# An AI app for Android: cloud and local models, SSH, MCP, and screen control

> Source: <https://discuss.huggingface.co/t/an-ai-app-for-android-cloud-and-local-models-ssh-mcp-and-screen-control/192000#post_1>
> Published: 2026-11-17 23:11:33+00:00

**Key takeaway:** ForgeMind 0.2.1-alpha ships five native LLM wire protocols, twelve cloud presets, LAN scanning for Ollama/vLLM/LM Studio, and a bidirectional MCP setup where the app acts as client (Streamable HTTP, HTTP+SSE, SSH_STDIO) and host on 127.0.0.1:8765.

```
I have been working on an AI app for Android — a native Kotlin workstation rather than a thin chat client around a single API key. The first public builds are ForgeMind 0.2.1-alpha. The public repository is GPL-3.0:

https://github.com/DIIIUA/ForgeMind

I wanted the phone to sit on the same contour as a laptop: cloud APIs when I need them, a LAN or Tailscale model when I do not, and enough local tools that an agent can do work without a second device in my hand.

The app talks to five native LLM wire protocols: OpenAI-compatible `/chat/completions` (SSE, with a non-SSE fallback), OpenAI `/responses`, Anthropic `/v1/messages`, Gemini `streamGenerateContent`, and Ollama's native `/api/chat`. There are twelve cloud presets (OpenAI, Anthropic, Gemini, OpenRouter, DeepSeek, xAI Grok, Groq, Mistral, Venice AI, NVIDIA NIM, Cloudflare Workers AI, Hugging Face), plus arbitrary custom HTTP gateways, plus a LAN scanner for servers such as Ollama, LM Studio, llama.cpp, KoboldCpp, vLLM, SGLang, LocalAI, Jan, GPT4All, OpenWebUI, and LiteLLM.

It can open SSH and SFTP (password or key) over a public IP, a VPN, or Tailscale. For Zero Trust I implemented Cloudflare Access SSH over `wss://`: the app sends `CF-Access-Client-Id` and `CF-Access-Client-Secret` and does not need a `cloudflared` binary on Android, and the server does not need an open SSH port.

MCP is bidirectional. As a client, ForgeMind supports Streamable HTTP, HTTP+SSE, and `SSH_STDIO` (it can spawn `npx`, `uvx`, or `python3` MCP servers on a remote host or in Termux). As a host, a built-in Streamable HTTP server listens on `127.0.0.1:8765` so a local agent, Termux, `adb forward`, or a tunnel can call the phone's tools.

On the device there are four file environments in a dual-pane manager, a Sora editor with pre-edit rollback snapshots, a persistent `/system/bin/sh`, optional Termux, and an 8-window agent browser isolated from the ordinary in-app browser.

Screen control is optional and off until you enable the Accessibility service. Once enabled, the app can read visible text and controls of other apps as a tagged tree and tap, type, or scroll by tag. Screenshots are on demand (Android 11+), not a background recording. Android still hides password fields, and `FLAG_SECURE` still blocks screenshots.

The current public build is a GitHub prerelease APK, `ForgeMind-v0.2.1-alpha.apk`, for Android 10+ (` minSdk 29`). There is no store listing yet. This is alpha software: watch early tool calls, especially writes, shells, and Accessibility actions. Secrets stay on the device under AES-256-GCM and are not in the repository.

Releases: https://github.com/DIIIUA/ForgeMind/releases/latest

I would be glad to answer questions about the protocol clients, the Cloudflare Access SSH path, or the Accessibility Window-Tag tree. If you try the app, I am more interested in a failed setup, a wrong tool call, or odd behaviour on your OEM skin than in a compliment.
```
