config

type AssistConfig struct {
	EnabledTools []string `toml:"enabled_tools"`
}

type AudioConfig struct {
	Backend        string `toml:"backend"`
	DeviceID       string `toml:"device_id"`
	OutputDeviceID string `toml:"output_device_id"`
	SampleRate     int    `toml:"sample_rate"`
	Channels       int    `toml:"channels"`
	FrameSizeMs    int    `toml:"frame_size_ms"`
	LatencyHint    string `toml:"latency_hint"`
}

type AuditConfig struct {
	Enabled         bool   `toml:"enabled"`
	RetentionDays   int    `toml:"retention_days"`
	EventLogEnabled bool   `toml:"event_log_enabled"` // wired in P2.1 (cpv.3.1) — Windows Event Log mirror
	OTLPEndpoint    string `toml:"otlp_endpoint"`     // wired in P2.2 (cpv.3.2) — OTLP exporter
	OTLPCertFile    string `toml:"otlp_cert_file"`
	OTLPKeyFile     string `toml:"otlp_key_file"`
	OTLPCAFile      string `toml:"otlp_ca_file"`
}

type Config struct {
	General        GeneralConfig        `toml:"general"`
	Audio          AudioConfig          `toml:"audio"`
	UI             UIConfig             `toml:"ui"`
	Vocabulary     VocabularyConfig     `toml:"vocabulary"`
	Assist         AssistConfig         `toml:"assist"`
	Shortcuts      ShortcutsConfig      `toml:"shortcuts"`
	ModelSelection ModelSelectionConfig `toml:"model_selection"`

	// ServerConnection points the device/local-target at a remote SpeechKit
	// Server-Target. Only consulted when at least one mode in ModelSelection
	// has mode_source = "server". Disabled by default; the desktop app runs
	// fully self-contained until a user opts a mode into server-side
	// execution (typically via onboarding or settings).
	ServerConnection ServerConnectionConfig `toml:"server_connection"`

	Local       LocalConfig       `toml:"local"`
	LocalLLM    LocalLLMConfig    `toml:"local_llm"`
	VPS         VPSConfig         `toml:"vps"`
	HuggingFace HuggingFaceConfig `toml:"huggingface"`
	Routing     RoutingConfig     `toml:"routing"`
	Update      UpdateConfig      `toml:"update"`
	Logging     LoggingConfig     `toml:"logging"`
	Audit       AuditConfig       `toml:"audit"`
	Telemetry   TelemetryConfig   `toml:"telemetry"`
	Feedback    FeedbackConfig    `toml:"feedback"` // legacy compat; prefer Store
	Store       StoreConfig       `toml:"store"`
	Providers   ProvidersConfig   `toml:"providers"`
	TTS         TTSConfig         `toml:"tts"`
	VoiceAgent  VoiceAgentConfig  `toml:"voice_agent"`

	// Server configures the standalone Linux server binary (cmd/speechkit-server).
	// All fields are optional; the desktop app (cmd/speechkit) ignores them entirely.
	Server    ServerConfig     `toml:"server"`
	Personas  []PersonaConfig  `toml:"personas"`
	Roles     []RoleConfig     `toml:"roles"`
	Sequences []SequenceConfig `toml:"sequences"`

	// Wakeword configures the always-on "Hey Quby" activation-word listener.
	// Read by cmd/speechkit (Device-Target) and any library embedder; the
	// Server-Target ignores this block in v1.
	Wakeword WakewordConfig `toml:"wakeword"`
}

type FeedbackConfig struct {
	SaveAudio          bool   `toml:"save_audio"`
	AudioRetentionDays int    `toml:"audio_retention_days"`
	DBPath             string `toml:"db_path"`
	MaxAudioStorageMB  int    `toml:"max_audio_storage_mb"`
}

type GeneralConfig struct {
	Language                 string `toml:"language"`
	Hotkey                   string `toml:"hotkey"` // Deprecated: legacy single-hotkey field kept for config file compat. Use DictateHotkey.
	DictateHotkey            string `toml:"dictate_hotkey"`
	AssistHotkey             string `toml:"assist_hotkey"`
	VoiceAgentHotkey         string `toml:"voice_agent_hotkey"`
	DictateHotkeyBehavior    string `toml:"dictate_hotkey_behavior"`
	AssistHotkeyBehavior     string `toml:"assist_hotkey_behavior"`
	VoiceAgentHotkeyBehavior string `toml:"voice_agent_hotkey_behavior"`
	DictateEnabled           bool   `toml:"dictate_enabled"`
	AssistEnabled            bool   `toml:"assist_enabled"`
	VoiceAgentEnabled        bool   `toml:"voice_agent_enabled"`
	AutoStartOnLaunch        bool   `toml:"auto_start_on_launch"`
	AgentHotkey              string `toml:"agent_hotkey"`
	AgentMode                string `toml:"agent_mode"`  // "assist" or "voice_agent" — determines what agent_hotkey triggers
	ActiveMode               string `toml:"active_mode"` // legacy compat
	HotkeyMode               string `toml:"hotkey_mode"` // legacy compat for single behavior setting
	AutoStopSilenceMs        int    `toml:"auto_stop_silence_ms"`
	FastModeSilenceMs        int    `toml:"fast_mode_silence_ms"` // silence threshold for Quick Capture auto-stop
	ModelDownloadDir         string `toml:"model_download_dir"`   // Default directory for downloaded local model files
}

type GoogleProviderConfig struct {
	Enabled      bool   `toml:"enabled"`
	APIKeyEnv    string `toml:"api_key_env"`
	STTAPIKeyEnv string `toml:"stt_api_key_env"`
	STTModel     string `toml:"stt_model"`
	UtilityModel string `toml:"utility_model"`
	AssistModel  string `toml:"assist_model"`
	AgentModel   string `toml:"agent_model"`
	// Region is the Google Cloud region the customer's API key / project is
	// pinned to. Default "europe-west3" (Frankfurt) reflects the EU-enterprise
	// compliance posture. US customers should explicitly set "us-central1".
	//
	// IMPORTANT: this field feeds the byok.key_updated audit event and the
	// settings UI. It does NOT redirect API traffic — the Gemini Live endpoint
	// is a single global WebSocket (generativelanguage.googleapis.com). Actual
	// data residency is controlled at the Google Cloud project level. Both the
	// project region AND this field must match for the audit event to be
	// accurate. See docs/compliance/byok-gemini-region-pinning.md.
	Region string `toml:"region"`
}

type GroqProviderConfig struct {
	Enabled      bool   `toml:"enabled"`
	APIKeyEnv    string `toml:"api_key_env"`
	STTModel     string `toml:"stt_model"`
	UtilityModel string `toml:"utility_model"`
	AssistModel  string `toml:"assist_model"`
	AgentModel   string `toml:"agent_model"`
}

type HuggingFaceConfig struct {
	Enabled      bool   `toml:"enabled"`
	Model        string `toml:"model"`
	UtilityModel string `toml:"utility_model"`
	AssistModel  string `toml:"assist_model"`
	AgentModel   string `toml:"agent_model"`
	TokenEnv     string `toml:"token_env"`
}

type InstallMode string

type InstallState struct {
	Mode      InstallMode `toml:"mode"`
	SetupDone bool        `toml:"setup_done"`
	DeviceID  string      `toml:"device_id"`
}

type LocalConfig struct {
	Enabled   bool   `toml:"enabled"`
	Model     string `toml:"model"`
	ModelPath string `toml:"model_path"`
	Port      int    `toml:"port"`
	GPU       string `toml:"gpu"`
}

type LocalLLMConfig struct {
	Enabled      bool   `toml:"enabled"`
	BaseURL      string `toml:"base_url"`
	Model        string `toml:"model"`
	ModelPath    string `toml:"model_path"`
	Port         int    `toml:"port"`
	GPU          string `toml:"gpu"`
	UtilityModel string `toml:"utility_model"`
	AssistModel  string `toml:"assist_model"`
	AgentModel   string `toml:"agent_model"`
}

type LoggingConfig struct {
	MaxFileSizeMB int    `toml:"max_file_size_mb"`
	MaxFiles      int    `toml:"max_files"`
	Level         string `toml:"level"` // "debug" | "info" | "warn" | "error" | "off"
}

type ModeModelSelection struct {
	PrimaryProfileID  string `toml:"primary_profile_id"`
	FallbackProfileID string `toml:"fallback_profile_id"`

	// ModeSource selects whether this mode runs locally (Framework kernel
	// in-process, default) or against a remote SpeechKit Server-Target
	// configured under [server_connection]. Empty string is treated as
	// ModeSourceLocal so existing configs keep behaving as before.
	ModeSource string `toml:"mode_source"`
}

type ModelSelectionConfig struct {
	Dictate    ModeModelSelection `toml:"dictate"`
	Assist     ModeModelSelection `toml:"assist"`
	VoiceAgent ModeModelSelection `toml:"voice_agent"`
}

type OllamaProviderConfig struct {
	Enabled      bool   `toml:"enabled"`
	BaseURL      string `toml:"base_url"`
	STTModel     string `toml:"stt_model"`
	UtilityModel string `toml:"utility_model"`
	AssistModel  string `toml:"assist_model"`
	AgentModel   string `toml:"agent_model"`
}

type OpenAIProviderConfig struct {
	Enabled       bool   `toml:"enabled"`
	APIKeyEnv     string `toml:"api_key_env"`
	STTModel      string `toml:"stt_model"`
	UtilityModel  string `toml:"utility_model"`
	AssistModel   string `toml:"assist_model"`
	AgentModel    string `toml:"agent_model"`
	TTSModel      string `toml:"tts_model"`
	TTSVoice      string `toml:"tts_voice"`
	RealtimeModel string `toml:"realtime_model"`
}

type OpenRouterProviderConfig struct {
	Enabled      bool   `toml:"enabled"`
	APIKeyEnv    string `toml:"api_key_env"`
	STTModel     string `toml:"stt_model"`
	UtilityModel string `toml:"utility_model"`
	AssistModel  string `toml:"assist_model"`
	AgentModel   string `toml:"agent_model"`
}

type OverlayFreePosition struct {
	X int `toml:"x"`
	Y int `toml:"y"`
}

type PersonaConfig struct {
	ID              string            `toml:"id"`
	DisplayName     string            `toml:"display_name"`
	Description     string            `toml:"description"`
	Voice           string            `toml:"voice"`
	Locale          string            `toml:"locale"`
	DefaultRole     string            `toml:"default_role"`
	DefaultSequence string            `toml:"default_sequence"`
	Tags            []string          `toml:"tags"`
	Metadata        map[string]string `toml:"metadata"`
}

type PolicyValues struct {
	UpdateEnabled             *bool
	UpdateManifestURL         string
	TelemetryUpdateCheck      *bool
	ProvidersEnforceLocalOnly *bool
	VoiceAgentAllowCloud      *bool
	AuditRetentionDays        *int
	AuditEventLogEnabled      *bool
	AuditOTLPEndpoint         string

	// Origin describes which registry hive first contributed a value.
	// One of "hklm-policies" | "hklm-defaults" | "hkcu" | "none" | "non-windows".
	Origin string
	// KeysFound counts all registry values that contributed to this overlay
	// (across all hives). Used in the policy.applied audit event.
	KeysFound int
}

type ProvidersConfig struct {
	OpenAI     OpenAIProviderConfig     `toml:"openai"`
	Groq       GroqProviderConfig       `toml:"groq"`
	Google     GoogleProviderConfig     `toml:"google"`
	Ollama     OllamaProviderConfig     `toml:"ollama"`
	OpenRouter OpenRouterProviderConfig `toml:"openrouter"`
}

type RoleConfig struct {
	ID                          string   `toml:"id"`
	DisplayName                 string   `toml:"display_name"`
	SystemPrompt                string   `toml:"system_prompt"`
	RefinementPrompt            string   `toml:"refinement_prompt"`
	Locale                      string   `toml:"locale"`
	VocabularyHint              string   `toml:"vocabulary_hint"`
	ToolAllowlist               []string `toml:"tool_allowlist"`
	Temperature                 float64  `toml:"temperature"`
	ThinkingEnabled             bool     `toml:"thinking_enabled"`
	ThinkingLevel               string   `toml:"thinking_level"`
	IncludeThoughts             bool     `toml:"include_thoughts"`
	ThinkingBudget              int      `toml:"thinking_budget"`
	AutomaticActivityDetection  bool     `toml:"automatic_activity_detection"`
	VADStartSensitivity         string   `toml:"vad_start_sensitivity"`
	VADEndSensitivity           string   `toml:"vad_end_sensitivity"`
	VADPrefixPaddingMs          int      `toml:"vad_prefix_padding_ms"`
	VADSilenceDurationMs        int      `toml:"vad_silence_duration_ms"`
	ActivityHandling            string   `toml:"activity_handling"`
	TurnCoverage                string   `toml:"turn_coverage"`
	ContextCompressionEnabled   bool     `toml:"context_compression_enabled"`
	ContextCompressionTriggerTk int64    `toml:"context_compression_trigger_tokens"`
	ContextCompressionTargetTk  int64    `toml:"context_compression_target_tokens"`
	EnableAffectiveDialog       bool     `toml:"enable_affective_dialog"`
}

type RoutingConfig struct {
	Strategy                string  `toml:"strategy"`
	PreferLocalUnderSeconds float64 `toml:"prefer_local_under_seconds"`
	ParallelCloud           bool    `toml:"parallel_cloud"`
	ReplaceOnBetter         bool    `toml:"replace_on_better"`
}

type SequenceConfig struct {
	ID          string               `toml:"id"`
	DisplayName string               `toml:"display_name"`
	Description string               `toml:"description"`
	Completion  string               `toml:"completion"` // "all_steps" | "explicit_close" | "max_turns"
	MaxTurns    int                  `toml:"max_turns"`
	Steps       []SequenceStepConfig `toml:"steps"`
}

type SequenceStepConfig struct {
	ID           string   `toml:"id"`
	Instruction  string   `toml:"instruction"`
	ExitCriteria string   `toml:"exit_criteria"`
	RequireTools []string `toml:"require_tools"`
	MaxTurns     int      `toml:"max_turns"`
}

type ServerAdminAuthSettings struct {
	Enabled       *bool  `json:"enabled,omitempty"`
	Username      string `json:"username,omitempty"`
	PasswordHash  string `json:"password_hash,omitempty"`
	PasswordValue string `json:"password,omitempty"`
}

type ServerAssistSettings struct {
	EnabledTools []string `json:"enabled_tools,omitempty"`
}

type ServerAuthSettings struct {
	Mode           string `json:"mode,omitempty"`
	BearerTokenEnv string `json:"bearer_token_env,omitempty"`
	GenerateToken  *bool  `json:"generate_token,omitempty"`
	TokenValue     string `json:"token_value,omitempty"`
}

type ServerConfig struct {
	ListenAddr        string   `toml:"listen_addr"`          // e.g. ":8080"
	PublicURL         string   `toml:"public_url"`           // external API base URL, e.g. https://speechkit.example.com/api
	Modes             []string `toml:"modes"`                // subset of ["dictation","assist","voiceagent"]; empty = all
	AuthMode          string   `toml:"auth_mode"`            // "none" | "bearer" | "edge_hmac" | "bearer_or_edge"
	BearerTokenEnv    string   `toml:"bearer_token_env"`     // env var name holding the bearer token
	BearerRole        string   `toml:"bearer_role"`          // optional role for static bearer callers, e.g. "admin"
	AdminAuthEnabled  bool     `toml:"admin_auth_enabled"`   // enables setup/admin UI username/password login
	AdminUsername     string   `toml:"admin_username"`       // setup/admin UI username; not used by API clients
	AdminPasswordHash string   `toml:"admin_password_hash"`  // bcrypt hash for setup/admin UI login
	EdgeAuthSecretEnv string   `toml:"edge_auth_secret_env"` // env var name holding the HMAC secret
	// SmokeTokenEnv names an optional env var that holds a public-friendly
	// demo bearer token. When set, the smoke UI on `/` embeds the token in
	// the rendered HTML so visitors can run all three modes without
	// configuring credentials. The smoke identity is tagged Source="smoke"
	// (Plan="demo") so handlers and rate-limiters can distinguish demo
	// traffic. Leave empty to disable smoke-from-page entirely; operators
	// must then paste their bearer token in the UI manually.
	SmokeTokenEnv         string   `toml:"smoke_token_env"`
	PublicBaseURL         string   `toml:"public_base_url"`     // public server URL used for returned client URLs
	TrustedProxyCIDRs     []string `toml:"trusted_proxy_cidrs"` // proxies allowed to supply X-Forwarded-* headers
	CORSAllowedOrigins    []string `toml:"cors_allowed_origins"`
	RateLimitRPS          float64  `toml:"rate_limit_rps"`
	RateLimitBurst        int      `toml:"rate_limit_burst"`
	MaxUploadMB           int      `toml:"max_upload_mb"`
	MaxVoiceAgentSessions int      `toml:"max_voiceagent_sessions"` // global cap
	MaxSessionsPerUser    int      `toml:"max_sessions_per_user"`
	TicketTTLSec          int      `toml:"ticket_ttl_sec"` // Voice Agent WS ticket TTL
	// VoiceAgentIdleTimeoutSec terminates a Voice Agent WebSocket session
	// after N seconds without any client- or provider-side activity.
	// Defaults to 900 (15 min). Set to 0 to disable the server-side idle
	// timeout (kernel-level idle handling stays in effect either way).
	VoiceAgentIdleTimeoutSec int                  `toml:"voiceagent_idle_timeout_sec"`
	LiveKit                  ServerLiveKitConfig  `toml:"livekit"`
	WhisperBinary            string               `toml:"whisper_binary"` // absolute path inside container
	WhisperPort              int                  `toml:"whisper_port"`   // loopback port for whisper.cpp server
	ModelDir                 string               `toml:"model_dir"`      // persistent volume, e.g. /var/lib/speechkit/models
	LogFormat                string               `toml:"log_format"`     // "json" | "text"
	LogLevel                 string               `toml:"log_level"`      // "debug" | "info" | "warn" | "error"
	Features                 ServerFeaturesConfig `toml:"features"`
}

type ServerConnectionConfig struct {
	// Enabled gates the entire server connection. When false, every mode is
	// forced to run locally regardless of its mode_source. Lets users keep
	// their server URL in config but temporarily flip back to fully local.
	Enabled bool `toml:"enabled"`

	// URL is the base URL of the speechkit-server, e.g.
	// "https://speechkit.example.com" or "http://localhost:8080".
	URL string `toml:"url"`

	// BearerTokenEnv names the env var that holds the bearer token sent in
	// the Authorization header. Defaults to SPEECHKIT_SERVER_TOKEN. The
	// value is never read from the TOML file itself — only the env var name
	// is configured here.
	BearerTokenEnv string `toml:"bearer_token_env"`

	// AuthMode selects how the resolved token is attached to outbound
	// requests. "bearer" sends Authorization: Bearer <token>. "api_key"
	// sends X-Api-Key: <token> for servers that use header-based API keys.
	// Empty/missing defaults to "bearer".
	AuthMode string `toml:"auth_mode"`

	// FallbackToLocal makes the device app fall back to the in-process
	// Framework kernel if a server call fails or the server is unreachable.
	// Useful for laptop deployments that may be offline; should be false
	// for kiosks that must never silently downgrade to local processing.
	FallbackToLocal bool `toml:"fallback_to_local"`

	// RequestTimeoutSec caps non-streaming HTTP calls (Dictation, Assist).
	// 0 means no explicit timeout (the underlying http.Client default
	// applies). Voice Agent WebSocket sessions are not affected.
	RequestTimeoutSec int `toml:"request_timeout_sec"`

	// ActiveTargetID selects the registered server target copied into the
	// compatibility fields above. Empty means the top-level URL/env/auth fields
	// are an ad-hoc single target.
	ActiveTargetID string `toml:"active_target_id"`

	// Targets is the optional local registry of SpeechKit server endpoints the
	// device can switch between. These are user/operator configured; product
	// builds must not inject private gateway/origin entries here.
	Targets []ServerConnectionTargetConfig `toml:"targets"`
}

type ServerConnectionTargetConfig struct {
	ID                string `toml:"id"`
	Label             string `toml:"label"`
	URL               string `toml:"url"`
	BearerTokenEnv    string `toml:"bearer_token_env"`
	AuthMode          string `toml:"auth_mode"`
	FallbackToLocal   bool   `toml:"fallback_to_local"`
	RequestTimeoutSec int    `toml:"request_timeout_sec"`
}

type ServerCredentialSettings struct {
	OpenAI      ServerProviderCredentialSettings `json:"openai,omitempty"`
	Groq        ServerProviderCredentialSettings `json:"groq,omitempty"`
	Google      ServerProviderCredentialSettings `json:"google,omitempty"`
	HuggingFace ServerProviderCredentialSettings `json:"huggingface,omitempty"`
	OpenRouter  ServerProviderCredentialSettings `json:"openrouter,omitempty"`
}

type ServerDictationSettings struct {
	Dictionary *string `json:"dictionary,omitempty"`
}

type ServerFeaturesConfig struct {
	Catalog      bool `toml:"catalog"`
	StorageReads bool `toml:"storage_reads"`
	Vocabulary   bool `toml:"vocabulary"`
	TTSDirect    bool `toml:"tts_direct"`
}

type ServerLLMSettings struct {
	Enabled      *bool  `json:"enabled,omitempty"`
	BaseURL      string `json:"base_url,omitempty"`
	UtilityModel string `json:"utility_model,omitempty"`
	AssistModel  string `json:"assist_model,omitempty"`
	AgentModel   string `json:"agent_model,omitempty"`
	HFRepo       string `json:"hf_repo,omitempty"`
}

type ServerLiveKitConfig struct {
	Enabled      bool   `toml:"enabled"`
	URL          string `toml:"url"`            // e.g. wss://livekit.example.com
	APIKeyEnv    string `toml:"api_key_env"`    // env var name holding the LiveKit API key
	APISecretEnv string `toml:"api_secret_env"` // env var name holding the LiveKit API secret
	TokenTTLSec  int    `toml:"token_ttl_sec"`  // join-token TTL
	RoomPrefix   string `toml:"room_prefix"`    // room name prefix for SpeechKit-managed rooms
}

type ServerModeProviderSettings struct {
	Dictation  ServerModeSetting `json:"dictation,omitempty"`
	Assist     ServerModeSetting `json:"assist,omitempty"`
	VoiceAgent ServerModeSetting `json:"voice_agent,omitempty"`
}

type ServerModeSetting struct {
	Enabled      *bool  `json:"enabled,omitempty"`
	ProviderKind string `json:"provider_kind,omitempty"`
	ProfileID    string `json:"profile_id,omitempty"`
	Model        string `json:"model,omitempty"`
}

type ServerModelSettings struct {
	Version            int                        `json:"version,omitempty"`
	OnboardingComplete bool                       `json:"onboarding_complete,omitempty"`
	OnboardingVersion  string                     `json:"onboarding_version,omitempty"`
	ServerAuth         ServerAuthSettings         `json:"server_auth,omitempty"`
	AdminAuth          ServerAdminAuthSettings    `json:"admin_auth,omitempty"`
	Modes              ServerModeProviderSettings `json:"modes,omitempty"`
	Credentials        ServerCredentialSettings   `json:"credentials,omitempty"`
	Dictation          ServerDictationSettings    `json:"dictation,omitempty"`
	Assist             ServerAssistSettings       `json:"assist,omitempty"`
	STT                ServerSTTSettings          `json:"stt,omitempty"`
	LLM                ServerLLMSettings          `json:"llm,omitempty"`
	VoiceAgent         ServerVoiceAgentSettings   `json:"voice_agent,omitempty"`
	TTS                ServerOptionalTTSSettings  `json:"tts,omitempty"`
}

type ServerOptionalTTSSettings struct {
	Enabled *bool `json:"enabled,omitempty"`
}

type ServerProviderCredentialSettings struct {
	Enabled *bool  `json:"enabled,omitempty"`
	Env     string `json:"env,omitempty"`
	Value   string `json:"value,omitempty"`
}

type ServerSTTSettings struct {
	Enabled *bool  `json:"enabled,omitempty"`
	URL     string `json:"url,omitempty"`
	Model   string `json:"model,omitempty"`
}

type ServerVoiceAgentSettings struct {
	Provider        string  `json:"provider,omitempty"`
	AgentProfileID  string  `json:"agent_profile_id,omitempty"`
	AgentSequenceID string  `json:"agent_sequence_id,omitempty"`
	PromptTemplate  *string `json:"prompt_template,omitempty"`
}

type ShortcutLocaleConfig struct {
	LeadingFillers []string `toml:"leading_fillers"`
	CopyLast       []string `toml:"copy_last"`
	InsertLast     []string `toml:"insert_last"`
	Summarize      []string `toml:"summarize"`
	QuickNote      []string `toml:"quick_note"`
}

type ShortcutsConfig struct {
	Locale map[string]ShortcutLocaleConfig `toml:"locale"`
}

type StoreConfig struct {
	Backend            string `toml:"backend"` // "sqlite" | "postgres" | registered name
	SQLitePath         string `toml:"sqlite_path"`
	PostgresDSN        string `toml:"postgres_dsn"`
	SaveAudio          bool   `toml:"save_audio"`
	AudioRetentionDays int    `toml:"audio_retention_days"`
	MaxAudioStorageMB  int    `toml:"max_audio_storage_mb"`
}

type TTSConfig struct {
	Enabled     bool           `toml:"enabled"`
	Strategy    string         `toml:"strategy"` // "cloud-first", "local-first", "cloud-only", "local-only"
	Voice       string         `toml:"voice"`    // Global default voice override
	Speed       float64        `toml:"speed"`    // Global speed 0.25-4.0, default 1.0
	Format      string         `toml:"format"`   // "mp3", "wav", "opus", "pcm"
	OpenAI      TTSOpenAI      `toml:"openai"`
	Google      TTSGoogle      `toml:"google"`
	HuggingFace TTSHuggingFace `toml:"huggingface"`
	Local       TTSLocal       `toml:"local"`
}

type TTSGoogle struct {
	Enabled bool   `toml:"enabled"`
	Voice   string `toml:"voice"` // e.g. "de-DE-Neural2-B"
}

type TTSHuggingFace struct {
	Enabled bool   `toml:"enabled"`
	Model   string `toml:"model"` // e.g. "parler-tts/parler-tts-mini-multilingual-v1.1"
}

type TTSLocal struct {
	Enabled   bool   `toml:"enabled"`
	Model     string `toml:"model"`
	ModelPath string `toml:"model_path"`
	Port      int    `toml:"port"`
}

type TTSOpenAI struct {
	Enabled bool   `toml:"enabled"`
	Model   string `toml:"model"` // "tts-1" or "tts-1-hd"
	Voice   string `toml:"voice"` // alloy, echo, fable, onyx, nova, shimmer
}

type TelemetryConfig struct {
	UpdateCheck bool `toml:"update_check"`
}

type UIConfig struct {
	OverlayEnabled          bool                           `toml:"overlay_enabled"`
	OverlayPosition         string                         `toml:"overlay_position"` // "top", "bottom", "left", "right"
	OverlayMovable          bool                           `toml:"overlay_movable"`
	OverlayFreeX            int                            `toml:"overlay_free_x"`
	OverlayFreeY            int                            `toml:"overlay_free_y"`
	OverlayMonitorPositions map[string]OverlayFreePosition `toml:"overlay_monitor_positions"`
	Visualizer              string                         `toml:"visualizer"`
	Design                  string                         `toml:"design"`
	AssistOverlayMode       string                         `toml:"assist_overlay_mode"`
	VoiceAgentOverlayMode   string                         `toml:"voice_agent_overlay_mode"`
}

type UpdateConfig struct {
	Enabled                bool   `toml:"enabled"`
	ManifestURL            string `toml:"manifest_url"`
	CheckIntervalHours     int    `toml:"check_interval_hours"`
	SignaturePinThumbprint string `toml:"signature_pin_thumbprint"` // optional Authenticode SHA-1 thumbprint; if set, installer signature verification additionally checks cert thumbprint matches (defense against compromised signing cert)
}

type VPSConfig struct {
	Enabled   bool   `toml:"enabled"`
	URL       string `toml:"url"`
	Model     string `toml:"model"`
	APIKeyEnv string `toml:"api_key_env"`
}

type VocabularyConfig struct {
	Dictionary string `toml:"dictionary"`
}

type VoiceAgentConfig struct {
	Enabled bool `toml:"enabled"`
	// Provider selects the backend that drives a Voice Agent session.
	// Supported values:
	//   ""          (default) — same as "gemini"
	//   "gemini"    — Google Gemini Live (cloud, GOOGLE_AI_API_KEY required)
	//   "cascaded"  — self-hosted whisper.cpp → Genkit agent LLM → TTS pipeline
	//                 (CPU-capable; no external realtime dependency)
	//   "moshi"     — self-hosted Kyutai Moshi Rust server (GPU required, M9b)
	//
	// The Server-Target reads this field via cmd/speechkit-server; the Device-
	// Target currently always uses "gemini" and ignores it.
	Provider               string `toml:"provider"`
	Model                  string `toml:"model"`             // Real-time model ID (e.g. "gemini-3.1-flash-live-preview")
	FallbackModel          string `toml:"fallback_model"`    // Fallback real-time model
	Voice                  string `toml:"voice"`             // Voice name for real-time model
	AgentProfileID         string `toml:"agent_profile_id"`  // Built-in Voice Agent profile ID; "default" preserves current behavior.
	AgentSequenceID        string `toml:"agent_sequence_id"` // Optional workflow sequence ID; empty uses the selected persona default.
	FrameworkPrompt        string `toml:"framework_prompt"`  // Durable host/framework instruction that defines the Voice Agent behavior
	RefinementPrompt       string `toml:"refinement_prompt"` // User-specific refinement appended to the framework prompt
	Instruction            string `toml:"instruction"`       // Legacy alias for FrameworkPrompt
	AutoStartOnLaunch      bool   `toml:"auto_start_on_launch"`
	CloseBehavior          string `toml:"close_behavior"` // "continue" keeps the conversation window in the taskbar; "new_chat" ends the current chat on close
	ReminderAfterIdleSec   int    `toml:"reminder_after_idle_sec"`
	DeactivateAfterIdleSec int    `toml:"deactivate_after_idle_sec"`
	// HoldReleaseGraceSec controls how long the Voice Agent stays open after
	// the user releases a hold-to-talk shortcut so the model has time to
	// deliver its reply. 0 (or unset) falls back to the kernel default
	// (10 seconds). The Device-Target hard-caps this at 30 seconds; values
	// above that are silently clamped at runtime so a misconfigured profile
	// cannot strand the user in a "still active" session.
	HoldReleaseGraceSec             int    `toml:"hold_release_grace_sec"`
	PipelineFallback                bool   `toml:"pipeline_fallback"` // Use STT -> Agent LLM -> optional TTS when the selected Voice Agent profile is not native realtime.
	ShowPrompter                    bool   `toml:"show_prompter"`     // Show live transcript prompter window
	EnableSessionSummary            bool   `toml:"enable_session_summary"`
	EnableInputTranscript           bool   `toml:"enable_input_transcript"`
	EnableOutputTranscript          bool   `toml:"enable_output_transcript"`
	EnableAffectiveDialog           bool   `toml:"enable_affective_dialog"`
	ThinkingEnabled                 bool   `toml:"thinking_enabled"`
	IncludeThoughts                 bool   `toml:"include_thoughts"`
	ThinkingBudget                  int    `toml:"thinking_budget"`
	ThinkingLevel                   string `toml:"thinking_level"`
	ContextCompressionEnabled       bool   `toml:"context_compression_enabled"`
	ContextCompressionTriggerTokens int64  `toml:"context_compression_trigger_tokens"`
	ContextCompressionTargetTokens  int64  `toml:"context_compression_target_tokens"`
	AutomaticActivityDetection      bool   `toml:"automatic_activity_detection"`
	ActivityHandling                string `toml:"activity_handling"`
	TurnCoverage                    string `toml:"turn_coverage"`
	VADStartSensitivity             string `toml:"vad_start_sensitivity"`
	VADEndSensitivity               string `toml:"vad_end_sensitivity"`
	VADPrefixPaddingMs              int    `toml:"vad_prefix_padding_ms"`
	VADSilenceDurationMs            int    `toml:"vad_silence_duration_ms"`
}

type WakewordAutoEndConfig struct {
	// SilenceCutoffSec is the duration without user audio activity (in
	// whole seconds) after which a wake-word-triggered session ends.
	// Zero falls back to the framework default (10s).
	SilenceCutoffSec int `toml:"silence_cutoff_sec"`

	// ExitPhrases is the case-insensitive substring list checked against
	// each user-transcript snippet. Empty falls back to the framework
	// default (DE+EN common closers: "danke", "tschuess", "ende", "stop",
	// "thanks", "bye", "goodbye", ...).
	ExitPhrases []string `toml:"exit_phrases"`
}

type WakewordConfig struct {
	// Enabled gates the entire feature. Default false (opt-in).
	Enabled bool `toml:"enabled"`

	// Backend selects the local detector implementation. The Windows app
	// ships Sherpa-ONNX KWS, LiveKit/openWakeWord ONNX, and STT phrase-match
	// as explicit selectable paths so test builds can compare detector
	// behaviour without silently falling back to a different implementation.
	Backend string `toml:"backend"`

	// PhraseID picks one of SpeechKit's curated wake phrases from
	// wakeword.DefaultCatalog (e.g. "hey_quby", "hey_computer",
	// "hey_jarvis", "hey_mira"). When set, the corresponding ONNX file is
	// resolved automatically and Phrase/ModelPath below are ignored. When
	// empty, the explicit Phrase + ModelPath fields are used instead
	// (custom phrase mode). Switching via this field is how users pick a
	// different wake phrase in settings without editing paths by hand.
	PhraseID string `toml:"phrase_id"`

	// Phrase is the display label of the trained wake phrase, surfaced in
	// the tray and status feed. It has NO effect on detection — the ONNX
	// model encodes the actual phrase(s). One model can be trained to
	// fire on multiple pronunciation variants (e.g. "Hey Cubi" and
	// "Hey Kubi" for the same brand "Quby") via target_phrases in the
	// training yaml; the display label here remains a single brand string.
	//
	// Ignored when PhraseID matches a catalog entry (the catalog's
	// DisplayName is used instead).
	Phrase string `toml:"phrase"`

	// ModelPath is the path to the trained phrase prediction model (.onnx).
	// Empty resolves to <data_dir>/models/wakeword/hey_quby.onnx at runtime.
	//
	// Ignored when PhraseID matches a catalog entry (the catalog's
	// FileName is resolved inside the wake-word models directory).
	ModelPath string `toml:"model_path"`

	// MelspecModelPath and EmbeddingModelPath point at the shared
	// openWakeWord upstream models. Empty values resolve to the same
	// directory as ModelPath with canonical filenames.
	MelspecModelPath   string `toml:"melspec_model_path"`
	EmbeddingModelPath string `toml:"embedding_model_path"`

	// DefaultMode is the runtime mode triggered when the wake phrase fires.
	// One of "dictate" | "assist" | "voice_agent". Defaults to voice_agent.
	DefaultMode string `toml:"default_mode"`

	// Threshold is the minimum probability to count a frame as a hit.
	// Range (0.0, 1.0]. Backend-specific defaults when this is 0:
	//   - LiveKit/openWakeWord: 0.5 (Wyoming/openWakeWord canonical)
	//   - Sherpa-onnx KWS: 0.25 (sherpa-onnx upstream default)
	//   - STT phrase match: 0 (substring match, no acoustic probability)
	// Use the in-app "Test wake word" self-test in Settings to calibrate
	// for your specific microphone + environment instead of guessing.
	Threshold float64 `toml:"threshold"`

	// MinConsecutiveFrames is the number of consecutive above-threshold
	// frames required before a trigger fires. Higher = fewer false-accepts,
	// more false-rejects. Defaults to 2.
	MinConsecutiveFrames int `toml:"min_consecutive_frames"`

	// CooldownMs is the minimum gap between two triggers, in milliseconds.
	// Defaults to 1500ms.
	CooldownMs int `toml:"cooldown_ms"`

	// DebugMode enables verbose detector diagnostics. When true the sidecars
	// emit per-decode score events (openWakeWord) or set the sherpa-onnx
	// ModelConfig.Debug flag (Sherpa KWS), and the host adapter forwards
	// those signals into the user-visible log feed. Default false — only flip
	// on while tuning a wake phrase, the score event stream is high-volume.
	DebugMode bool `toml:"debug_mode"`

	// AutoEnd controls the framework-level auto-end policy applied to any
	// session that the wake-word triggered. Wake-word-origin Voice-Agent
	// activations terminate automatically on silence after this many
	// seconds, or when the user utters one of the configured exit
	// phrases. Empty values fall back to wakeword.DefaultAutoEndConfig
	// (10s silence + DE/EN exit phrases) so a TOML without an [auto_end]
	// block still gets the framework baseline. There is intentionally no
	// hard-cap on session duration — Voice-Agent is designed for
	// multi-hour dialogs and a forced cap would break regular use.
	AutoEnd WakewordAutoEndConfig `toml:"auto_end"`
}