README.md

metadata

title: Jarvis Mobile Agent
emoji: 🤖
colorFrom: blue
colorTo: purple
sdk: docker
pinned: false

🤖 Jarvis — AI Automation Planner for Mobile Devices

Jarvis is an intelligent AI automation planner that converts user voice/text commands into structured JSON task plans for mobile device automation.

Architecture

Mobile Application
      ↓
HTTP POST → /v1/chat/completions
      ↓
Jarvis AI (Reason → Plan → Output)
      ↓
JSON automation plan returned
      ↓
Mobile automation engine executes locally

Project Structure

AI-Agent/
├── app.py                  # Main app — Gradio UI + FastAPI endpoints
├── jarvis/
│   ├── __init__.py
│   ├── agent.py            # Core agent (Reason → Plan → Execute)
│   ├── tools.py            # Dynamic tool registry
│   ├── security.py         # Restricted-action enforcement
│   └── prompts.py          # System prompt builder
├── tool_registry.json      # Tool definitions (editable without code changes)
├── requirements.txt
├── Dockerfile
└── README.md

API Endpoint

POST /v1/chat/completions

OpenAI-compatible chat completions endpoint.

Request:

{
  "messages": [
    { "role": "user", "content": "Download a galaxy image and send it to Arun" }
  ]
}

Response:

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "model": "Qwen/Qwen2-1.5B-Instruct",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "{\"steps\":[{\"tool\":\"download_image\",\"query\":\"galaxy\"},{\"tool\":\"send_message\",\"contact\":\"Arun\",\"message\":\"Sending the galaxy image\"}]}"
      },
      "finish_reason": "stop"
    }
  ]
}

GET /health

Returns service status and available tools.

GET /v1/models

Lists the currently loaded model.

Supported Tools

Tool	Description	Parameters
open_app	Open a device application	app_name
search_google	Internet search	query
download_image	Download an image	query
download_video	Download a video	url
send_message	Send a message to a contact	contact, message
product_search	Search e-commerce platforms	query
set_alarm	Create alarm/reminder	time
save_file	Download and store a file	url

New tools can be added by editing tool_registry.json — no code changes needed.

Security Restrictions

Jarvis refuses to automate:

• Banking / payment apps (GPay, PhonePe, Paytm, etc.)
• Financial transactions
• Password managers
• Personal identity data (Aadhaar, PAN, SSN, etc.)

Restricted requests return:

{ "error": "This action is restricted for security reasons." }

Deployment to Hugging Face Spaces

Option A — Gradio SDK (recommended)

1. Create a new Space at huggingface.co/new-space
2. Choose SDK: Gradio
3. Push this repository to the Space
4. Add your HF_TOKEN as a Space secret (Settings → Secrets)
5. The Space will be available at:

   POST https://<your-space>.hf.space/v1/chat/completions
   

Option B — Docker SDK

1. Create a new Space with SDK: Docker
2. Push this repository (the Dockerfile is included)
3. Add HF_TOKEN as a Space secret

Environment Variables

Variable	Default	Description
JARVIS_MODEL	Qwen/Qwen2-1.5B-Instruct	HF model ID for the planner LLM
HF_TOKEN	(none)	Hugging Face API token for gated models
PORT	7860	Server port

Local Development

pip install -r requirements.txt
export HF_TOKEN="hf_..."
python app.py

The Gradio UI will be available at http://localhost:7860 and the API at http://localhost:7860/v1/chat/completions.

Mobile Integration (Android Example)

val client = OkHttpClient()
val json = """{"messages":[{"role":"user","content":"$userCommand"}]}"""
val body = json.toRequestBody("application/json".toMediaType())

val request = Request.Builder()
    .url("https://Valtry-mobile-agent.hf.space/v1/chat/completions")
    .post(body)
    .build()

client.newCall(request).enqueue(object : Callback {
    override fun onResponse(call: Call, response: Response) {
        val plan = JSONObject(response.body!!.string())
            .getJSONArray("choices")
            .getJSONObject(0)
            .getJSONObject("message")
            .getString("content")
        // Parse and execute the automation plan
    }
    override fun onFailure(call: Call, e: IOException) { /* handle */ }
})