Angel: public-domain Android app runs tool-using LLM agents on-device with llama.cpp and MCP
ihaveaboyfriendsorry · reddit · 2026-10-09
A developer open-sourced Angel, a public-domain (Unlicense) Android app that runs tool-using LLM agents on your phone with no telemetry and data staying on-device.
Key capabilities:
- Providers: native APIs + tool calling for OpenAI, Gemini, Anthropic, and DeepSeek, or on-device GGUF models via vendored llama.cpp (CPU, ARM-optimized kernels); local models use their native tool-call format, so tool calling works offline
- Tools exposed: androidshell (uid 2000 via Shizuku, no root), readscreen (uiautomator UI tree), linuxshell (proot distro), plus full MCP support (featured servers, official registry, custom)
- Safety model: per-tool Ask/Allow/Deny, editable commands before execution, per-command kill, encrypted on-device credentials
- Tech: Kotlin/Compose, minSdk 26, ARM64; tests cover provider contracts, storage/crypto, Shizuku integration, and real on-device model download + generation
The author is seeking feedback on provider adapter edge cases, on-device tool-call reliability on small models, and failing MCP registry installs. Repo and APK are public.
Related event: Open-Source Android App Angel Runs LLM Agents and MCP Locally(2 posts)→
More from coding & agent
- Atlassian Launches Agentic Multiplayer Protocol (AMP) for Human-Agent Collaboration — davidhoang · 2026-10-09
- Flock: Open-Source Tool Runs Multiple Claude Accounts Side by Side on Desktop — AccomplishedCraft389 · 2026-10-09
- app-store-screenshots hits 7.2k GitHub stars: AI agent skill scaffolds store-ready screenshot editor — tom_doerr · 2026-10-09
- rhp 2.0 Ships MCP Server So AI Agents Generate Editorial-Style Charts in One Prompt — bezdazen · 2026-10-09
- LLM-as-a-Verifier: Weaker Model Verifies Stronger One, Hits 69.2% SOTA on Terminal-Bench 4 — Azaliamirh · 2026-10-09
- TestSprite Season 4 developer contest offers $4,000 prize pool for Claude Code and Codex users — JaynitMakwana · 2026-10-09