tencent/TCAndon-Router
TCAndon-Router is a 4 billion parameter reasoning-centric multi-intent routing module developed by Tencent, designed for agent selection in multi-agent systems. This model excels at dynamically onboarding new agents without retraining and provides transparent, interpretable routing decisions. It achieves state-of-the-art performance on real-world enterprise datasets like HWU64, MINDS14, SGD, and Tencent Cloud ITSM, making it ideal for complex intent-routing scenarios and resolving agent conflicts.
Loading preview...
TCAndon-Router: Adaptive Reasoning for Multi-Agent Systems
TCAndon-Router, developed by Tencent, is a 4 billion parameter reasoning-centric multi-intent routing module specifically designed for agent selection within multi-agent systems. It addresses the critical need for efficient and accurate agent routing in complex enterprise applications.
Key Capabilities & Features
- Dynamic Agent Onboarding: New agents or intents can be added simply by appending their descriptions, eliminating the need for retraining the model.
- Transparent Routing Decisions: Provides interpretable routing, enhancing explainability, robustness, and cross-domain generalization, which simplifies post-deployment analysis.
- Conflict Resolution: Effectively manages agent conflicts arising from overlapping responsibilities, preserving all relevant agents and merging their outputs via a Refining Agent for a single, high-quality final response.
- State-of-the-Art Performance: Achieves leading results on large-scale, real-world enterprise datasets including HWU64, MINDS14, SGD, and Tencent Cloud ITSM (QCloud), outperforming models like Claude-Sonnet-4.5 and DeepSeek-v3.1-terminus in several benchmarks.
- Training Methodology: Trained using Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (DAPO).
When to Use TCAndon-Router
- Multi-Agent Systems: Ideal for scenarios requiring intelligent routing of user queries to appropriate agents.
- Intent-Routing Applications: Applicable to any system needing to classify and route user intents, such as skill selection for agents.
- Enterprise Solutions: Designed for real-world enterprise applications where dynamic adaptability and interpretability are crucial.
- Complex Query Handling: Particularly effective when dealing with queries that might involve multiple agents or overlapping responsibilities, ensuring comprehensive responses.