livekit/livekit

Features

• Realtime AI Session Managers - Provides a platform for building low-latency, multimodal AI agents that process audio, video, and text in live communication sessions.
• Real-Time Media Servers - Provides programmable infrastructure for routing, scaling, and managing real-time audio and video streams across global networks.
• Agent Orchestrators - Manages, coordinates, and executes autonomous agent workflows and their associated persistent data.
• Voice Agents - Orchestrates low-latency, bidirectional voice agents with automatic interruption handling.
• Realtime Voice Conversation Facilitators - Facilitates low-latency, two-way spoken interactions by connecting to language models capable of natural dialogue.
• Voice Activity Detection - Provides real-time voice activity detection to identify when a user is speaking, enabling natural turn-taking and interruption handling.
• Multimodal Agent Capabilities - Processes speech, video, and text to enable real-time, multimodal interactions between users and automated agents.
• Speech-to-Text Integrations - Sequences speech-to-text, language models, and text-to-speech pipelines into interactive, context-aware conversational workflows.
• Conversational Audio Streams - Provides bidirectional audio streaming pipelines for natural, low-latency conversational voice interactions.
• Voice-Enabled Agents - Provides real-time voice interaction capabilities including preemptive response generation to minimize latency in conversational AI agents.
• Virtual Room Orchestrators - Manages virtual spaces where participants interact through media tracks and triggers events to monitor activity in real time.
• Selective Forwarding Units - Routes media streams between participants and agents using a selective forwarding unit to manage signaling, network traversal, and quality of service.
• Telephony Management Systems - Integrates AI agents into public telephone networks via SIP trunking and automated call management.
• Realtime Avatar Renderers - Streams synchronized, interactive virtual characters into voice-based AI applications for lifelike visual communication.
• Agent Deployment Management - Provides comprehensive infrastructure for hosting, managing, and scaling agent deployments in production environments.
• Agent Lifecycle Management - Manages the full lifecycle of automated agents, including initialization, execution, and graceful shutdown.
• Programmatic Participants - Enables building programmatic participants that join rooms to process media and interact with users.
• Conversational Turn Detection - Analyzes speech context to determine if a user has finished speaking or intends to continue, preventing premature interruptions.
• Session Lifecycle Managers - Manages the lifecycle of voice AI interactions by coordinating speech-to-text, language models, and text-to-speech pipelines.
• AI Agent Infrastructure - Provides back-end systems and registries that support the deployment, integration, and tracking of AI agents.
• Language Model Integrations - Connects voice agents to external language models to process conversational input and generate intelligent, context-aware responses.
• Voice Activity Detection - Identifies when a user is speaking to enable natural turn-taking and interruption handling.
• AI Model Integrations - Integrates various AI models through managed inference services for real-time conversational processing.
• Unified Model Interfaces - Connects voice agents to various speech-to-text, large language, and text-to-speech models through a unified interface for real-time conversational processing.
• Unified Speech Pipelines - Converts spoken audio to text and generates lifelike speech using integrated AI models within a unified pipeline.
• Speech-to-Text Services - Converts spoken audio into written text while identifying individual participants to distinguish between speakers.
• Speech-to-Text Utilities - Transcribes spoken audio into written text using specialized models for regional languages with real-time streaming support.
• Conversational Pipelines - Processes audio by chaining speech-to-text transcription, language model generation, and text-to-speech synthesis to provide modular control over each stage of the conversation.
• Room Provisioning Services - Creates a persistent or temporary virtual space for participants to join, with configurable limits on participant count and automatic cleanup timers.
• Room Management - Creates, lists, and deletes communication rooms while allowing configuration of participant limits, timeouts, and automated recording or streaming settings.
• Turn Completion Detection - Analyzes audio input to identify when a user has finished speaking, using configurable detection modes and endpointing delays.
• WebRTC Facilitators - Manages real-time audio, video, and data streams between participants and automated agents using secure, scalable communication protocols.
• Agent Handoffs - Passes session authority between specialized agents to enforce distinct roles or updated instructions during a conversation.
• Agent Deployment Servers - Deploys agent services in production with load balancing, overload protection, and graceful shutdown capabilities.
• Agentic Model Integrations - Connects conversational agents to external language models via managed services or API keys.
• Interruption Sensitivity Configuration - Adjusts timing windows around agent speech to prevent misidentifying user corrections or late transcripts as background noise.
• Conversation Validators - Executes conversation turns to verify agent behavior, including tool calls, handoffs, and multi-turn interactions.
• Agent Evaluation Frameworks - Assesses agent decision-making, action success, and conversation quality through automated scoring and feedback loops.
• Preemptive Response Generation - Starts processing before a user finishes speaking to reduce perceived latency in back-and-forth conversations.
• Agent Tool Execution - Enables agents to execute custom user-defined functions and tools during live voice conversations.
• Model Provider Integrations - Integrates various language and speech services through plugins or managed infrastructure to power agent intelligence.
• AI Agent Integration Platforms - Provides a platform for integrating automated AI agents into real-time communication sessions.
• AI Hosting Platforms - Offers managed infrastructure for hosting and scaling interactive AI agents that process real-time audio, video, and data streams.
• Audio Processing - Applies noise cancellation and signal conditioning to input audio to ensure high-quality voice recognition and interaction.
• Detection Model Configurations - Configures automated speech activity detection models to determine user turn completion.
• Language Model Integrations - Connects to hosted language models to generate conversational responses within agent workflows.
• Large Language Models - Connects conversational agents to external language models for real-time text processing and response generation.
• LLM Provider Integrations - Provides a unified interface for connecting to diverse language model providers with support for automatic fallbacks.
• Local AI Inference - Executes machine learning models directly on local hardware resources to ensure privacy and reduce latency.
• Realtime Processing Pipelines - Handles audio input and output within a single model to minimize latency and improve the expressiveness of conversational responses.
• User Interruption Detection - Distinguishes between intentional user interruptions and conversational backchanneling to ensure agents respond appropriately.
• Realtime Avatar Integration - Integrates photorealistic digital characters into voice applications to provide synchronized visual feedback and lifelike presence during conversations.
• Speaker Diarization - Identifies individual speakers within mixed audio tracks to isolate voices and improve transcription accuracy.
• Speech Synthesis Services - Provides interfaces for connecting to third-party speech synthesis providers to generate high-quality audio output.
• Text-to-Speech - Converts text into natural-sounding audio for voice agents using specialized models.
• Speech-to-Speech Models - Processes audio input directly through specialized models to generate natural spoken output without intermediate text conversion.
• Tool Discovery - Loads tool definitions on demand based on model queries to reduce token usage and maintain accuracy.
• Graceful Shutdowns - Ensures clean termination of services by allowing in-flight requests to complete before shutting down.
• Speech-to-Text Pipelines - Chains specialized speech-to-text, language, and text-to-speech services to build high-performance, modular voice processing workflows.
• Video Input Processing - Captures live video frames from user cameras or screen shares to provide real-time visual context to AI models.
• Media Streaming - Publishes and subscribes to audio, video, and data streams, supporting multiple simultaneous inputs like camera, microphone, and screen sharing.
• Participant Management - Retrieves, updates, or removes participants from a room and modifies their metadata or permissions to control access and session behavior.
• Programmatic Participants - Executes custom code that joins real-time media rooms to process audio, video, or data streams as an active participant.
• Token-Based Authentication - Validates session connections using cryptographically signed tokens that encode participant identity and permissions.
• Agent Dispatching - Triggers the connection of automated agents to rooms via API calls or configuration rules.
• Connection Establishment Protocols - Facilitates the initiation of secure, low-latency connections to virtual rooms for media and data exchange.
• SIP Bridges - Connects third-party SIP services to rooms to enable inbound and outbound calling capabilities.
• Telephony Services - Connects voice agents to public telephone networks using SIP trunks to facilitate inbound and outbound calling.
• Session Lifecycle Management - Handles the graceful termination of active sessions, including participant disconnection and resource cleanup.
• Agent Performance Monitoring - Tracks real-time metrics, session status, and error logs through a centralized dashboard to diagnose and resolve operational issues.
• Graceful Shutdown Mechanisms - Finishes active requests before closing connections during a server shutdown to prevent data loss.
• Conversational Test Suites - Validates conversational responses and logic by simulating user input and asserting that the agent meets specific intent requirements.
• Interactive Video Avatar Generators - Integrates photorealistic digital characters into conversational AI sessions to provide visual feedback and human-like interaction during voice-based exchanges.
• Agent Orchestration - Orchestrates the lifecycle of voice AI interactions, from input collection to output delivery.
• Agent Session Management - Stores and retrieves session data across agent transitions to maintain continuity.
• Warm Transfer Orchestration - Dials a human agent via telephony protocols, manages hold music, and hands off conversation context for seamless handovers.
• Service Registrars - Connects agent code to a central server to wait for incoming job requests and report current capacity.
• Agent Capability Extensions - Integrates external tools and functions, including Model Context Protocol servers, to allow agents to perform actions or retrieve data.
• Speech Initiation - Triggers the agent to speak a predefined message or generate a dynamic response to start or continue a conversation.
• Multi-Agent Session Facilitators - Hands off an active conversation to a different agent dynamically to handle specific tasks or user requests.
• Agent Tool Definitions - Creates modular, namespace-prefixed functions for agent capabilities.
• Agent Configuration Schemas - Provides standardized formats for defining agent behavior, persona, identity, and conversational flow using natural language prompts.
• Conversation History Management - Maintains and passes conversation history between agents during handoffs to ensure continuity.
• Speech State Monitors - Monitors whether an agent is currently speaking to coordinate follow-up actions or graceful session termination.
• External Service Integrations - Connects AI agents to third-party memory, storage, and processing services to extend conversational capabilities.
• Media System Integrators - Connects sessions to external recording, streaming, and telephony networks through dedicated service interfaces.
• External Tool Execution - Supports the execution of external model-provider tools like web search and data retrieval within agent workflows.
• Multimodal Input Handlers - Handles audio and video streams from participants, including noise cancellation and adaptive frame rate sampling.
• Local Model Integrations - Connects voice agents to locally hosted language models via compatible APIs to process text without cloud dependencies.
• Synthetic Speech Generation - Converts text into natural-sounding audio using cloud-based voice synthesis services.
• Model Provider Configurations - Executes external tasks like web searching directly through model provider infrastructure.
• Multimodal AI Pipelines - Routes media and data streams through modular pipelines to enable multimodal AI interactions.
• Full-Duplex Multimodal Interaction - Coordinates simultaneous input and output across multiple modalities for natural communication.
• Speech Synthesis - Overrides default audio generation by routing speech synthesis through dedicated text-to-speech engines.
• Synchronized Transcription Streams - Streams agent speech as text to frontend clients in sync with audio playback, providing live captions and transcript data.
• Communication - Open-source WebRTC SDKs for low-latency video and data streaming.
• Media and Communication - Scalable WebRTC SFU for real-time video.
• Video Conferencing - Modern platform for scalable WebRTC conferencing with SDKs.
• Voice And Video - Listed in the “Voice And Video” section of the Awesome Selfhosted awesome list.
• Agent-Integrated Functions - Triggers user-defined functions to perform side effects or interact with external systems during a conversation.
• Unified Transcription Pipelines - Converts spoken audio into text by routing requests to multiple third-party providers through a unified interface.
• Audio Input Cleaning - Applies noise cancellation, automatic gain control, and voice isolation to audio streams to improve speech recognition.
• Media Client SDKs - Enables consistent media track exchange across web, mobile, and game engines through a unified programming model.
• Inbound Call Routers - Directs incoming calls to specific rooms using configurable dispatch rules and custom participant attributes.
• Network Reliability - Automatically negotiates network paths through TURN relays to maintain stable media streams across restrictive firewalls.
• Outbound Call Initiators - Initiates phone calls programmatically by passing trunk configuration or using stored credentials to create participants.
• Remote Procedure Calls - Routes tool calls from the server to a connected client via remote procedure calls to perform actions.
• Telephony Provisioning Services - Connects public telephone networks to communication rooms using native phone numbers or external SIP trunking services.
• End-to-End Encryption - Applies end-to-end encryption to data in transit and manages access through token-based authentication.
• Token-Based Authentication - Validates connections to virtual rooms using cryptographically signed tokens that encode participant identity and permission levels.
• Agent Entrypoint Registrars - Registers primary functions for job execution on agent servers.
• Streaming Latency Optimizers - Transmits synthesized speech over persistent connections to reduce latency and provide word-level timing data.
• AI Session Monitoring - Captures transcripts, audio recordings, and interaction logs from live sessions to identify performance issues and refine agent instructions.
• Workflow Orchestrators - Sequences multiple tasks into ordered groups that share conversation context for complex, multi-step interactions.
• Event-Driven Agent Loops - Manages the initialization, execution, and graceful shutdown of automated participants through a centralized job management framework.
• Multi-Agent Coordination Systems - Sequences ordered tasks into groups that share context, support backtracking, and return summarized results.
• Utterance Scheduling - Enforces timing and cadence constraints on agent speech to ensure natural conversational flow.
• Agent Plugin Frameworks - Customizes agent behaviors by swapping out components like text-to-speech or language models for individual tasks.
• Agent Task Execution - Executes short-lived, scoped units of work that perform specific objectives and return typed results to the controlling agent.
• Permission Configurators - Restricts or grants capabilities for agents to publish or subscribe to media and data streams within a session.
• Reasoning Effort Configurations - Adjusts the reasoning depth of supported models to control the complexity of generated responses.
• Agent Task Initiations - Triggers proactive spoken responses from the agent on demand during active conversational sessions.
• Conversational and Voice Interaction - Applies background noise cancellation, preemptive response generation, and turn detection to improve conversational responsiveness and clarity.
• Execution Control Policies - Controls agent tool-calling behavior with step limits and timeouts to prevent infinite loops or hanging processes.
• Agentic Task Orchestrators - Models complex, multi-step processes by chaining together specialized agents, reusable tasks, and handoff logic.
• Audio Noise Cancellation - Processes audio streams in realtime to remove background noise and improve speech clarity for participants and automated agents.
• Context Injection - Retrieves and adds relevant information to the conversation history in real-time based on user input to improve response accuracy.
• Conversation History Management - Maintains and manipulates the history of a conversation to control the information provided to language models.
• Conversation State Persistence - Maintains and manipulates ordered message histories to provide language models with the necessary context for coherent interactions.
• Conversation State Managers - Exposes real-time events for tracking speaking states and interruption status to manage conversational flow.
• External Tool Integration - Exposes external functions to an AI model, allowing it to perform actions or retrieve data from external systems during a conversation.
• Indic - Converts text into natural-sounding audio for Indian languages and English with configurable speaking styles.
• Knowledge Retrieval Tools - Performs web searches and queries knowledge bases to retrieve information during a conversation.
• Language Model Optimizers - Optimizes payload delivery and encoding for language models to minimize response latency in voice agents.
• Multilingual Speech Translation - Converts spoken input into a different language during a live session to facilitate real-time multilingual communication.
• Custom Tool Logic Interfaces - Implements specialized toolsets with custom initialization and cleanup routines to manage external resources.
• Prompt Caches - Caches persistent context to reuse static prompts and documents, reducing token costs.
• Multilingual Transcription - Converts spoken audio into text for Indian languages, English, and code-mixed speech using specialized models.
• Hybrid Synthesis Pipelines - Pairs real-time speech-understanding models with separate text-to-speech engines for balanced performance.
• Speech Synthesis Gateways - Streams audio output from multiple third-party providers through a unified interface to simplify integration.
• Speech Synthesis Markup Controls - Applies markup tags to text input to control prosody, emphasis, and pronunciation for more natural and engaging synthetic audio output.
• Structured Data Collection Workflows - Executes prebuilt voice workflows to capture and validate information like names, contact details, and dates from spoken input.
• Tool Result Transformers - Intercepts and modifies data returned by external tools to manage context size and formatting for language models.
• Media Stream Subscriptions - Controls the reception of media streams by enabling or disabling subscriptions to specific tracks, either automatically or through manual selection.
• Session Recorders - Triggers recording processes automatically upon room creation to capture all participant activity.
• Session Export Utilities - Records or streams live room content, individual participant feeds, or specific media tracks to files, HLS segments, or external RTMP services.
• Response Streaming - Flushes segments of generated text to the speech engine immediately to provide low-latency feedback.
• Telephony Input Capture - Collects numerical data from callers using spoken digits or dual-tone multi-frequency signals for menu navigation.
• Telephony Number Provisioning - Purchases and manages local or toll-free phone numbers for use within telephony applications.
• DTMF Signaling - Transmits DTMF tones to external phone systems to navigate automated menus or interact with telephony-based services.
• Concurrent Task Limiters - Controls parallel task execution to manage system resource consumption.
• Cleanup Hooks - Schedules automatic resource release upon function exit.
• Event Monitoring Systems - Triggers webhooks and event notifications to track and respond to state changes involving rooms, participants, and media streams in real time.
• Session Termination - Disconnects an automated agent from a call and removes the room to ensure resources are freed after a conversation.
• Agent Debugging Interfaces - Monitors agent events, tool execution, and model performance in real-time through a visual interface to analyze behavior and timing.
• Avatars - Generates real-time interactive virtual avatars for voice AI applications.
• Conversational Avatar Animators - Integrates animated virtual avatars that synchronize facial expressions with agent audio output.
• Conversational Interface Embeds - Adds a voice or chat interface to any website by inserting a script tag that renders a launcher button and interaction pop-up.
• Media Controllers - Provides programmatic APIs for managing media playback state and subscription status.
• Speech Synthesis Providers - Connects external text-to-speech services through modular plugins to access provider-specific features or custom account configurations.
• Agent Context Management - Pre-populates an agent session with user-specific data or task metadata before the conversation begins.
• Agent Integration Adapters - Connects external graph-based agent workflows to voice sessions using integration adapters for real-time interaction.
• Agent State Persistence - Persists task execution state across agent handoffs to ensure continuity.
• Persistent Agent Integrators - Connects voice agents to stateful language models that maintain long-term memory and autonomous task execution.
• Tool Access Controls - Limits which tools an agent can access for a single specific interaction to enforce staged workflows or security.
• Turn Duration Limiters - Provides configurable limits on user turn length to ensure balanced and responsive voice conversations.
• Connection Monitors - Tracks network health metrics like latency and packet loss for each participant to surface signal strength or connection status in the user interface.
• Wake Word Detection - Identifies specific wake words in audio streams to activate voice agents hands-free without requiring manual user input.
• Voice Profiling - Extracts metadata from audio streams including age, emotion, pitch, and gender to provide insights into the speaker's vocal profile.
• Text to Speech Services - Converts text to speech using a curated selection of models with integrated billing and rate limiting.
• Transcription Model Selectors - Allows choosing the optimal model for specific languages or specialized transcription tasks.
• Transcription Buffering - Buffers audio segments using voice activity detection to support models that cannot process streaming input in real time.
• Transcription Timing Synchronizers - Forwards aligned transcript data alongside generated audio to ensure precise timing between speech and text.
• External Knowledge Integrators - Connects agents to external databases and APIs to perform retrieval-augmented generation and access real-time data.
• Tool Mocking - Simulates tool outputs and error conditions during testing to isolate agent logic from external dependencies.
• Visual Response Generation - Streams images or renders virtual avatars to provide lifelike visual responses to users during active communication sessions.
• Voice Cloning Tools - Creates synthetic voice profiles from short audio samples to enable personalized speech generation.
• Provider Failover Mechanisms - Switches automatically to an alternative provider if the primary text-to-speech service fails, maintaining the session using a pre-synchronized voice clone.
• Human Approval Gates - Requires explicit confirmation or state validation within tool logic to ensure sensitive operations only occur after user verification.
• Language Detection Tools - Identifies the spoken language of an audio stream in real-time, including support for mid-stream language switching.
• Conversation History Condensation - Condenses long conversation logs into a brief summary before transferring to a new agent to optimize context window usage.
• Model Inference Servers - Connects to speech models through a hosted service to eliminate the need for individual API keys and reduce latency.
• Model Execution Tools - Enables language models to perform external tasks like web searching, file analysis, or sandboxed code execution during a conversation.
• Model Response Aggregation - Awaits and aggregates complete streaming outputs including text, tool invocations, and usage metadata.
• Model Task Retries - Defines a sequence of alternative speech or language models to automatically retry requests if the primary service fails or becomes unavailable.
• Custom Model Logic Interfaces - Allows developers to define bespoke model behavior through standard interfaces for specialized processing requirements.
• Structured Output Parsers - Parses complex JSON payloads from model responses to separate control directives from the actual spoken content.
• Text Transformation Pipelines - Applies custom or built-in filters to modify text output before it reaches the speech synthesis engine.
• Local Speech Synthesis - Converts text to spoken audio using an open-weight model that runs on local infrastructure.
• Speech Synthesis Markup - Modifies synthesized audio output using inline tags to control pacing, pauses, and vocal styles.
• Pipeline Customizers - Modifies agent behavior at specific stages of the processing pipeline by injecting custom logic for speech or language events.
• Media Track Subscriptions - Controls which media tracks are captured by an active recording process by dynamically updating subscription settings during the session.
• Conversational Feedback Tools - Plays audio responses or filler messages during long-running tool calls to maintain conversational flow.
• Process Prewarming - Initializes worker processes and loads heavy resources in advance to reduce latency when starting new jobs.
• Task Interruption Policies - Configures whether background tasks are cancelled or allowed to finish when a user interrupts the agent.
• Cloud Agent Deployers - Automates the containerization and registration of agent services to cloud infrastructure for production hosting.
• Job Filtering - Evaluates incoming job requests to accept or reject them based on server capacity or specific job requirements.
• Load Balancing Policies - Monitor resource utilization or concurrency to dynamically throttle the acceptance of new jobs based on custom thresholds.
• Audio Processing - Applies real-time transformations like time-stretching and volume adjustment to audio streams before delivery.
• Facial Expression Modulators - Injects specific emotional states into avatar performance to influence facial expressions during conversation.
• Automatic Gain Controls - Adjusts incoming audio volume automatically to ensure consistent input quality regardless of microphone distance.
• Playback Synchronizers - Coordinates the timing between agent audio generation and avatar visual playback for accurate lip-syncing.
• Broadcast Messaging - Sends custom data payloads to participants in a room using reliable or lossy delivery modes.
• Participant Interaction Hooks - Links agents to specific room participants and manages session behavior based on user presence.
• Participant Transfer Services - Supports call transfers and multi-room broadcasting by moving participant media tracks between rooms.
• Voice Platform Integrations - Streams bidirectional audio between external messaging or voice platforms and rooms without requiring SIP infrastructure.
• Managed Inference Authentication - Accesses speech recognition services through a managed inference layer to eliminate the need for individual provider API keys.
• System Usage Monitoring - Tracks media quality, agent behavior, and usage metrics through a centralized dashboard and analytics interface.
• Agent Input and Output Validators - Executes automated conversational test cases against user inputs to verify that agent responses align with expected outcomes.
• LLM-Driven Avatar Frameworks - Adjusts the facial state of a digital character dynamically during conversation via direct commands or LLM triggers.
• Visual Input Integration - Processes live video and image data to provide visual context for automated responses.
• Input Processing Logic - Monitors incoming chat messages to trigger agent responses and allows for custom handling logic.

Star history