🤖 ATRI - Emotionally Evolving AI Companion

Your personal AI who remembers, reflects, and grows alongside you

🌐 Language: English | 简体中文

An AI companion that remembers, grows, and maintains emotional continuity

🚀 Quick Start • ✨ Features • 🖼️ Screenshots • 📚 Documentation

---

💡 What is ATRI?

ATRI is an Android app + cloud backend AI companion project. Unlike ordinary chatbots, she has:

📱 ATRI on Your Phone Chat with her anytime, anywhere Send images and documents

📔 Nightly Diary She records what happened today Written from her perspective

🧠 Long-term Memory Diaries become "memories" Recalled in future conversations

🌟 What Makes It Different?

Traditional Chatbots	ATRI's Approach
Every conversation starts fresh	📚 Remembers everything important via diary + vector memory
Emotions change instantly	🎭 PAD 3D emotion model + natural decay, emotions have inertia
One-size-fits-all responses	💕 Intimacy system affects speaking style, relationships grow
May fabricate memories	🔍 Tool registration mechanism, actively verifies when needed

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     📱 Android App (Kotlin)                      │
│              Jetpack Compose • Room • DataStore                  │
└──────────────────────────┬──────────────────────────────────────┘
                           │ HTTPS + Token Auth
                           ▼
┌─────────────────────────────────────────────────────────────────┐
│               ☁️ Cloudflare Worker (TypeScript)                  │
│    ┌─────────┐    ┌─────────┐    ┌─────────────────────┐        │
│    │   D1    │    │   R2    │    │     Vectorize       │        │
│    │Database │    │ Storage │    │   Vector Database   │        │
│    └─────────┘    └─────────┘    └─────────────────────┘        │
└──────────────────────────┬──────────────────────────────────────┘
                           │ OpenAI-Compatible API
                           ▼
┌─────────────────────────────────────────────────────────────────┐
│                   🤖 AI Model Service (Swappable)                │
│        OpenAI • Claude • Gemini • DeepSeek • Local Models        │
└─────────────────────────────────────────────────────────────────┘

---

🚀 Quick Start

📋 Prerequisites

Requirement	Description
💻	Computer (Windows / macOS / Linux)
☁️	Cloudflare account (free)
🔑	OpenAI API Key or compatible API
📦	Node.js 18+
🐍	Python 3.8+

1️⃣ Deploy the Backend

🪟 Option A: Windows One-Click Deploy (Recommended for beginners)

1. Double-click scripts/deploy_cf.bat
2. Follow the prompts to enter:
   • Worker name (press Enter for default)
   • D1 database name (press Enter for default)
   • R2 bucket name (press Enter for default)
   • Vectorize index name (press Enter for default)
   • OPENAI_API_KEY (required)
   • EMBEDDINGS_API_KEY (required for vector memory, SiliconFlow recommended)
   • Other optional secrets (can skip)
3. The script will automatically create resources and deploy
4. Copy the Worker URL when done

🍎 Option B: macOS / Linux Manual Deploy

# 1. Clone the project
git clone https://github.com/your-username/ATRI.git
cd ATRI

# 2. Install dependencies
cd worker && npm install

# 3. Login to Cloudflare
npx wrangler login

# 4. Create D1 database
npx wrangler d1 create atri_diary
# Copy the database_id from output and paste into worker/wrangler.toml

# 5. Initialize database tables
npx wrangler d1 execute atri_diary --file=db/schema.sql

# 6. Create R2 bucket
npx wrangler r2 bucket create atri-media

# 7. Create Vectorize index
npx wrangler vectorize create atri-memories --dimensions=1024 --metric=cosine

# 8. Set secrets
npx wrangler secret put OPENAI_API_KEY

# 9. Sync prompts and deploy
cd .. && python3 scripts/sync_shared.py
cd worker && npx wrangler deploy

After successful deployment, you'll see the Worker URL:

✨ https://atri-worker.your-subdomain.workers.dev

🔐 Configure Secrets

Secret	Description	Required
OPENAI_API_KEY	Chat model API key	✅
EMBEDDINGS_API_KEY	Embeddings API key for vector memory (SiliconFlow recommended)	✅
APP_TOKEN	Client access token to protect API	Recommended

2️⃣ Install the Android App

📥 Download APK from Releases

Build a signed release APK (recommended)

1. Go to ATRI/
2. Copy keystore.properties.example -> keystore.properties (do not commit it)
3. Generate a keystore (Windows example):
   • keytool -genkeypair -v -keystore keystore\\atri-release.jks -alias atri -keyalg RSA -keysize 2048 -validity 10000
4. Fill passwords in ATRI/keystore.properties
5. Build:
   • cd ATRI && .\\gradlew.bat :app:assembleRelease

Output: ATRI/app/build/outputs/apk/release/app-release.apk

3️⃣ Initial Setup

Step	Action
1	Welcome Screen: Set your nickname and avatar
2	Settings (tap ⚙️ icon): Enter Worker URL and App Token
3	Select Model: Choose a model based on your API
4	Start Chatting ✨

---

⚠️ Troubleshooting

❓ Deploy script says "node not found"

Install Node.js 18+: https://nodejs.org/

❓ Deploy script says "Python not found"

Install Python 3.8+: https://www.python.org/downloads/

❓ wrangler login keeps spinning

Check your network connection. You may need a VPN in some regions.

❓ Chat not responding

1. Verify Worker URL is correct
2. Check if OPENAI_API_KEY is valid
3. Check Worker logs in Cloudflare dashboard

❓ Diary not generating

Diaries are generated daily at 23:59 Beijing time. There must be conversation records for that day. You can also manually trigger regeneration on the diary page.

❓ How to use other AI services?

Any OpenAI-compatible API works:

1. Edit OPENAI_API_URL in worker/wrangler.toml
2. Optional: Configure separate DIARY_API_URL / DIARY_MODEL (diary generation can use a different model)
3. Embeddings (vector memory): by default it's https://api.siliconflow.cn/v1 + BAAI/bge-m3
   • Set EMBEDDINGS_API_KEY (SiliconFlow API key)
   • Optional: override EMBEDDINGS_API_URL / EMBEDDINGS_MODEL
4. Redeploy: cd worker && npx wrangler deploy

---

✨ Key Features

🎭 In-Character Authentic personality defined in prompts.json

💬 Context Memory Today's conversations inform responses

📖 Auto Diary Nightly reflections from her perspective

🧠 Long-term Memory Vector-stored memories awakened when needed

🖼️ Rich Media Send images or docs she understands them

🔬 Technical Highlights

Feature	Description
PAD Emotion Model	3D emotion coordinates (Pleasure/Arousal/Dominance) + natural decay
Intimacy System	Relationship temperature affects reply style, fades without maintenance
Tool Registration	Model actively verifies memories, doesn't fabricate
Split Architecture	Chat and diary can use different upstreams independently
Signed Access Control	Path-based signatures solve model dropping query params

---

🖼️ UI Preview

Welcome Set nickname and avatar	Chat Immersive chat experience	Sidebar Status and quick actions
Diary View ATRI's diary entries	Settings Configure backend and model

---

📁 Project Structure

.
├── ATRI/                    # 📱 Android App
│   ├── app/src/main/
│   │   ├── java/me/atri/
│   │   │   ├── data/        # Data layer (API, DB, Repository)
│   │   │   ├── di/          # Dependency Injection (Hilt)
│   │   │   ├── ui/          # UI layer (Compose)
│   │   │   └── utils/       # Utilities
│   │   └── res/             # Resources
│   └── build.gradle.kts
│
├── worker/                  # ☁️ Cloudflare Worker
│   ├── src/
│   │   ├── routes/          # API routes
│   │   ├── services/        # Core services
│   │   ├── jobs/            # Scheduled jobs
│   │   └── utils/           # Utility functions
│   ├── db/schema.sql        # Database schema
│   └── wrangler.toml        # Worker config
│
├── shared/                  # 📝 Shared Config
│   └── prompts.json         # Personality and prompts
│
└── scripts/                 # 🔧 Deployment Scripts
    ├── deploy_cf.bat        # Windows one-click deploy
    └── sync_shared.py       # Sync prompts

---

📚 Learn More

Document	Content
📘 Tech Architecture Blueprint	Design philosophy, data flow, API contracts, extension guide
📝 Personality Definition	ATRI's personality, diary generation, memory system prompts
🗄️ Database Schema	D1 database table definitions

---

🤝 Contributing

Contributions are welcome! Feel free to open issues or submit pull requests.

Before contributing code, we recommend reading the Tech Architecture Blueprint to understand the system design.

---

📄 License

This project is licensed under the PolyForm Noncommercial License 1.0.0.

In short:

• ✅ Personal learning, research, non-commercial use
• ❌ Commercial use requires separate authorization

---

If this project helps you, consider giving it a ⭐

_{Built with ❤️ for those who believe AI can be more than just a tool}