Macaron-A2UI: A Model for Generative UI in Personal Agents
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916042692886528 |
|---|---|
| author | Kong, Fancy Zheng, Congjie Zhuang, Murphy Yang, Rio Zhang, Sueky Fu, Hao Jin, Gene Cao, Song Chen, Kaijie Chen, Andrew Ma, Pony |
| author_facet | Kong, Fancy Zheng, Congjie Zhuang, Murphy Yang, Rio Zhang, Sueky Fu, Hao Jin, Gene Cao, Song Chen, Kaijie Chen, Andrew Ma, Pony |
| contents | As personal agents evolve to handle complex, user-centric tasks, static plain-text chat is rapidly becoming a bottleneck. Generative UI emerges as the necessary new interface layer, dynamically synthesizing the right controls, options, and state from the interaction context in real time. We present Macaron-A2UI, a model for Generative UI in personal agents. Our goal is to move beyond text-only interaction by enabling agents to generate natural language together with lightweight, executable UI actions for information collection, preference refinement, confirmation, and multi-goal organization. We build a large-scale Generative UI corpus from heterogeneous dialogue sources, introduce A2UI-Bench for controlled evaluation, and train 30B, 235B and 754B models with parameter-efficient LoRA-based supervised fine-tuning followed by reward-driven reinforcement learning. The best Macaron-A2UI model reaches 75.6 overall on A2UI-Bench without explicit schema hints, surpassing the strongest full-schema frontier baseline. We release the models, benchmark, and evaluation protocol to support future work on Generative UI for personal agents. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_24830 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Macaron-A2UI: A Model for Generative UI in Personal Agents Kong, Fancy Zheng, Congjie Zhuang, Murphy Yang, Rio Zhang, Sueky Fu, Hao Jin, Gene Cao, Song Chen, Kaijie Chen, Andrew Ma, Pony Human-Computer Interaction As personal agents evolve to handle complex, user-centric tasks, static plain-text chat is rapidly becoming a bottleneck. Generative UI emerges as the necessary new interface layer, dynamically synthesizing the right controls, options, and state from the interaction context in real time. We present Macaron-A2UI, a model for Generative UI in personal agents. Our goal is to move beyond text-only interaction by enabling agents to generate natural language together with lightweight, executable UI actions for information collection, preference refinement, confirmation, and multi-goal organization. We build a large-scale Generative UI corpus from heterogeneous dialogue sources, introduce A2UI-Bench for controlled evaluation, and train 30B, 235B and 754B models with parameter-efficient LoRA-based supervised fine-tuning followed by reward-driven reinforcement learning. The best Macaron-A2UI model reaches 75.6 overall on A2UI-Bench without explicit schema hints, surpassing the strongest full-schema frontier baseline. We release the models, benchmark, and evaluation protocol to support future work on Generative UI for personal agents. |
| title | Macaron-A2UI: A Model for Generative UI in Personal Agents |
| topic | Human-Computer Interaction |
| url | https://arxiv.org/abs/2605.24830 |