@speechos/core 0.2.0 → 0.2.2

package/dist/livekit.d.cts

@@ -3,7 +3,7 @@
  * Handles room connections, audio streaming, and transcription requests
  */
 import { Room } from "livekit-client";
-import type { LiveKitTokenResponse, ErrorSource } from "./types.js";
+import type { LiveKitTokenResponse, ErrorSource, CommandDefinition, CommandResult, VoiceSessionOptions } from "./types.js";
 /**
  * A deferred promise with timeout support.
  * Encapsulates resolve/reject/timeout in a single object for cleaner async handling.
@@ -31,21 +31,38 @@ declare class LiveKitManager {
     private room;
     private tokenData;
     private micTrack;
+    private cachedTokenData;
+    private tokenCacheTimestamp;
+    private tokenPrefetchPromise;
+    private tokenRefreshTimer;
+    private autoRefreshEnabled;
     private pendingTranscript;
     private pendingEditText;
+    private pendingCommand;
     private pendingTrackSubscribed;
-    private preWarmPromise;
     private editOriginalText;
+    private sessionSettings;
     /**
-     * Pre-warm resources for faster connection
-     * Call this when user shows intent (e.g., expands widget)
-     * Only fetches token - mic permission is requested when user clicks Dictate
+     * Check if the cached token is still valid (within TTL)
      */
-    preWarm(): Promise<void>;
+    private isCachedTokenValid;
+    /**
+     * Pre-fetch a LiveKit token for later use
+     * Call this early (e.g., when widget expands) to reduce latency when starting a voice session.
+     * If a prefetch is already in progress, returns the existing promise.
+     * If a valid cached token exists, returns it immediately.
+     */
+    prefetchToken(): Promise<LiveKitTokenResponse>;
     /**
      * Fetch a LiveKit token from the backend
+     * Uses cached token if valid, otherwise fetches a fresh one.
+     * Includes language settings and user vocabulary which are stored in the VoiceSession.
      */
     fetchToken(): Promise<LiveKitTokenResponse>;
+    /**
+     * Internal method to fetch a fresh token from the server
+     */
+    private fetchTokenFromServer;
     /**
      * Connect to a LiveKit room (fresh connection each time)
      */
@@ -65,8 +82,13 @@ declare class LiveKitManager {
     private handleDataMessage;
     /**
      * Publish microphone audio track
+     * Uses the device ID from session settings if set
      */
     enableMicrophone(): Promise<void>;
+    /**
+     * Log information about the current microphone track
+     */
+    private logMicrophoneInfo;
     /**
      * Disable microphone audio track
      */
@@ -76,10 +98,24 @@ declare class LiveKitManager {
      */
     sendDataMessage(message: object): Promise<void>;
     /**
-     * Start a voice session
-     * Connects to room, enables microphone, and waits for agent to subscribe to our track
+     * Start a voice session with pre-connect audio buffering
+     * Fetches a fresh token, then enables mic with preConnectBuffer to capture audio while connecting.
+     * Agent subscription happens in the background - we don't block on it.
+     *
+     * @param options - Session options including action type and parameters
      */
-    startVoiceSession(): Promise<void>;
+    startVoiceSession(options?: VoiceSessionOptions): Promise<void>;
+    /**
+     * Wait for the agent to subscribe to our audio track in the background
+     * Handles timeout errors without blocking the main flow
+     */
+    private waitForAgentSubscription;
+    /**
+     * Enable microphone with pre-connect buffering
+     * This starts capturing audio locally before the room is connected,
+     * buffering it until the connection is established.
+     */
+    private enableMicrophoneWithPreConnectBuffer;
     /**
      * Stop the voice session and request the transcript
      * Returns a promise that resolves with the transcript text
@@ -101,11 +137,47 @@ declare class LiveKitManager {
      * Alias for requestEditText - granular API naming
      */
     stopAndEdit(originalText: string): Promise<string>;
+    /**
+     * Request command matching using the transcript as input
+     * Sends command definitions to the backend, which matches the user's speech against them
+     * Returns a promise that resolves with the matched command or null if no match
+     * @throws Error if timeout occurs waiting for command result
+     */
+    requestCommand(commands: CommandDefinition[]): Promise<CommandResult | null>;
+    /**
+     * Alias for requestCommand - granular API naming
+     */
+    stopAndCommand(commands: CommandDefinition[]): Promise<CommandResult | null>;
     /**
      * Disconnect from the current room
      * Clears the token so a fresh one is fetched for the next session
      */
     disconnect(): Promise<void>;
+    /**
+     * Invalidate the cached token
+     * Call this when settings change that would affect the token (language, vocabulary)
+     */
+    invalidateTokenCache(): void;
+    /**
+     * Start auto-refreshing the token while the widget is expanded.
+     * Call this after a voice session completes to immediately fetch a fresh token
+     * (since each command requires its own token) and keep it fresh for subsequent commands.
+     */
+    startAutoRefresh(): void;
+    /**
+     * Stop auto-refreshing the token.
+     * Call this when the widget collapses or user navigates away.
+     */
+    stopAutoRefresh(): void;
+    /**
+     * Schedule a token refresh before the current cache expires.
+     * Handles computer sleep by checking elapsed time on each refresh attempt.
+     */
+    private scheduleTokenRefresh;
+    /**
+     * Perform the auto-refresh, handling computer sleep scenarios.
+     */
+    private performAutoRefresh;
     /**
      * Get the current room instance
      */
@@ -122,11 +194,6 @@ declare class LiveKitManager {
      * Check if microphone is enabled
      */
     isMicrophoneEnabled(): boolean;
-    /**
-     * Clear the cached token
-     * Used when user identity changes to ensure next session gets a fresh token
-     */
-    clearToken(): void;
 }
 export declare const livekit: LiveKitManager;
 export {};

package/dist/livekit.d.ts

@@ -3,7 +3,7 @@
  * Handles room connections, audio streaming, and transcription requests
  */
 import { Room } from "livekit-client";
-import type { LiveKitTokenResponse, ErrorSource } from "./types.js";
+import type { LiveKitTokenResponse, ErrorSource, CommandDefinition, CommandResult, VoiceSessionOptions } from "./types.js";
 /**
  * A deferred promise with timeout support.
  * Encapsulates resolve/reject/timeout in a single object for cleaner async handling.
@@ -31,21 +31,38 @@ declare class LiveKitManager {
     private room;
     private tokenData;
     private micTrack;
+    private cachedTokenData;
+    private tokenCacheTimestamp;
+    private tokenPrefetchPromise;
+    private tokenRefreshTimer;
+    private autoRefreshEnabled;
     private pendingTranscript;
     private pendingEditText;
+    private pendingCommand;
     private pendingTrackSubscribed;
-    private preWarmPromise;
     private editOriginalText;
+    private sessionSettings;
     /**
-     * Pre-warm resources for faster connection
-     * Call this when user shows intent (e.g., expands widget)
-     * Only fetches token - mic permission is requested when user clicks Dictate
+     * Check if the cached token is still valid (within TTL)
      */
-    preWarm(): Promise<void>;
+    private isCachedTokenValid;
+    /**
+     * Pre-fetch a LiveKit token for later use
+     * Call this early (e.g., when widget expands) to reduce latency when starting a voice session.
+     * If a prefetch is already in progress, returns the existing promise.
+     * If a valid cached token exists, returns it immediately.
+     */
+    prefetchToken(): Promise<LiveKitTokenResponse>;
     /**
      * Fetch a LiveKit token from the backend
+     * Uses cached token if valid, otherwise fetches a fresh one.
+     * Includes language settings and user vocabulary which are stored in the VoiceSession.
      */
     fetchToken(): Promise<LiveKitTokenResponse>;
+    /**
+     * Internal method to fetch a fresh token from the server
+     */
+    private fetchTokenFromServer;
     /**
      * Connect to a LiveKit room (fresh connection each time)
      */
@@ -65,8 +82,13 @@ declare class LiveKitManager {
     private handleDataMessage;
     /**
      * Publish microphone audio track
+     * Uses the device ID from session settings if set
      */
     enableMicrophone(): Promise<void>;
+    /**
+     * Log information about the current microphone track
+     */
+    private logMicrophoneInfo;
     /**
      * Disable microphone audio track
      */
@@ -76,10 +98,24 @@ declare class LiveKitManager {
      */
     sendDataMessage(message: object): Promise<void>;
     /**
-     * Start a voice session
-     * Connects to room, enables microphone, and waits for agent to subscribe to our track
+     * Start a voice session with pre-connect audio buffering
+     * Fetches a fresh token, then enables mic with preConnectBuffer to capture audio while connecting.
+     * Agent subscription happens in the background - we don't block on it.
+     *
+     * @param options - Session options including action type and parameters
      */
-    startVoiceSession(): Promise<void>;
+    startVoiceSession(options?: VoiceSessionOptions): Promise<void>;
+    /**
+     * Wait for the agent to subscribe to our audio track in the background
+     * Handles timeout errors without blocking the main flow
+     */
+    private waitForAgentSubscription;
+    /**
+     * Enable microphone with pre-connect buffering
+     * This starts capturing audio locally before the room is connected,
+     * buffering it until the connection is established.
+     */
+    private enableMicrophoneWithPreConnectBuffer;
     /**
      * Stop the voice session and request the transcript
      * Returns a promise that resolves with the transcript text
@@ -101,11 +137,47 @@ declare class LiveKitManager {
      * Alias for requestEditText - granular API naming
      */
     stopAndEdit(originalText: string): Promise<string>;
+    /**
+     * Request command matching using the transcript as input
+     * Sends command definitions to the backend, which matches the user's speech against them
+     * Returns a promise that resolves with the matched command or null if no match
+     * @throws Error if timeout occurs waiting for command result
+     */
+    requestCommand(commands: CommandDefinition[]): Promise<CommandResult | null>;
+    /**
+     * Alias for requestCommand - granular API naming
+     */
+    stopAndCommand(commands: CommandDefinition[]): Promise<CommandResult | null>;
     /**
      * Disconnect from the current room
      * Clears the token so a fresh one is fetched for the next session
      */
     disconnect(): Promise<void>;
+    /**
+     * Invalidate the cached token
+     * Call this when settings change that would affect the token (language, vocabulary)
+     */
+    invalidateTokenCache(): void;
+    /**
+     * Start auto-refreshing the token while the widget is expanded.
+     * Call this after a voice session completes to immediately fetch a fresh token
+     * (since each command requires its own token) and keep it fresh for subsequent commands.
+     */
+    startAutoRefresh(): void;
+    /**
+     * Stop auto-refreshing the token.
+     * Call this when the widget collapses or user navigates away.
+     */
+    stopAutoRefresh(): void;
+    /**
+     * Schedule a token refresh before the current cache expires.
+     * Handles computer sleep by checking elapsed time on each refresh attempt.
+     */
+    private scheduleTokenRefresh;
+    /**
+     * Perform the auto-refresh, handling computer sleep scenarios.
+     */
+    private performAutoRefresh;
     /**
      * Get the current room instance
      */
@@ -122,11 +194,6 @@ declare class LiveKitManager {
      * Check if microphone is enabled
      */
     isMicrophoneEnabled(): boolean;
-    /**
-     * Clear the cached token
-     * Used when user identity changes to ensure next session gets a fresh token
-     */
-    clearToken(): void;
 }
 export declare const livekit: LiveKitManager;
 export {};

package/dist/speechos.d.cts

@@ -4,7 +4,7 @@
  * Provides both low-level and high-level APIs for voice interaction.
  * This is the main entry point for headless usage of SpeechOS.
  */
-import type { SpeechOSConfig } from "./types.js";
+import type { SpeechOSCoreConfig, CommandDefinition, CommandResult } from "./types.js";
 import { state } from "./state.js";
 import { events } from "./events.js";
 /**
@@ -20,7 +20,7 @@ declare class SpeechOSCore {
      * Initialize the SDK with configuration
      * @param config - Configuration options including apiKey
      */
-    init(config: SpeechOSConfig): void;
+    init(config: SpeechOSCoreConfig): void;
     /**
      * Check if the SDK is initialized
      */
@@ -84,6 +84,22 @@ declare class SpeechOSCore {
      * Call this after edit() when user stops speaking
      */
     stopEdit(): Promise<string>;
+    /**
+     * One-shot command: connect, wait for agent, record voice, match against commands
+     * Automatically handles the full voice session lifecycle
+     *
+     * @param commands - Array of command definitions to match against
+     * @returns The matched command result or null if no match
+     */
+    command(commands: CommandDefinition[]): Promise<CommandResult | null>;
+    private _commandCommands?;
+    private _commandResolve?;
+    private _commandReject?;
+    /**
+     * Stop command recording and get the matched command
+     * Call this after command() when user stops speaking
+     */
+    stopCommand(): Promise<CommandResult | null>;
     /**
      * Cancel the current operation
      */
@@ -99,7 +115,7 @@ declare class SpeechOSCore {
     /**
      * Get the current config
      */
-    getConfig(): SpeechOSConfig;
+    getConfig(): SpeechOSCoreConfig;
     private ensureInitialized;
     private cleanup;
     /**

package/dist/speechos.d.ts

@@ -4,7 +4,7 @@
  * Provides both low-level and high-level APIs for voice interaction.
  * This is the main entry point for headless usage of SpeechOS.
  */
-import type { SpeechOSConfig } from "./types.js";
+import type { SpeechOSCoreConfig, CommandDefinition, CommandResult } from "./types.js";
 import { state } from "./state.js";
 import { events } from "./events.js";
 /**
@@ -20,7 +20,7 @@ declare class SpeechOSCore {
      * Initialize the SDK with configuration
      * @param config - Configuration options including apiKey
      */
-    init(config: SpeechOSConfig): void;
+    init(config: SpeechOSCoreConfig): void;
     /**
      * Check if the SDK is initialized
      */
@@ -84,6 +84,22 @@ declare class SpeechOSCore {
      * Call this after edit() when user stops speaking
      */
     stopEdit(): Promise<string>;
+    /**
+     * One-shot command: connect, wait for agent, record voice, match against commands
+     * Automatically handles the full voice session lifecycle
+     *
+     * @param commands - Array of command definitions to match against
+     * @returns The matched command result or null if no match
+     */
+    command(commands: CommandDefinition[]): Promise<CommandResult | null>;
+    private _commandCommands?;
+    private _commandResolve?;
+    private _commandReject?;
+    /**
+     * Stop command recording and get the matched command
+     * Call this after command() when user stops speaking
+     */
+    stopCommand(): Promise<CommandResult | null>;
     /**
      * Cancel the current operation
      */
@@ -99,7 +115,7 @@ declare class SpeechOSCore {
     /**
      * Get the current config
      */
-    getConfig(): SpeechOSConfig;
+    getConfig(): SpeechOSCoreConfig;
     private ensureInitialized;
     private cleanup;
     /**

package/dist/state.d.cts

@@ -9,9 +9,12 @@ import type { SpeechOSState, StateChangeCallback, UnsubscribeFn } from "./types.
 declare class StateManager {
     private state;
     private subscribers;
+    /** Cached immutable snapshot for useSyncExternalStore compatibility */
+    private snapshot;
     constructor(initialState: SpeechOSState);
     /**
-     * Get the current state (returns a copy to prevent mutations)
+     * Get the current state snapshot (returns a stable reference for React)
+     * This returns an immutable frozen object that only changes when setState is called.
      */
     getState(): SpeechOSState;
     /**

package/dist/state.d.ts

@@ -9,9 +9,12 @@ import type { SpeechOSState, StateChangeCallback, UnsubscribeFn } from "./types.
 declare class StateManager {
     private state;
     private subscribers;
+    /** Cached immutable snapshot for useSyncExternalStore compatibility */
+    private snapshot;
     constructor(initialState: SpeechOSState);
     /**
-     * Get the current state (returns a copy to prevent mutations)
+     * Get the current state snapshot (returns a stable reference for React)
+     * This returns an immutable frozen object that only changes when setState is called.
      */
     getState(): SpeechOSState;
     /**

package/dist/types.d.cts

@@ -18,22 +18,60 @@ export interface ServerErrorMessage {
  */
 export type ErrorSource = "init" | "connection" | "timeout" | "server";
 /**
- * Configuration options for initializing SpeechOS
+ * Backend type for voice sessions
+ * - 'websocket': Direct WebSocket connection (lower latency, recommended)
+ * - 'livekit': LiveKit WebRTC connection (legacy)
  */
-export interface SpeechOSConfig {
-    /** API key for authentication with SpeechOS backend */
-    apiKey?: string;
+export type VoiceBackend = "websocket" | "livekit";
+/**
+ * Configuration options for initializing SpeechOS Core
+ */
+export interface SpeechOSCoreConfig {
+    /** API key for authentication with SpeechOS backend (required) */
+    apiKey: string;
     /** Optional user identifier for tracking which end user is using the SDK */
     userId?: string;
     /** Backend host URL for API calls (default: https://app.speechos.ai) */
     host?: string;
-    /** Position of the widget on screen (used by client package) */
-    position?: "bottom-center" | "bottom-right" | "bottom-left";
-    /** Custom z-index for widget overlay (used by client package) */
-    zIndex?: number;
     /** Enable debug logging */
     debug?: boolean;
 }
+/**
+ * Session settings passed when starting a voice session
+ * Contains user preferences for transcription and processing
+ */
+export interface SessionSettings {
+    /** Input language code for speech recognition (e.g., "en-US", "es", "fr") */
+    inputLanguageCode?: string;
+    /** Output language code for transcription formatting */
+    outputLanguageCode?: string;
+    /** Whether to apply AI formatting (removes filler words, adds punctuation) */
+    smartFormat?: boolean;
+    /** Custom vocabulary terms to improve transcription accuracy */
+    vocabulary?: string[];
+    /** Text snippets with trigger phrases that expand to full text */
+    snippets?: Array<{
+        trigger: string;
+        expansion: string;
+    }>;
+    /** Audio input device ID (empty string for system default) */
+    audioDeviceId?: string;
+}
+/**
+ * Options for starting a voice session
+ */
+export interface VoiceSessionOptions {
+    /** Callback when microphone is ready and capturing */
+    onMicReady?: () => void;
+    /** Action type for this session */
+    action?: SpeechOSAction;
+    /** Text to edit (for edit action) */
+    inputText?: string;
+    /** Command definitions (for command action) */
+    commands?: CommandDefinition[];
+    /** User settings for this session */
+    settings?: SessionSettings;
+}
 /**
  * LiveKit token response from the backend
  */
@@ -43,10 +81,59 @@ export interface LiveKitTokenResponse {
     room: string;
     identity: string;
 }
+/**
+ * User vocabulary data sent with transcription/edit requests
+ * Includes custom vocabulary terms for improved transcription accuracy
+ * and text snippets that can be expanded from trigger phrases
+ */
+export interface UserVocabularyData {
+    /** Custom vocabulary terms to improve transcription of domain-specific words */
+    vocabulary: string[];
+    /** Text snippets with trigger phrases that expand to full text */
+    snippets: Array<{
+        /** Short trigger phrase the user speaks */
+        trigger: string;
+        /** Full text to expand the trigger into */
+        expansion: string;
+    }>;
+}
 /**
  * Available actions that can be triggered from the widget
  */
-export type SpeechOSAction = "dictate" | "edit";
+export type SpeechOSAction = "dictate" | "edit" | "command";
+/**
+ * Definition of a command argument
+ */
+export interface CommandArgument {
+    /** Name of the argument (used as key in the result) */
+    name: string;
+    /** Description of what this argument represents */
+    description: string;
+    /** Type of the argument value */
+    type?: "string" | "number" | "integer" | "boolean";
+    /** Whether this argument is required (default: true) */
+    required?: boolean;
+}
+/**
+ * Definition of a command that can be matched
+ */
+export interface CommandDefinition {
+    /** Unique name/identifier for the command */
+    name: string;
+    /** Description of what this command does (helps LLM match intent) */
+    description: string;
+    /** Arguments that can be extracted from the user's speech */
+    arguments?: CommandArgument[];
+}
+/**
+ * Result of a successful command match
+ */
+export interface CommandResult {
+    /** Name of the matched command */
+    name: string;
+    /** Extracted argument values */
+    arguments: Record<string, unknown>;
+}
 /**
  * Recording/dictation states
  */
@@ -109,6 +196,10 @@ export interface SpeechOSEventMap {
         text: string;
         originalText: string;
     };
+    /** Emitted when command matching completes (null if no command matched) */
+    "command:complete": {
+        command: CommandResult | null;
+    };
     /** Emitted when transcribed text is inserted into a form field */
     "transcription:inserted": {
         text: string;
@@ -120,6 +211,11 @@ export interface SpeechOSEventMap {
         editedContent: string;
         element: HTMLElement;
     };
+    /** Emitted when user settings change (language, snippets, vocabulary, smartFormat) */
+    "settings:changed": {
+        /** Type of setting that changed */
+        setting: "language" | "snippets" | "vocabulary" | "smartFormat";
+    };
     /** Emitted when an error occurs */
     error: {
         code: string;