Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

AbuZar-Ansarii/Jarvis

Productivity

A local-first Android voice assistant that automates on-screen tasks through accessibility services and connects to local or cloud LLMs.

135
57
Dart
MIT
15 days old

Project overview

J.A.R.V.I.S. (PrivateAgent) is a Flutter/Dart Android app that combines a voice assistant, an on-device UI automation agent, and a sci-fi HUD. Its core mechanism is an accessibility service that dumps the current screen's view hierarchy, sends a compact representation to an LLM, and then injects taps, text input, scrolls, and swipes to carry out multi-step tasks. It can talk to an OpenAI-compatible endpoint, including a local Ollama instance, so the LLM call can stay on-device or go to a cloud provider you configure.

Beyond automation, the README documents a wide set of add-on features: wake-word listening, a floating overlay assistant, voice translation, a walkie-talkie-style Telegram remote control, sensor-based air gestures, motion-triggered sentry mode with camera capture, local network scanning, device diagnostics, and smart-home control via Home Assistant. These are best treated as separate capabilities you opt into, each requiring its own Android permission.

Build and run

Prerequisites are Flutter SDK 3.12.2 or newer, Android compile SDK 35, minimum SDK 26 (Android 8.0+), and JDK 17+.

git clone https://github.com/AbuZar-Ansarii/Jarvis.git
cd Jarvis
flutter pub get
flutter analyze
flutter test test/jarvis_features_test.dart
flutter build apk --release
adb install -r build/app/outputs/flutter-apk/app-release.apk

The release APK lands at build/app/outputs/flutter-apk/app-release.apk. Note the README's own clone snippet still contains a placeholder repository URL (clone section); use the project's actual repository address instead. A prebuilt release is also advertised on the Releases page.

After installing, configure an LLM provider in Settings. For offline use, point the base URL at a local Ollama endpoint such as http://127.0.0.1:11434/v1; cloud presets exist for DeepSeek, OpenAI, Gemini, OpenRouter, and NVIDIA NIM. Then grant the Accessibility service and other permissions through the in-app Permissions Hub before automation can work.

Permissions and constraints

The app depends heavily on special Android permissions: microphone for voice, Accessibility service for screen inspection and touch injection, system alert window for the floating HUD, notification listener for email/SMS triage, SMS and contacts access, and storage management for file operations. Battery-optimization exemption is recommended for wake-word listening while the screen is off. Shizuku is listed as an optional privileged path for elevated shell commands without root.

Accessibility automation cannot read apps that set FLAG_SECURE, such as many banking apps, and clicks can miss targets when the screen is not in the expected orientation. Local-first refers to on-device storage and the option of a local LLM; the default voice engine streams Microsoft Edge neural TTS audio over the network and falls back to Android TTS offline. Remote control through Telegram sends commands and screenshots through Telegram's infrastructure. The project is MIT-licensed.

HotRepo's take

The project sits in the system-tools niche rather than a general assistant: its value is in using Android accessibility APIs to operate other apps, which no web-based assistant can do. The README is unusually detailed for a solo project, with a directory map, architecture diagram, and explicit troubleshooting notes. But the documented surface is extremely broad, and the HotRepo record has no GitHub daily-ranking entry, so the descriptions should be read as author claims rather than verified behavior. The star curve over the observed day (roughly 107 to 113) shows a small, slow-moving audience; the 57 forks relative to that count hint at people trying to build or modify it themselves.

Written by AI from the project's README, updated . The project's own docs are the source of truth.

You might also like

  1. adewaskar/jarvisA browser-based voice assistant with an Iron Man interface that uses Claude Code to run your tools and MCP servers.TypeScript836
  2. isair/jarvisA local-first desktop AI assistant with always-on voice, personal memory, MCP tools and dictation for macOS, Windows and Linux.Python2,305
  3. TencentCloud/OctopA self-hosted multi-user, multi-agent assistant platform for households and small teams that want one private instance.Python10,919
  4. sigoden/aichatA terminal LLM client that unifies 20+ providers with shell command generation, chat REPL, RAG, tools and agents.Rust10,548
  5. CapriChooms/CapriCoreA Windows desktop AI companion with local or API models, voice, memory and permission-based desktop tools.JavaScript103