ClawMobile: Rethinking Smartphone-Native Agentic Systems

Hongchao Du; Shangyu Wu; Qiao Li; Riwei Pan; Jinheng Li; Youcheng Sun; Chun Jason Xue

doi:10.1145/3805621.3807655

ClawMobile: Rethinking Smartphone-Native Agentic Systems

Hongchao Du (Co-first Author)
, Shangyu Wu (Co-first Author)
, Qiao Li
, Riwei Pan
, Jinheng Li
, Youcheng Sun
, Chun Jason Xue

Department of Computer Science

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

1 Downloads (CityUHK Scholars)

Abstract

Smartphones represent a uniquely challenging environment for agentic systems. Unlike cloud or desktop settings, mobile devices combine constrained execution contexts, fragmented control interfaces, and rapidly changing application states. As large language models (LLMs) evolve from conversational assistants to action-oriented agents, achieving reliable smartphone-native autonomy requires rethinking how reasoning and control are composed.We introduce ClawMobile as a concrete exploration of this design space. ClawMobile adopts a hierarchical architecture that separates high-level language reasoning from structured, deterministic control pathways, improving execution stability and reproducibility on real devices. Using ClawMobile as a case study, we distill the design principles for mobile LLM runtimes and identify key challenges in efficiency, adaptability, and stability. We argue that building robust smartphone-native agentic systems demands principled coordination between probabilistic planning and deterministic system interfaces. The implementation is open-sourced 1 to facilitate future exploration. © 2026 the owner/author(s).

Original language	English
Title of host publication	EuroMLSys '26 - Proceedings of the 2026 the 6th Workshop on Machine Learning and Systems
Place of Publication	New York, NY
Publisher	Association for Computing Machinery
Pages	370-376
Number of pages	7
ISBN (Print)	9798400726057
DOIs	https://doi.org/10.1145/3805621.3807655
Publication status	Published - Apr 2026
Event	6th Workshop on Machine Learning and Systems (EuroMLSys 2026) - Edinburgh, United Kingdom Duration: 27 Apr 2026 → … https://euromlsys.eu/

Publication series

Name	EuroMLSys - Proceedings of the European Workshop on Machine Learning and Systems

Conference

Conference	6th Workshop on Machine Learning and Systems (EuroMLSys 2026)
Abbreviated title	EuroMLSys '26
Place	United Kingdom
City	Edinburgh
Period	27/04/26 → …
Internet address	https://euromlsys.eu/

Research Keywords

Agent Systems
LLMs
Mobile Devices

Publisher's Copyright Statement

This full text is made available under CC-BY 4.0. https://creativecommons.org/licenses/by/4.0/

Access Output

10.1145/3805621.3807655

491109109.pdfFinal Published version, 491 KBLicence: CC BY

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

Du, H., Wu, S., Li, Q., Pan, R., Li, J., Sun, Y., & Xue, C. J. (2026). ClawMobile: Rethinking Smartphone-Native Agentic Systems. In EuroMLSys '26 - Proceedings of the 2026 the 6th Workshop on Machine Learning and Systems (pp. 370-376). (EuroMLSys - Proceedings of the European Workshop on Machine Learning and Systems). Association for Computing Machinery. https://doi.org/10.1145/3805621.3807655

@inproceedings{9b660e62fef344eeb9c173262728608e,

title = "ClawMobile: Rethinking Smartphone-Native Agentic Systems",

abstract = "Smartphones represent a uniquely challenging environment for agentic systems. Unlike cloud or desktop settings, mobile devices combine constrained execution contexts, fragmented control interfaces, and rapidly changing application states. As large language models (LLMs) evolve from conversational assistants to action-oriented agents, achieving reliable smartphone-native autonomy requires rethinking how reasoning and control are composed.We introduce ClawMobile as a concrete exploration of this design space. ClawMobile adopts a hierarchical architecture that separates high-level language reasoning from structured, deterministic control pathways, improving execution stability and reproducibility on real devices. Using ClawMobile as a case study, we distill the design principles for mobile LLM runtimes and identify key challenges in efficiency, adaptability, and stability. We argue that building robust smartphone-native agentic systems demands principled coordination between probabilistic planning and deterministic system interfaces. The implementation is open-sourced 1 to facilitate future exploration. {\textcopyright} 2026 the owner/author(s).",

keywords = "Agent Systems, LLMs, Mobile Devices",

author = "Hongchao Du and Shangyu Wu and Qiao Li and Riwei Pan and Jinheng Li and Youcheng Sun and Xue, \{Chun Jason\}",

year = "2026",

month = apr,

doi = "10.1145/3805621.3807655",

language = "English",

isbn = "9798400726057",

series = "EuroMLSys - Proceedings of the European Workshop on Machine Learning and Systems",

publisher = "Association for Computing Machinery",

pages = "370--376",

booktitle = "EuroMLSys '26 - Proceedings of the 2026 the 6th Workshop on Machine Learning and Systems",

address = "United States",

note = "6th Workshop on Machine Learning and Systems (EuroMLSys 2026), EuroMLSys '26 ; Conference date: 27-04-2026",

url = "https://euromlsys.eu/",

}

Du, H , Wu, S , Li, Q , Pan, R , Li, J, Sun, Y & Xue, CJ 2026, ClawMobile: Rethinking Smartphone-Native Agentic Systems. in EuroMLSys '26 - Proceedings of the 2026 the 6th Workshop on Machine Learning and Systems. EuroMLSys - Proceedings of the European Workshop on Machine Learning and Systems, Association for Computing Machinery, New York, NY, pp. 370-376, 6th Workshop on Machine Learning and Systems (EuroMLSys 2026), Edinburgh, United Kingdom, 27/04/26. https://doi.org/10.1145/3805621.3807655

ClawMobile: Rethinking Smartphone-Native Agentic Systems. / Du, Hongchao (Co-first Author); Wu, Shangyu (Co-first Author); Li, Qiao et al.
EuroMLSys '26 - Proceedings of the 2026 the 6th Workshop on Machine Learning and Systems. New York, NY: Association for Computing Machinery, 2026. p. 370-376 (EuroMLSys - Proceedings of the European Workshop on Machine Learning and Systems).

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

TY - GEN

T1 - ClawMobile

T2 - 6th Workshop on Machine Learning and Systems (EuroMLSys 2026)

AU - Du, Hongchao

AU - Wu, Shangyu

AU - Li, Qiao

AU - Pan, Riwei

AU - Li, Jinheng

AU - Sun, Youcheng

AU - Xue, Chun Jason

PY - 2026/4

Y1 - 2026/4

N2 - Smartphones represent a uniquely challenging environment for agentic systems. Unlike cloud or desktop settings, mobile devices combine constrained execution contexts, fragmented control interfaces, and rapidly changing application states. As large language models (LLMs) evolve from conversational assistants to action-oriented agents, achieving reliable smartphone-native autonomy requires rethinking how reasoning and control are composed.We introduce ClawMobile as a concrete exploration of this design space. ClawMobile adopts a hierarchical architecture that separates high-level language reasoning from structured, deterministic control pathways, improving execution stability and reproducibility on real devices. Using ClawMobile as a case study, we distill the design principles for mobile LLM runtimes and identify key challenges in efficiency, adaptability, and stability. We argue that building robust smartphone-native agentic systems demands principled coordination between probabilistic planning and deterministic system interfaces. The implementation is open-sourced 1 to facilitate future exploration. © 2026 the owner/author(s).

AB - Smartphones represent a uniquely challenging environment for agentic systems. Unlike cloud or desktop settings, mobile devices combine constrained execution contexts, fragmented control interfaces, and rapidly changing application states. As large language models (LLMs) evolve from conversational assistants to action-oriented agents, achieving reliable smartphone-native autonomy requires rethinking how reasoning and control are composed.We introduce ClawMobile as a concrete exploration of this design space. ClawMobile adopts a hierarchical architecture that separates high-level language reasoning from structured, deterministic control pathways, improving execution stability and reproducibility on real devices. Using ClawMobile as a case study, we distill the design principles for mobile LLM runtimes and identify key challenges in efficiency, adaptability, and stability. We argue that building robust smartphone-native agentic systems demands principled coordination between probabilistic planning and deterministic system interfaces. The implementation is open-sourced 1 to facilitate future exploration. © 2026 the owner/author(s).

KW - Agent Systems

KW - LLMs

KW - Mobile Devices

UR - http://www.scopus.com/inward/record.url?scp=105038658979&partnerID=8YFLogxK

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-105038658979&origin=recordpage

U2 - 10.1145/3805621.3807655

DO - 10.1145/3805621.3807655

M3 - RGC 32 - Refereed conference paper (with host publication)

SN - 9798400726057

T3 - EuroMLSys - Proceedings of the European Workshop on Machine Learning and Systems

SP - 370

EP - 376

BT - EuroMLSys '26 - Proceedings of the 2026 the 6th Workshop on Machine Learning and Systems

PB - Association for Computing Machinery

CY - New York, NY

Y2 - 27 April 2026

ER -

Du H , Wu S , Li Q , Pan R , Li J, Sun Y et al. ClawMobile: Rethinking Smartphone-Native Agentic Systems. In EuroMLSys '26 - Proceedings of the 2026 the 6th Workshop on Machine Learning and Systems. New York, NY: Association for Computing Machinery. 2026. p. 370-376. (EuroMLSys - Proceedings of the European Workshop on Machine Learning and Systems). Epub 2026 Apr 28. doi: 10.1145/3805621.3807655

ClawMobile: Rethinking Smartphone-Native Agentic Systems

Abstract

Publication series

Conference

Research Keywords

Publisher's Copyright Statement

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Cite this