Qwen Audio Agent is a real-time voice runtime that keeps AI agents talking, working, and present. Normal voice assistants go silent or stall while they think, search, or run a tool; this runtime keeps the conversation flowing while the agent works in the background, then naturally reports back: “It’s ready.”
The project comes from the Qwen Audio team and is developed on GitHub, where it has earned nearly 3,000 stars under the Apache-2.0 license. It is moving fast: version 2.0.0 landed in September 2026 with a rebuilt orchestration runtime, a unified client protocol, and ACP/A2A backend integration, and updates continue through October 2026. The runtime pairs speech recognition and text-to-speech frontends with working agents: it advertises support for connecting coding and general agents such as Claude Code, Codex, and OpenCode, so a spoken request can kick off real work while the voice presence keeps acknowledging, updating, and conversing. Recent releases added a desktop conversation panel, remote mobile access, embeddable gateway and realtime provider extensions, agent skill installation and management, multimodal input in the terminal UI, and expanded voice and video model support. Memory and knowledge libraries help the agent stay consistent across a session. The team ships scenario examples including customer service, smart cockpit, and digital human deployments, plus a user guide and an arXiv technical report for those who want the architecture in depth.
Qwen Audio Agent is free and open source under the Apache-2.0 license. You supply your own models and infrastructure; any hosted model APIs you connect are billed by their providers.
It is for voice AI developers and agent builders who find turn-based, wait-in-silence interaction unacceptable, and for teams prototyping voice-first customer service, in-car assistants, and digital human experiences on an open runtime.