Nastech

Nastech is an open-source Android AI assistant that automates your phone through plain‑language commands. It offers over 80 device tools, AI‑authored workflows, scheduled jobs, an in‑app browser driven by the AI, keyless web search, a Linux workspace, SSH, screen automation, file management, music controls, cloud voice transcription, cloud‑first model providers, and a remote Telegram bot. Every capability is opt‑in.

What can it do?

Tell it what to do in plain language, and the phone runs it in the background while you go about your day.

Examples: - "Every weekday at 9am, summarise my unread WhatsApp into one Telegram message." - "If my home server's disk fills up, ping me." - "Watch my notifications. If anything from my boss comes in, forward it to Telegram." - "Find the PDF on my phone that mentions 'invoice' and read me the first paragraph." - "Take a screenshot every 30 minutes for the next 4 hours so I can see what I actually did all afternoon." - "Use Termux to build me a webpage listing everything you can do, then open it in my browser." - "When I plug in headphones at home WiFi after 7pm, start my evening playlist."

Each is a one‑line setup.

Features

| Feature | Default State | Scope & Description | | :--- | :--- | :--- | | Device Control | Off (opt‑in) | Tap, swipe, scroll, type, take screenshots, open apps, adjust brightness/volume, post notifications, read battery/WiFi/signal/location/sensors, read contacts & SMS, send SMS, set wallpaper, read/write NFC tags, sign/encrypt with Android Keystore, access storage, manage ZIP archives. Over 80 tools. | | Shizuku | Off (opt‑in) | Run shell commands with higher privileges than a normal app – no root, no Termux required. Useful for granting permissions, inspecting system settings, and querying other packages. | | Workflows & Schedules | Off (opt‑in) | Set triggers (WiFi, Bluetooth, headphones, geofence, app launch, notifications, time, charging, screen state, etc.) and conditions (battery levels, sunrise/sunset, day of week, foreground app). Run tasks on any cadence (e.g., every Monday at 8am, every two hours). Survives reboots. | | Cloud AI and Voice | On (default) | Uses cloud‑first model providers: OpenAI‑compatible services, OpenRouter (PKCE), Ollama Cloud, OpenCode Zen. Optional ElevenLabs TTS and Scribe transcription. No local model downloads or GPU required. | | Telegram Bot | Off (opt‑in) | Control the assistant from anywhere via Telegram. Send questions, photos, PDFs, voice notes. Approval prompts use Yes/No buttons; multi‑choice questions appear in chat. Works through proxy where Telegram is blocked. | | In‑App Browser | On (built‑in) | Real browser that the AI can drive: click cookies, fill search boxes, scroll, read pages. Streams screenshots to chat. Floating chat pill lets you talk to the AI without leaving the page. Built‑in article extraction keeps token costs low. | | Web Search & Fetch | On (built‑in) | DuckDuckGo search works with no API key (default). 19 alternative engines available with your own key. Web fetch and extraction are enabled by default – pull any page, decode charset, paginate long docs. Blocked from private network targets. | | File Manager | Off (opt‑in) | Find, read, save, copy, move, rename, delete files. System folders outside the app sandbox remain off‑limits. | | Workspace | Off (opt‑in) | Real Linux environment on the phone. Run shell commands, read/write/patch files, browse results in a built‑in file manager with text editor and previews. Long‑running tasks survive turns and can be backgrounded. | | SSH | Off (opt‑in) | Save server credentials once. Run commands, upload/download files, tail logs, start long‑running servers – all from chat. Works on WiFi or cellular. | | Music & Media | Off (opt‑in) | Control Android media playback (play, pause, volume) from chat or Telegram. Queue survives force‑stops. | | Skills | On (bundled) | Drop a Markdown skill file and the AI gains new capabilities. Bundled catalog includes QR generator, Wikipedia query, piano, interactive map, etc. Two skills enabled by default. | | Sub‑Agents | On (optional) | Dispatch focused sub‑agents to clean side‑contexts, optionally on cheaper models. Run multiple in parallel. Manage profiles in Settings. | | Context Compaction | On (auto) | Summarises older conversation turns automatically when the context window fills, keeping recent tool calls intact. Manual trigger also available. | | Doctor | On (built‑in) | Built‑in health checkup for permissions, services, databases, network, Termux, Shizuku, MCP servers, skills, workspaces, Telegram proxy, and more. Auto‑fix available. | | MCP Servers | Off (opt‑in) | Connect Model Context Protocol servers – the AI gains whatever tools they expose. All changes require approval. | | Notifications & External Triggers | Off (opt‑in) | Read, summarise, and forward incoming notifications from apps you choose. Whitelist starts empty. External apps (Tasker, ADB) can hand tasks via the External Automation Intent API. |

Quick Start

1. Install - Download the latest APK. - Allow install from unknown sources, then open. - If you have an old debug build installed, uninstall it first (different signing). - Existing installation with a different application ID? Export a backup in Settings before installing, then restore after setup.

2. Add an LLM Provider - Settings → Providers → pick one → paste your API key. - Options: OpenRouter, Codex (OAuth), Grok (OAuth), Local (LiteRT on‑device), AICore (Gemini Nano on Pixel 8/9/10+).

3. Turn on what you want - Settings → Assistants → tap your assistant → Local Tools – enable the categories you need. - Without tools enabled, Nastech remains a focused chat experience.

4. (Optional) Telegram Bot - Message @BotFather for a token. - Message @userinfobot for your numeric user ID. - Tell the assistant: "Set up the Telegram bot. Token is <token>. My user id is <id>. Set me as the default chat. Enable it."

Requirements

- Architecture: arm64 or x86_64 - Android 8.0+ (API 26), targets API 37 - Storage: ~80 MB - LLM Provider: OpenAI, Google, Anthropic, OpenRouter, Codex, Grok, Ollama, or any OpenAI‑compatible endpoint. Google sign‑in available as an alternative to a Gemini API key. AICore on Pixel devices also supported.

Languages

The interface ships in English, 简体中文, 繁體中文, 日本語, 한국어, Русский, and العربية. The app follows your system language and falls back to English. RTL languages (Arabic, Persian, Urdu) render correctly in chat. Screen readers (TalkBack) are fully supported.

Safety & Privacy

Three layers of protection: - Per‑assistant toggles – every tool starts off; you enable only what you want. - Per‑call approval – tools that change something ask before running. - HARDLINE floor – genuinely dangerous commands (wipe, reboot, fork bombs, system file destruction) are blocked unconditionally.

Additional safeguards: - Passwords and API keys never hit log files. - Cloud backups skip saved credentials. - The Telegram bot ignores everyone except your allowlist. - Web fetches are blocked if they resolve to a private network address (LAN or cloud metadata endpoints).

License

GNU Affero General Public License v3.0