DEV Community: CR4CODE

Termux Assistant AI v3.7 — что нового

CR4CODE — Fri, 16 Oct 2026 08:42:27 +0000

Что нового

v3.7 — Форум сообщества + мульти-движки

Termux Assistant AI v3.7 — это не просто обновление. Это шаг к экосистеме, где ИИ-ассистент на Android перестаёт быть «чатом» и становится полноценным инструментом для жизни, работы и творчества.

💬 Форум сообщества — теперь у проекта есть дом

Открыт форум: обсуждения, баг-репорты, идеи, голосование за функции, раздел «Сценарии и промпты». Вход по e-mail, тёмная тема, уведомления, вложения. Больше не нужно писать в личку — всё открыто и структурировано.
Форум: termux-assistant-ai.ru/forum
Категории: Новости, Установка, Ошибки, Идеи, Сценарии, Разработка
Регистрация за 30 секунд, вход через e-mail

🧠 Мульти-движки: DeepSeek, Claude, Perplexity, Gemini, Mistral

Раньше ассистент работал только с DeepSeek. Теперь — реестр движков: каждый сайт адаптируется своим JS-адаптером. Управляй несколькими ИИ одновременно: один отвечает в чате, другой ищет, третий пишет код.

🏠 Пример: умный дом без специалиста и без затрат

Купил одноплатник или мини-ПК для умного дома? Обычно нужен айтишник, чтобы всё настроить: Home Assistant, MQTT, автоматизации, скрипты. Теперь это делает ассистент — прямо с телефона, голосом или текстом. Бесплатно. Без облаков и подписок.

🎨 Плитка «Форум» в приложении

Хаб пополнился плиткой «Форум» — доступ к сообществу в один тап. Плюс кнопка «Скачать APK» и переход на сайт с форума.

🛡️ Усиленная безопасность

Подтверждение опасных команд (git push, установка пакетов) — всплывающее окно поверх любого приложения. Allowlist + audit-журнал всех действий ИИ.

Возможности — безграничны, если есть фантазия

Наш ассистент — это не «ещё один чат с ИИ». Это руки для нейросети внутри твоего Android. Идеи:
Автоматизация рутины: бэкапы, парсинг, отчёты
Домашний сервер и умный дом без специалиста
Вайб-кодинг: собрать APK, игру, скрипт — прямо с телефона
Личный DevOps: деплой, мониторинг, уведомления
Творчество: генерация и обработка контента, озвучка
Работа: прототипы, автоматизация процессов, интеграции
Единственный лимит — твоё воображение.

📦 Что внутри

Форум сообщества (полный стек: PostgreSQL + FastAPI + Jinja2)
Мульти-движки ИИ (Claude, Perplexity, Gemini, Mistral)
Плитка «Форум» в Hub
Подтверждение опасных команд + audit
Реальный APK для Android 7+

🔗 Ссылки

Скачать APK: termux-assistant-ai.ru
Форум: termux-assistant-ai.ru/forum
GitHub: github.com/CR4CODE/termux-assistant-ai-releases
Telegram: t.me/termux_assistant_ai
ВКонтакте: vk.com/termuxai
#Termux #AI #Android #DeepSeek #OpenSource #умныйдом #автоматизация #вайбкодинг

Ссылки

🌐 Сайт: https://termux-assistant-ai.ru
📦 GitHub: https://github.com/CR4CODE/Termux-Assistant-AI/releases
💬 Telegram: https://t.me/termux_assistant_ai

Termux Assistant AI — Android-приложение, которое умеет не только говорить, но и делать: собирает APK, постит в соцсети, чинит себя.

Termux Assistant AI v3.5.0 — AI with hands inside Android

CR4CODE — Sat, 10 Oct 2026 12:40:41 +0000

Most AI assistants can only talk. They write code, explain concepts, give advice — but they can't do anything in the real world. You copy their answer and paste it somewhere yourself.

Termux Assistant AI breaks this paradigm. It's an AI assistant that lives inside Android through Termux and has a real shell on your phone. Not a sandbox. Not an emulator. The actual Linux environment with real filesystem access.

What's new in 3.5.0

📤 DeepSeek chat export

A new screen in the app: paste a public share link, get the full conversation as Markdown. It uses Playwright with a stealth browser to bypass the AWS WAF protection, then extracts every message with proper roles.

The result is saved to /sdcard/ai-tasker/chats/ — visible in any file manager.

🧑🏫 Mentor mode

Step-by-step tutorials inside other apps. A scenario engine with element highlighting, powered by Accessibility API. Voice guidance included.

🔗 Smart social accounts

The "Social" section now shows connection status for VK and Telegram. If a token is missing, it offers to configure it. A "Check" button validates the token through the API — groups.getById for VK, getMe for Telegram.

🎨 Unified UI

60+ SVG icons from Bootstrap Icons replaced every emoji in the interface. Consistent style across all screens, colored toasts with context, animations and neon glow.

Critical fix: auto-cycle

Long bash blocks (heredoc with Cyrillic text) were getting truncated in the middle. The root cause: DeepSeek renders a long code block as multiple TextView nodes in the WebView, and the app was only grabbing the first one. Now all parts are collected and joined.

This bug affected any multi-line Python script submitted through the auto-cycle.

Auto-increment versionCode

The build script now reads the maximum versionCode from the installed APK and the Download folder, then bumps it by 1. The classic "app won't install" problem (downgrade refused) is solved permanently.

What it can do

Build APKs directly on the phone (aapt2 + javac + d8 + apksigner)
Install them via the app
Post to VK, Telegram, Dev.to
Work with git, GitHub, gh CLI
Read logs, restart services, patch itself

I gave a language model hands inside Android — and it built me an app factory

CR4CODE — Wed, 07 Oct 2026 06:34:44 +0000

Most LLM assistants are "brains in a jar." They can reason, write code, explain concepts — but they can't do anything in the real world. Everything they produce is text that a human copies and pastes somewhere.

I decided to fix that. Termux Assistant AI is an Android app that bridges a WebView with DeepSeek Chat and simultaneously gives the assistant direct shell access inside Termux via local HTTP servers. In effect — the language model gets hands.

The project is fully open source (MIT): github.com/CR4CODE/Termux-Assistant-AI.

What Termux is, and why it's the key

Termux is a terminal emulator for Android with a full Linux environment — no root required. Bash, Python, git, ssh, package manager — all working. Essentially a regular Linux distro in your pocket.

The key insight: if I have Termux, I have a shell. And if a language model has a shell, it can run arbitrary commands, work with files, compile code, publish releases.

Two questions remained:

How to bridge the WebView app with Termux?
How to safely pass commands from the model to the shell?

Architecture: three layers

The system consists of three isolated layers.

Layer 1: WebView + DeepSeek

The main app is an Android WebView that loads chat.deepseek.com. The user chats with the model as usual. But the app intercepts the "Send" button and can insert prompts into the input field and read responses from the DOM.

Implemented via a JavaScript bridge VibeWebBridge: methods sendMessage, readLastAnswer, lastAnswerHash, isThinking, isStopped execute JS in the WebView and return results through callbacks.

Response reading works by finding [class*=markdown] elements and hashing the last block to distinguish a new response from an old one.

Layer 2: HTTP server in Termux (port 8767)

Inside Termux runs a Python server ai-tasker-server. It listens on 127.0.0.1:8767 and accepts tasks from the app:
the app accepts tasks:

POST /task
Body: {"task": "auto:uptime"}
Response: {"status":"success","output":"...","exit_code":0}

Routes are grouped by prefix:

auto: — direct bash via bash -c
project_* — read/write/build Android projects
vk_, tg_ — posting to VK and Telegram
release_* — release pipeline: CHANGELOG, tag, GitHub Release
bot_* — PiarBot management
apply_patches: — apply patches to source files
log: — read logs

A key design decision: auto: commands go straight to bash, without LLM-based classification. The first version tried to guess the command type, which caused hangs on non-existent routes. Now it's just subprocess.run(["bash","-c", code], timeout=120).

Layer 3: VibeService (port 8768)

A separate Android service that manages the WebView and exposes an HTTP API:

POST /send — insert text into input field and press "Send"
GET /read — read last DeepSeek response
GET /hash — hash of last response (for change detection)
GET /thinking — is the model generating right now?
GET /stopped — is the model stuck in "Stopped" state?
POST /install-apk — install APK from path

This service is how the assistant talks to DeepSeek and runs autonomous loops.

Full development cycle: from prompt to APK

This is the most interesting part — Vibe mode. You write one task, and the system does the rest.

Workflow:

User runs vibe-run <project> "<task>"
Script collects project tree and content of all .java/.xml files
Builds a prompt requiring files back in format === FILE: path === ... === END FILE ===
Sends to DeepSeek via VibeService (/send)
Waits for response stabilization: hash unchanged for N seconds in a row
Parses response, applies files to project
Builds APK via aapt2 → javac → d8 → apksigner
If build fails — sends error log back to the model and retries (up to 5 iterations)
On success — APK is handed over for installation via /install-apk

The key trick is self-healing. If the model stops in "Stopped" state (e.g., token limit), the background watchdog vibe-auto-resume checks /stopped every 5 seconds and automatically sends "continue". Without this, the autonomous run just hangs.

Real use cases

Fast prototypes — idea → working APK in minutes
Phone automation — scheduled tasks, parsing, data dispatch
Content publishing — scheduled VK/Telegram posts with LLM-generated text
Release management — build, CHANGELOG, tag, GitHub Release, social announcement

Honest limitations

Phone resources. Heavy projects (Gradle, Kotlin, Compose) won't fit — the pipeline is tuned for lightweight Java+XML
Fragility of autonomous runs. One typo in a patch can crash the server, so all changes are backed up
Dependency on an external LLM. DeepSeek Chat is not local. A local option via llama.cpp in Termux is the next step
No vision. The assistant "sees" only the WebView DOM, not the whole screen

What's next

Voice control via Termux API (TTS/STT)
Local LLM via llama.cpp for full autonomy
Vision — OCR + uiautomator to control any app
APK marketplace — build → sign → publish to own channel

Conclusion

Termux Assistant AI is an experiment: what happens if you give a language model hands? It turns out — a lot. The model builds apps, ships releases, fixes its own bugs, runs social media. All inside a single Android phone, without a PC or cloud infrastructure.

The project is fully open source:

Code: github.com/CR4CODE/Termux-Assistant-AI
Landing: cr4code.github.io/Termux-Assistant-AI
License: MIT

I'm not a programmer. I don't know a single line of code. Everything you see was built by one person and one AI that became my hands.

Questions and criticism are welcome in the comments.