Configuration

Pia stores its configuration in %APPDATA%\Pia\settings.json. Property names are camelCase, and enum values are written as their names ("Dark", "Parakeet").

Caution

Pia owns this file and rewrites it whenever a setting changes. Edit it only while Pia is closed, or your changes will be overwritten. Everything below is reachable from the settings UI — hand-editing is for scripted rollouts and the handful of keys marked no UI.

To push settings to managed devices instead, use Enterprise Policy — it takes the same property names.

Appearance and language

Setting	Type	Default	Description
theme	"System" | "Dark" | "Light"	"System"	UI theme
uiLanguage	"EN" | "DE" | "FR"	"EN"	Interface language
targetLanguage	"EN" | "DE" | "FR"	unset	Default output language for Optimize
targetSpeechLanguage	"Auto" | "EN" | "DE" | "FR"	"Auto"	Speech-recognition language

Startup and windows

Setting	Type	Default	Description
launchAtStartup	bool	true	Start Pia when you sign in to Windows
startMinimized	bool	false	Launch hidden in the system tray
autoCaptureSelectedText	bool	true	Paste the selected text in when a hotkey triggers Pia
defaultWindowMode	"Optimize" | "Assistant"	"Optimize"	Mode used by the New Window button
autoUpdateEnabled	bool	true	Allow auto-update on launch

Optimize

Setting	Type	Default	Description
defaultOutputAction	"CopyToClipboard" | "AutoType" | "PasteToPreviousWindow"	"CopyToClipboard"	What happens when Optimize finishes
autoTypeDelayMs	integer	10	AutoType delay in milliseconds
defaultTemplateId	GUID	unset	Default optimization template

Speech

Setting	Type	Default	Description
sttBackend	"Parakeet" | "Whisper"	"Parakeet"	Speech-to-text engine
whisperModel	"Tiny" | "Base" | "Small" | "Medium" | "Large"	"Base"	Whisper model size (ignored by Parakeet)
ttsEnabled	bool	false	Enable text-to-speech
ttsVoiceModelKey	string	"en_US-lessac-medium"	TTS voice identifier

Providers and personas

Setting	Type	Default	Description
defaultProviderId	GUID	unset	Provider used when no per-mode default applies
useSameProviderForAllModes	bool	true	Use one provider across Optimize and Assistant
modeProviderDefaults	object	{}	Window mode → provider GUID
modePersonaDefaults	object	{}	Window mode → persona GUID

Assistant and files

Setting	Type	Default	Description
assistantFilesFolder	string	unset	The only folder the assistant may touch. Empty ⇒ no file access
assistantDefaultWorkingDirectory	string	"Playground"	Subfolder new chats start in. Empty ⇒ the folder root
assistantFileToolsEnabled	bool	true	Allow reading and writing files in that folder
assistantGitToolsEnabled	bool	true	Allow git commands in that folder
assistantSuggestionsEnabled	bool	false	Show suggested prompts on an empty chat
assistantAgentModeDefault	bool	false	Start new chats with the lever on Agent (no UI)
alwaysAllowedTools	array	[]	Standing tool grants. Managed from the settings page

Chat history

Setting	Type	Default	Description
chatHistoryRetentionDays	integer	180	Delete chats not opened for this many days (1–730). A favorite chat is exempt
chatAutoTitleEnabled	bool	false	Generate a short title per chat (costs an extra call)

Agent runs

See the Agent Runs guide for what each limit means.

Setting	Type	Default	Description
agentMaxSteps	integer	24	Maximum steps in an interactive run
maxToolRoundsPerStep	integer	24	Maximum tool rounds within one step
agentWallClockMinutes	integer	20	Time limit for an interactive run
agentMaxReplans	integer	2	Re-plans allowed after a step fails
agentPlanReasoningTurnEnabled	bool	false	Add a tool-free thinking turn before planning
agentRunAutoApproveBuiltInWrites	bool	false	Skip approval for Pia’s own write tools in agent runs and voice mode
agentPersonaRoster	object	{}	Operating mode → persona GUIDs the planner may assign to steps
scheduledMaxSteps	integer	24	Maximum steps in a background or scheduled run
scheduledWallClockMinutes	integer	45	Time limit for a background or scheduled run
scheduledMaxReplans	integer	2	Re-plans allowed in a background or scheduled run
maxParallelBackgroundRuns	integer	2	Background runs at the same time; extras queue
maxParallelRequestsPerProvider	integer	4	Concurrent requests Pia will make to one provider (no UI)

Meetings and transcription

Setting	Type	Default	Description
meetingAttendeeDisplayName	string	unset	Name shown to other participants. Empty ⇒ “‹your name›’s assistant”
meetingTranscriptFolder	string	unset	Folder the Save… dialog opens in
meetingBrowserSelection	"BundledChromium" | "SystemChrome" | "SystemEdge" | "SystemDefault"	"BundledChromium"	Which browser Pia drives
meetingAttendeeShowBrowserWindow	bool	false	Show the browser window instead of joining hidden
enableMeetingDiarization	bool	true	Split the transcript by speaker
meetingSmartSpeakerDetection	bool	true	Let Pia tune and correct speaker assignments itself
speakerEmbeddingThreshold	float	0.50	Voice-matching sensitivity (smart detection off)
meetingMaxSpeakers	integer	0	Cap on distinct speakers. 0 ⇒ no limit
meetingMinSpeechSeconds	float	1.5	Speech needed before a speaker is identified
meetingAttendeeRosterSnapshotMinutes	integer	2	How often the participant list is re-read (no UI)

Memory and privacy

Setting	Type	Default	Description
autoIngestSources	bool	true	Compile documents dropped into the vault’s sources folder
privacy.tokenizationEnabled	bool	true	Replace personal data with tokens before sending text to a provider
privacy.piiKeywords	array	[]	Your own PII keywords

Sync and server

Setting	Type	Default	Description
syncEnabled	bool	false	Enable Pia Cloud sync
serverUrl	string	unset	Server URL, for self-hosting
trustSelfSignedCertificates	bool	false	Accept self-signed TLS certificates

Keys Pia manages itself

settings.json also holds state that isn’t a preference: sign-in tokens, encryption key material and device IDs, sync timestamps and ETags, window position and size, unsent drafts, column widths, and schema-version markers. Leave these alone — editing them can sign you out or, in the case of the encryption keys, cut you off from your own data.