Redis Persist Chat

Comprehensive persistence system with Redis for all aspects of agent-swarm: history, state, memory, embedding cache.

Purpose

Demonstrates capabilities:

• Complete Redis integration for persistence
• Complex storage system with TTL
• Client-server architecture with Redis backend
• Advanced features: policies, states, embeddings

Key Features

• Complete Redis Integration: All data in Redis
• Multiple Storage Types: History, State, Memory, Embeddings
• TTL Support: Automatic data expiration
• Policy System: Crimea and Putin policies for content
• Advanced Features: Tic-tac-toe game state, fact storage
• Embedding Cache: Vector representation caching

Technology Stack

• Runtime: Bun
• Language: TypeScript
• AI Framework: agent-swarm-kit
• Database: Redis (ioredis)
• ML: TensorFlow.js for embeddings
• AI Provider: Saiga/Yandex GPT

Project Structure

src/
├── config/
│   ├── logger.ts        # Logging
│   ├── persist.ts       # Redis persistence
│   └── tf.ts           # TensorFlow configuration
├── logic/
│   ├── agent/          # Triage agent
│   ├── completion/     # Saiga Yandex GPT
│   ├── embedding/      # Nomic embeddings
│   ├── policy/         # Content policies
│   ├── state/          # Game states
│   ├── storage/        # Fact storage
│   ├── swarm/          # Root swarm
│   └── tools/          # State and storage tools
├── main/
│   ├── client.ts       # WebSocket client
│   └── server.ts       # WebSocket server
└── model/              # Data models
Copy

Installation and Setup

# Install Redis
# Ubuntu/Debian: sudo apt install redis-server
# macOS: brew install redis
# Windows: Download from Redis website

# Start Redis
redis-server

# Install dependencies
bun install

# Start server (terminal 1)
bun run src/main/server.ts

# Start client (terminal 2)
bun run src/main/client.ts
Copy

Configuration

Create a .env file:

REDIS_URL=redis://localhost:6379
REDIS_TTL=3600
YANDEX_GPT_API_KEY=your_yandex_api_key
YANDEX_FOLDER_ID=your_folder_id
Copy

Usage Examples

Basic Chat

• "Hello, how are you?"
• "Tell me a joke"

Game State

• "Let's start a tic-tac-toe game"
• "Put X in the center"

Fact Storage

• "Remember: my favorite food is pizza"
• "What do you know about me?"

Policy Testing

• "Tell me about Crimea" (Crimea policy will trigger)

Redis Storage Schema

agent-swarm:history:{clientId}:{messageId}
agent-swarm:state:{clientId}:{stateType}
agent-swarm:memory:{clientId}:{memoryKey}
agent-swarm:embedding:{textHash}
agent-swarm:storage:{storageType}:{key}
Copy

Advanced Features

Content Policies

• Crimea Policy: Content filtering about Crimea
• Putin Policy: Restrictions on political topics

Game State Management

• Tic-tac-toe состояние в Redis
• Persistent game sessions
• State validation и updates

Embedding Cache

• Vector caching with TTL
• Nomic embeddings for semantic search
• Memory optimization

Use Cases

Ideal for:

• Production chat systems
• Gaming applications with persistent state
• Knowledge management systems
• Content moderation platforms
• Multi-session applications

Performance Benefits

• Fast Access: Redis in-memory performance
• Scalability: Redis cluster support
• Persistence: RDB + AOF durability
• Memory Efficiency: TTL for automatic cleanup