TaoLive Digital Avatar Agent Technical Report: Training Agents to Evolve with Their Harness

2026-08-16

AI Summary

This study addresses the challenge of balancing low latency with dynamic strategy adaptation for digital humans in live-streaming e-commerce by proposing Harness, a perception-aware training framework. This paradigm innovatively incorporates runtime configurations into the training distribution, enabling model behavior decoupling and retraining-free dynamic evolution through state augmentation, a three-stage training pipeline comprising SFT, distillation, and proxy RL, and a production-grade simulator. Experimental results demonstrate that the 35B model achieves a score of 94.8 in real-world question answering with a P50 latency of only 3.4 seconds. Furthermore, it exhibits significantly superior instruction-following capabilities compared to baselines and general-purpose large language models, effectively resolving the trade-off between overfitting in compact models and high latency in larger counterparts.

Abstract

AI-powered digital-avatar streamers in live e-commerce must answer product questions, engage viewers, and execute changing business strategies in real time. This requires low latency, factual and effective replies, and rapid adaptation to updated campaign, compliance, and style requirements. We develop an evolvable Harness that decouples Skills, Hooks, system prompts, and tools from model weights, allowing runtime behavior to change without retraining. However, Harness evolution creates a moving execution environment: compact models fine-tuned on one configuration may memorize names, schemas, and prompt templates rather than follow the Harness currently provided, while stronger zero-shot models are too slow for real-time use. We address this tension with Harness-Aware Training (HAT), which makes Harness states part of the training distribution. HAT applies task-preserving Harness-State Augmentation (HSA) to Skills, tool schemas, prompt structures, and interaction constraints, and comprises three stages: HSA-based supervised fine-tuning, general on-policy distillation to recover general capabilities, and HSA-based agentic reinforcement learning in a production-informed live-room simulator. Across four evaluation sets with more than 4,500 cases, our compact 35B model scores 94.8 on real-world Live-Stream QA, versus 80.3 for the base model and 93.0 for the strongest evaluated general LLM, while scoring 94.6 on Harness-Variant QA and retaining 83.5 on IFEval. By contrast, fixed-Harness SFT reduces IFEval by 7.7 points. In a controlled complete-agent replay on one NVIDIA H20 GPU with MTP enabled, the system achieves 3.407 s P50 and 8.114 s P95 latency. These results show that HAT produces a latency-feasible compact agent that remains effective under evaluated Harness changes without sacrificing general instruction following.

Problem

• Digital Avatar Agent
• Live E-commerce
• Harness Evolution
• Real-time Adaptation
• Low Latency

Innovation

• Harness-Aware Training
• Harness-State Augmentation
• Agentic Reinforcement Learning
• Digital Avatar Agent
• On-policy Distillation