I Can't Write Code. But I Shipped a Native Android Terminal IDE with Claude Code + Codex Running Natively — No Termux Required.

RYO ITABASHI — Tue, 28 Apr 2026 14:07:07 +0000

Update: v0.1.0 Released

Since my last post, everything changed.

The WebView terminal is gone. Termux is no longer required.
Claude Code and Codex now run natively on Android —
in the same process as your shell, with zero TCP, zero IPC.

This is what shipped.

The Problem I Kept Running Into

Claude Code broke in Termux. Codex broke in Termux.
Every update was a gamble.

So I directed the AI to throw away the entire architecture
and rebuild it from scratch.

What Changed

Before (Plan A):

WebView terminal via ttyd
Termux required
TCP socket between terminal and app
Died every time I switched apps

After (Plan B):

Native PTY via JNI forkpty — same process, zero IPC, zero TCP
Termux not required
bash, Node.js, Python 3, git, curl, ripgrep, jq, tmux, vim, sqlite3 bundled inside the APK as .so files
Executed via /system/bin/linker64 to bypass SELinux's execve restrictions on app_data_file context

As far as I know, this is the only React Native app in the
world with an embedded native terminal emulator running
in-process via JNI.

Claude Code + Codex on Android

Both now run natively through Shelly's managed runtime:

Claude Code 2.1.121 (Opus 4.7, 1M context) ✅
Codex v0.124.0-termux (GPT-5.5) ✅
Gemini CLI 0.39.1 ✅

The runtime hot-swaps updates without an APK rebuild.
Broken versions are logged and cooled down automatically.

If Claude Code or Codex broke in your Termux setup —
this is the maintained path.

4 live panes simultaneously:

Top left: Claude Code terminal
Top right: Codex (GPT-5.5)
Bottom left: Cerebras AI pane with READING TERMINAL
Bottom right: Browser pane with YouTube

What Works Today

Native terminal with inline command blocks
Multi-pane layout (up to 4 live panes)
Cross-pane intelligence — AI reads terminal output automatically
Multi-agent routing: Claude, Gemini, Cerebras, Groq, Perplexity, Codex, Local LLM via llama.cpp
File editing with per-hunk InlineDiff accept/reject
@team multi-model consensus
SSH profiles
CRT mode (scanlines + phosphor green + vignette)

Known Limitations

Claude Code first-run OAuth requires credential transplant
Ports monitor blocked by Android 10+ SELinux (bug #99)
Test coverage is thin

Try It

GitHub: https://github.com/RYOITABASHI/Shelly

APK: GitHub Releases

License: GPLv3

Rough edges exist. That's why it's open source.

I Can't Write Code. But I Built a 100,000-Line Terminal IDE on My Phone.

RYO ITABASHI — Sat, 04 Apr 2026 16:35:18 +0000

I can't write code.

I'm not an engineer. I've never written a line of TypeScript. I have no formal training in computer science.

But I built a 100,000-line terminal IDE — by talking to AI.

Every architectural decision is mine. The code is not. It was created through conversation with Claude Code, running inside Termux on a Samsung Galaxy Z Fold6. No desktop. No laptop. Just a foldable phone and an AI that can execute commands.

Today I'm releasing it as open source.

GitHub: github.com/RYOITABASHI/Shelly

The Problem

You're running Claude Code in the terminal. It throws an error. You copy it. You switch to ChatGPT. You paste. You ask "what went wrong?" You copy the fix. You switch back. You paste. You run it.

Seven steps. Every single time.

The terminal and the chat live in different worlds. You are the copy-paste bridge between them.

The Solution

Shelly puts Chat and Terminal side by side — and connects them with AI.

Say "fix the error on the right". Shelly reads the terminal output, understands the error, and generates an executable command. Tap ▶ Run and the fix lands directly in the Terminal.

No copy. No paste. No tab switching.

What makes it different

Native terminal emulator. Not a WebView. Kotlin code derived from Termux's terminal-emulator library, rendering on an Android Canvas. Connected via a C helper (forkpty()) over TCP localhost. As far as I know, this is the only React Native app with an embedded native terminal emulator.

12 AI agents. @claude, @gemini, @codex, @cerebras, @local, @perplexity, @team, @plan, @arena, @actions. The input router automatically selects the best AI for the task, or you choose with @mentions.

It runs on your phone. Not in the cloud. Not on a desktop. Shelly + Termux gives you a full development environment in your pocket.

Key Features

Cross-pane intelligence — AI reads terminal output, suggests fixes, one-tap execution
ActionBlock — Code blocks in AI responses have [▶ Run] buttons
Approval Proxy — Terminal [Y/n] prompts become native chat buttons
Auto-savepoint — Game-like save system. Every change auto-committed. Revert with one tap.
Plan Mode — AI generates step cards. Execute or skip each step.
Arena Mode — Same prompt, two AIs, blind comparison. Vote for the better one.
Click-to-Edit — Tap any element in the preview panel, tell the AI what to change
VoiceChain — Speak → execute → TTS response. Hands-free development.
GitHub integration — AI suggests when to push. @actions sets up CI/CD without YAML.
5-level command safety — Every command is risk-assessed before execution
Local LLM support — Run models on-device via llama.cpp. Fully offline capable.

The keyboard too

The keyboard you see in the screenshots is Nacre — an 11,000-line Android IME I built (also through AI) to solve the input problem on mobile.

github.com/RYOITABASHI/Nacre

Built with

Layer	Technology
Framework	Expo 54 / React Native 0.81
Language	TypeScript + Kotlin + C
Terminal	Native emulator (Termux-derived) + Direct PTY (C, forkpty)
State	Zustand
i18n	900+ keys, English/Japanese

Why I built this

I'm a creative director. I wanted to use Claude Code on my phone, but Termux was too intimidating. So I made a chat interface that hides the complexity while keeping the full power.

Then I realized the real problem wasn't the terminal — it was the gap between the terminal and the AI. So I connected them.

Then the WebView kept dying every time I switched apps. So I directed the AI to replace the entire rendering layer with a native terminal emulator written in Kotlin and C.

100,000 lines later, I still can't write code. But I can describe what I need, and the AI builds it.

Try it