# Asynchronous Delegation

2 related articles

Subscribe to updates

arXiv Oct 9, 2026
DuplexAgent coordinates full-duplex voice with delegated agents through six editable modules, and its recursive loop lifts fixed-core collaboration from 54.6 to 85.3
The work presents DuplexAgent, which writes the collaboration between a full-duplex interaction model and a delegation pool of reasoning LLMs and coding agents as six editable modules (Policy, Intake, Hold, Worker, Deliver, Memory), plus Duplex-Harness-RSI, a closed loop in which a simulator generates timed conversations and attributes failures, a reasoning LLM acts as Exam Planner to choose the next exam, and a coding agent acts as Harness Editor to revise the modules named by the diagnosis, with paired development and protection gates deciding retention; on VoiceChat and Qwen-Realtime the retained harness improves spoken knowledge, executable tool use, and fixed-core collaboration.
The work presents DuplexAgent, which writes the collaboration between a full-duplex interaction model and a delegation pool of reasoning LLMs and coding agents as six editable modules (Policy, Intake, Hold, Worker, Deliver, Memory), plus Duplex-Harness-RSI, a closed loop in which a simulator generates timed conversations and attributes failures, a reasoning LLM acts as Exam Planner to choose the next exam, and a coding agent acts as Harness Editor to revise the modules named by the diagnosis, with paired development and protection gates deciding retention; on VoiceChat and Qwen-Realtime the retained harness improves spoken knowledge, executable tool use, and fixed-core collaboration.
Yingda Shen（First author）
DuplexAgent Duplex-Harness-RSI Full-duplex voice agent collaboration Collaboration simulator Asynchronous Delegation
arXiv Sep 28, 2026
Realtime-Venus splits real-time dialogue from background work into a dual-loop runtime with two 9B frontends, leading online models on six of eight video benchmarks
The work presents Realtime-Venus, a proactive full-duplex interaction system with two separately trained 9B models, Realtime-Venus-Omni for audio-visual interaction and Realtime-Venus-Audio for spoken interaction, which align continuous perception, conversational control, native speech generation, and private delegation requests on a shared causal timeline while Realtime-Venus-Harness executes background capabilities asynchronously in a dual-loop runtime, achieving the highest scores among compared online models on six of eight video benchmarks and leading or matching on several audio understanding, spoken question answering, and full-duplex overlap metrics.
The work presents Realtime-Venus, a proactive full-duplex interaction system with two separately trained 9B models, Realtime-Venus-Omni for audio-visual interaction and Realtime-Venus-Audio for spoken interaction, which align continuous perception, conversational control, native speech generation, and private delegation requests on a shared causal timeline while Realtime-Venus-Harness executes background capabilities asynchronously in a dual-loop runtime, achieving the highest scores among compared online models on six of eight video benchmarks and leading or matching on several audio understanding, spoken question answering, and full-duplex overlap metrics.
Ant Group（First author, Corresponding author）
Realtime-Venus Realtime-Venus-Harness Full-Duplex Interaction Asynchronous Delegation Long-Video Memory

You’re all caught up

DuplexAgent coordinates full-duplex voice with delegated agents through six editable modules, and its recursive loop lifts fixed-core collaboration from 54.6 to 85.3

Realtime-Venus splits real-time dialogue from background work into a dual-loop runtime with two 9B frontends, leading online models on six of eight video benchmarks