Index / DroidClaw · updated Sep 4, 2026
DroidClaw
unitedbyai/droidclaw · healthy · rank 23 of 43 by stars
DroidClaw turns old Android phones into autonomous AI agents that read the screen via accessibility trees and tap/type through ADB. Its standout trick is delegating tasks to on-device apps like ChatGPT or Gemini without needing API keys.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 45 / 100 · how these are scored
higher is safer
higher is safer
higher is safer
higher is safer
higher is riskier
Model access
Read from the repository, not written by a model · 29 files examined
3 providers · runs locally · gateway support
Pinned models
llama-3.3-70b-versatile released 2024-12-06 — 21 months old (from the public model catalogue)
Pin last edited 7 months ago
Several defaults are pinned; the project picks one per provider rather than shipping a single default.
Evidence
Decision
Why choose DroidClaw over OpenClaw?
Why choose this
- Controls real Android devices via accessibility tree and ADB
- Uses installed apps like a human, avoiding API key requirements
- Web dashboard with scheduled goals and remote device pairing
Tradeoffs
- Narrower scope: Android UI automation only, not a general assistant
- Early-stage (v0.5.x) with active bug fixes in core session handling
- No skill/plugin registry comparable to OpenClaw's ecosystem
Best fit
- Repurposing old Android phones as AI agents
- Automating mobile apps without APIs or integrations
- Delegating tasks to on-device apps like ChatGPT or Gemini
Avoid if
- You need desktop or server-side automation
- You require strong sandboxing or enterprise security controls
- You want a mature, stable release (still pre-1.0)
Evidence is solid for architecture and activity from the README and recent commits, but community sentiment is thin with zero Reddit matches and only indirect web results. Security and privacy assessments are inferred rather than documented.
AI layer reviewed Aug 24, 2026 · how this is written
Star activity
1,558 stars today
Overview
DroidClaw is a TypeScript-based AI agent that controls Android phones through a perception → reasoning → action loop. It dumps the accessibility tree via ADB, parses interactive UI elements, diffs screens to detect changes, and lets an LLM decide the next tap, type, or swipe — repeating until the goal is complete or the step budget runs out. The project ships an Android APK, a web dashboard, and a Bun-run kernel, with recent work adding scheduled goals via QStash and a shared session module for web-started goals.
Its most distinctive capability is API-free delegation: the agent can open apps like ChatGPT, Gemini, or Google Search on the device itself, submit requests, and bring results back — using apps exactly as a human would. This makes it a compelling way to repurpose old Android hardware into useful agents without building integrations.
Compared to OpenClaw, DroidClaw is narrower and more specialized: it focuses on mobile UI automation rather than general local assistance, and it lacks a skill registry or plugin ecosystem. It is also early-stage (v0.5.3) with active fixes around session management and scheduling, so expect rough edges. Security posture is a concern — the agent has broad control over the device, and no sandboxing model is documented.