Own the voice agent runtime.
A source-checked map of 9 open frameworks and libraries for realtime voice agents. Licenses and maintenance signals checked 2026-07-17.
Framework is not the same as platform
Some projects provide a media pipeline; others include telephony, a visual builder, or deployment. The missing operational layers still become part of your cost.
TEN Framework
A cross-language framework and runtime for realtime conversational voice agents, with reusable extensions and a visual agent builder.
02Pipecat
A vendor-neutral framework for realtime voice and multimodal agents with broad STT, LLM, TTS, transport, and telephony integrations.
03LiveKit Agents
A realtime agent framework built around programmable participants, open WebRTC clients, model plugins, turn detection, and media rooms.
04Bolna
An end-to-end voice-agent orchestration framework for ASR, LLM, TTS, WebSockets, and telephony, with an optional hosted platform.
05Dograh
A self-hostable voice AI platform with a visual workflow builder, BYOK speech stacks, telephony, campaigns, and Docker deployment.
06VideoSDK AI Agents
A production-oriented framework for voice and multimodal agents that join VideoSDK rooms and connect models, workers, and client devices.
07Vocode Core
A modular library for streaming LLM conversations across microphones, phone calls, and meetings with replaceable speech providers.
08FastRTC
A Python library for adding low-latency WebRTC and WebSocket audio or video streams to applications, including voice-agent patterns.
09OpenAI Agents SDK for Voice
The open Agents SDK includes voice pipeline patterns for connecting speech input, agent workflows, tools, and streamed audio output.
Choose the boundary you want to own
A lower platform bill can mean a larger engineering and operations bill. Start with media ownership, telephony, deployment, and compliance before comparing model integrations.