DEPLOY.md

Image Encoder Microservice 🖼️

Standalone service for vision-language learning

What It Does

Converts any image into:

• ✅ 512-dimensional CLIP embeddings
• ✅ Text representation for LLMs
• ✅ Auto-sends to cosmo-ai for learning
• ✅ Runs on FREE HuggingFace CPU tier

Quick Deploy

1. Create HuggingFace Space:

   • Name: image-encoder
   • SDK: Docker
   • Hardware: CPU basic (free tier)

2. Upload files from /image-encoder/ folder

3. Update COSMO_AI_URL in app.py to your cosmo-ai URL

4. Deploy! ✅

How It Works

[User uploads image]
    ↓
[image-encoder processes with CLIP]
    ↓
[Converts to embedding + text]
    ↓
[Sends to cosmo-ai /api/feed/vision]
    ↓
[Cosmo-AI stores and learns]

API Usage

Encode Image

import requests
import base64

# Read image
with open("photo.jpg", "rb") as f:
    img_data = base64.b64encode(f.read()).decode()

# Encode and send to cosmo-ai
response = requests.post(
    "https://YOUR-USERNAME-image-encoder.hf.space/encode",
    json={
        "image_base64": img_data,
        "send_to_cosmo": True  # Auto-send to cosmo-ai
    }
)

print(response.json())

Upload File

files = {"file": open("image.png", "rb")}
response = requests.post(
    "https://YOUR-USERNAME-image-encoder.hf.space/encode/upload",
    files=files,
    params={"send_to_cosmo": True}
)

Features

✅ CPU Optimized

• Uses CLIP-vit-base-patch32 (150MB)
• Fast inference on free CPU
• No GPU needed

✅ Auto Keepalive

• Pings itself every 20 minutes
• Prevents space from sleeping
• Always available

✅ Cosmo-AI Integration

• Automatic data sending
• Vision feed endpoint
• Learning from visual data

Cosmo-AI Integration

Cosmo-AI now has /api/feed/vision endpoint that:

• ✅ Receives image embeddings
• ✅ Stores up to 1000 images
• ✅ Provides stats and samples
• ✅ Enables vision-language learning

Check stats:

curl https://shubhjn-cosmo-ai.hf.space/api/feed/vision/stats

Architecture

┌─────────────────┐      ┌──────────────────┐
│ image-encoder   │──────▶│   cosmo-ai     │
│  (CPU Space)    │ REST  │   (Main App)     │
│                 │ API   │                  │
│ • CLIP Model    │       │ • LLM            │
│ • Encoding      │       │ • Chat           │
│ • Keepalive     │       │ • Vision Feed    │
└─────────────────┘       └──────────────────┘

Free & Independent

• 🆓 Both services run on FREE HuggingFace tier
• 🔓 No restrictions, no API limits
• 🚀 Independent microservices
• ♾️ Keepalive prevents sleeping

Deploy and enjoy vision-enabled AI! 🎉