
Name playlists, albums, artists or tracks once, then say 'playlist Morning'. You can also import your own Spotify playlists from Spotify's account data export (Privacy settings, Download your data): choose the ZIP or its folder; only files named playlist*.json are read, locally, and only playlists you own are kept. Nothing is saved until Apply and Close.
music_library newer than v0.30.0-beta.2
Names you can say to play a Spotify playlist, album, artist or track. Add, delete and import changes are held until Apply and Close.
Components
Install and manage optional parts. Choose Check what's available to contact GitHub; downloads are SHA-256 verified. The Gesture lane needs the Gesture control component.
gesture.enabled
Optional webcam hand-pose commands. Needs the Gesture control component.
Integrations
Connect optional services with explicit permission. Todoist is the only integration.
tasks.todoist.enabled newer than v0.30.0-beta.2
Off by default. When on, tasks you add by voice are also sent to Todoist's servers. Tasks are always stored on this computer first.
tasks.todoist.api_token newer than v0.30.0-beta.2
Your own Todoist API token. The field is hidden until Todoist is switched on. Use Test connection after pasting it.
Config-file keys
These keys have no control in Settings. They live in config.json in your Samsara folder (%USERPROFILE%\.samsara\). Close Samsara before editing it, and keep a copy first.
accessibility
accessibility.ava_captions_linger_s
Config-only caption timing override; the existing captions UI deliberately exposes enable and mute, not timing tuning.
accessibility.ava_captions_position
Placement state written by dragging the caption window; not a separate Settings preference.
agent_channel
agent_channel.allow_writes newer than v0.30.0-beta.2
CLI agents are read-only by default.
Default off: GPT and Claude questions run read-only. Switching it on lets the command line agents write files.
agent_channel.cwd newer than v0.30.0-beta.2
Working directory for subscription CLI agents.
The folder the command line agents run in. It must exist; if it does not, the failure is spoken as the CLI not being installed.
agent_channel.timeout_s newer than v0.30.0-beta.2
Maximum seconds to wait for a CLI answer.
agent_channel.voice_preamble newer than v0.30.0-beta.2
Spoken-response instruction sent with each CLI question.
audio
audio.input_sensitivity
Compatibility schema value with no current capture consumer; microphone calibration and speech threshold are the active controls.
ava
ava.catalog_answers newer than v0.30.0-beta.2
Local answers to command catalog questions.
Lets Ava answer 'what commands do I have for X' from the command catalog without calling any model.
ava_command_session
ava_command_session.shortlist_size
Interpreter candidate-budget tuning; the supported user choice is whether command-miss interpretation is enabled.
ava_invocations
ava_invocations
Config-only whole-utterance invocation vocabulary; kept outside Settings until an editor can validate session phrase collisions.
Whole-utterance phrases that switch to Ava: hey ava, so ava, oracle.
ava_memory
ava_memory.recall_sessions newer than v0.30.0-beta.2
Config-only limit for summaries recalled in a reply.
How many summaries are recalled in a reply, 0 to 30. Default 3.
ava_memory.sessions_kept newer than v0.30.0-beta.2
Config-only limit for dated summaries in assistant memory.
How many dated session summaries are kept in the memory file, 0 to 365. Default 30.
benchmark
benchmark.max_samples
Config-only benchmark storage cap; Help exposes collection consent while retaining the bounded default cap.
claude_voice
claude_voice.enabled newer than v0.30.0-beta.2
Desktop Claude reply reading is enabled by default; advanced config-only override.
Reads Claude desktop replies aloud after you send a message by voice. Default on.
claude_voice.reply_timeout_s newer than v0.30.0-beta.2
Advanced config-only timeout for a desktop Claude reply.
claude_voice.spoken_word_cap newer than v0.30.0-beta.2
Advanced config-only limit for each spoken Claude reply chunk.
click
click.button
Compatibility command metadata; actual mouse buttons belong to command records, not global preferences.
click.type
Compatibility command metadata; actual click actions belong to command records, not global preferences.
comfy
comfy.auto_open newer than v0.30.0-beta.2
Advanced image completion preference; open the image remains available by voice.
comfy.base_url newer than v0.30.0-beta.2
Local ComfyUI server address; loopback only.
Must be a loopback address. Default http://127.0.0.1:8001.
comfy.default_workflow newer than v0.30.0-beta.2
Name of an explicitly imported local image workflow.
comfy.prompt_prefix newer than v0.30.0-beta.2
Owner-editable text prepended verbatim to image prompts; empty by default.
comfy.prompt_suffix newer than v0.30.0-beta.2
Owner-editable text appended verbatim to image prompts; empty by default.
comfy.timeout_s newer than v0.30.0-beta.2
Local image rendering deadline in seconds.
How long to wait for an image to render. Default 600 seconds.
comfy.workflows newer than v0.30.0-beta.2
Imported API graphs and their exact input mappings; configured by the workflow import tool.
command
command.button
Phase 2A compatibility key; command_mode.button is the current Settings control.
command.trigger_mode
Phase 2A compatibility key; command_mode.mode is the current Settings control.
command_mode
command_mode.ava_tts_half_duplex
Developer override for echo suppression while Ava speaks; keep the default protection for ordinary use.
command_mode.dictate_commit_word
Config-only single-word session protocol override; changing the commit vocabulary needs the session token validation, not an unrestricted text field.
The one word that pastes a staged hands-free thought. Default finish. A single word only.
command_mode.personal_alias_offer_timeout_s
Config-only tuning for the specialist spoken alias-offer flow, separate from ordinary session inactivity.
command_mode.preview_position
Placement state owned by the draggable preview and its Settings reset action.
command_mode.session_streaming_preview
Config-only legacy session-preview switch; separate from the dedicated CapsLock streaming-capture preference.
correction_capture
correction_capture.max_edit_ratio
Config-only correction-training rejection threshold; not an ordinary text-editing preference.
debug
debug.dump_hotkey_buffers
Developer-only raw audio capture for buffer investigations; deliberately excluded from everyday Settings.
diagnostics
diagnostics.watch_code_changes newer than v0.30.0-beta.2
Source builds only; it has no control in Settings. Checks once a minute whether Samsara's code changed on disk and then offers Restart Samsara from the tray. Packaged installs never show it.
dictation
dictation.signoff_filter
Developer escape hatch for the default-on hallucinated-signoff filter during false-positive evaluation.
dispatch_voice
dispatch_voice.enabled newer than v0.30.0-beta.2
Agora voice dispatch and its event feed.
Voice dispatch to the Agora queue. A developer feature.
dispatch_voice.poll_s newer than v0.30.0-beta.2
Seconds between Agora dispatch feed checks.
ducking
ducking.enabled
Legacy fallback for configurations without ducking.recording_mode; the Audio recording-mode selector owns the current behavior.
formatting_tokens
formatting_tokens.enabled
Config-only escape hatch for spoken formatting-token interpretation; normal dictation keeps formatting tokens enabled.
intent
intent.shadow_enabled
Developer-only shadow execution experiment that records dictated text without changing actions; kept config-only.
intent.shadow_retention_days
Retention tuning for the config-only shadow execution experiment, not general log retention.
ollama
ollama.enabled
Legacy provider flag; the Ava master and consent-aware provider policy own current activation.
ollama.host
Config-only local-server endpoint override; normal setup uses the local default and provider consent checks.
ollama.max_response_length
Config-only conversational response-length cap; not the spoken acknowledgement character limit.
ollama.model
Local conversational model selection is owned by the Ava setup guide; distinct from the command-miss model in Settings.
ollama.safety_gate_enabled
Developer compatibility override for local-model safety checks; not an ordinary user preference.
ollama.timeout_seconds
Config-only local conversational request deadline; separate from cloud and editing timeouts.
onboarding
onboarding.lessons
Practice progress recorded by onboarding lessons; not a free-form Settings dictionary.
onboarding.next_step
Resume location owned by the first-run workflow.
onboarding.requested_ai
Pending provider choice owned by setup; the Ava page owns the active consent-aware provider policy.
onboarding.setup_status
Setup progress owned by the first-run workflow; not a manually editable readiness claim.
onboarding.version
Setup workflow version owned by onboarding migrations.
queue_voice
queue_voice.python newer than v0.30.0-beta.2
Interpreter for the existing queue note script; advanced config-only path.
queue_voice.queue_root newer than v0.30.0-beta.2
Owner-maintained prompt queue location; advanced config-only path.
queue_voice.reports_root newer than v0.30.0-beta.2
Owner-maintained prompt reports location; advanced config-only path.
smart_corrections
smart_corrections.keep_alive
Config-only Ollama duration override; it accepts backend duration syntax, not an ordinary Settings on/off value.
smart_corrections.min_words
Config-only correction eligibility threshold; the visible controls choose correction modes and repair behavior.
smart_corrections.timeout_s
Config-only correction-backend deadline tuning; retain the bounded runtime default for ordinary use.
transcription
transcription.mode
Compatibility schema value; cleanup_mode is the current dictation cleanup preference.
tts
tts.rate
Phase 2A compatibility name; tts.speed is the current speech-speed preference.
tts.self_hear_grace_ms newer than v0.30.0-beta.2
Capture interlock grace after Ava speech ends.
How long after Ava stops speaking captures that lie wholly inside her speech are still dropped. Default 400 ms.
ui
ui.idle_animation
Config-only animation override for the legacy listening indicator, separate from indicator visibility.
ui.live_surface.focus_hotkey
Experimental live-surface binding; config-only until it joins the validated Settings shortcut/collision editor.
ui.live_surface.idle_delay_s
Experimental live-surface tuning; its public Settings surface remains the opt-in switch until the replacement UI is released.
ui.live_surface.legacy
Internal migration backup of previous floating-window settings; never a preference editor.
ui.live_surface.partials
Experimental partial-transcription policy; not a second control for the existing streaming capture path.
ui.live_surface.pause_hotkey
Experimental live-surface binding; config-only until it joins the validated Settings shortcut/collision editor.
ui.live_surface.placement
Per-monitor geometry written by the live surface itself, not a text-editable preference.
ui.live_surface.reset_hotkey
Experimental live-surface binding; config-only until it joins the validated Settings shortcut/collision editor.
ui.live_surface.schema_version
Internal live-surface migration version, not a user preference.
ui.live_surface.show_idle_mark
Experimental live-surface tuning; its public Settings surface remains the opt-in switch until the replacement UI is released.
wake_word_config
wake_word_config.opens_session
Config-only opt-in to the alternative latched wake route, which also requires command_mode.mode=toggle.
Lets a wake phrase open a hands-free session. Needs the session button in toggle mode.