ModalityDance/PalmClaw

[EMNLP 2026 Demo] "PalmClaw: A Native On-Device Agent Framework for Mobile Phones"

What it solves

PalmClaw provides a way to run a full AI agent stack directly on an Android device. It eliminates the need for a PC to deploy and operate an agent, offering a more private, local-first runtime that leverages mobile hardware and files.

How it works

PalmClaw operates as a native Android application. It uses an agent loop where an LLM decides on actions, calls tools, and generates responses based on context provided by memory and skills. It can receive inputs from a local chat interface or through integrated external channels (such as Telegram, Discord, Slack, Feishu, WeCom, and Email).

Who it’s for

It is designed for users who want a private AI assistant on their phone, as well as researchers and developers building native on-device agent frameworks for mobile phones.

Highlights

  • Mobile-native deployment: Runs directly on Android with access to local hardware and files.
  • Multi-channel integration: Supports connecting the agent to platforms like Telegram, Discord, Slack, and Email.
  • Full agent stack: Includes built-in support for memory, skills, and tools within a single mobile runtime.
  • Android sandbox isolation: Utilizes the Android App sandbox for a safer runtime boundary.

Related

  • Project
  • Project
  • Project
  • Project
  • Project