	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > ADK with Redis for Fast Agent State Storage

ADK with Redis for Fast Agent State Storage

Author: Venkata Sudhakar

ADK agents maintain state across turns using session services. The default InMemorySessionService is lost when the process restarts. For production, you need a persistent store. Redis is ideal - it offers microsecond reads, TTL-based expiry, and horizontal scaling. ShopMax India uses Redis to maintain shopping cart and customer preference state across agent sessions.

You implement a custom session service by extending the ADK BaseSessionService class. The service stores session data as JSON in Redis with a configurable TTL, and loads it back on each request.

The below example shows how ShopMax India implements a Redis-backed session service for the product recommendation agent.

import redis
import json
from google.adk.sessions import BaseSessionService, Session
from typing import Optional

class RedisSessionService(BaseSessionService):
    def __init__(self, host="localhost", port=6379, ttl_seconds=3600):
        self.client = redis.Redis(host=host, port=port, decode_responses=True)
        self.ttl = ttl_seconds

def _key(self, app_name: str, session_id: str) -> str:
        return f"adk:{app_name}:session:{session_id}"

def create_session(self, app_name: str, user_id: str, **kwargs) -> Session:
        session = Session(app_name=app_name, user_id=user_id)
        key = self._key(app_name, session.id)
        self.client.setex(key, self.ttl, json.dumps(session.to_dict()))
        return session

def get_session(self, app_name: str, session_id: str) -> Optional[Session]:
        key = self._key(app_name, session_id)
        data = self.client.get(key)
        if not data:
            return None
        return Session.from_dict(json.loads(data))

def update_session(self, session: Session) -> None:
        key = self._key(session.app_name, session.id)
        self.client.setex(key, self.ttl, json.dumps(session.to_dict()))

Using the Redis session service with an ADK agent,

It gives the following output,

Session created: sess_7f3a91 (stored in Redis)

For video editing under Rs 60,000, I recommend:
1. ASUS VivoBook 15 OLED - Rs 54,990 (Ryzen 7, 16GB RAM)
2. Lenovo IdeaPad Slim 5 - Rs 57,500 (Intel i7, 512GB SSD)
3. HP Pavilion 15 - Rs 59,999 (Ryzen 5, dedicated GPU)

Best pick: ASUS VivoBook for colour accuracy on the OLED display.

In production at ShopMax India, use Google Cloud Memorystore for Redis with VPC peering to your Cloud Run services. Set TTL to match your session timeout policy and monitor cache hit rates in Cloud Monitoring to ensure Redis is serving most session lookups.

Send your comments, suggestions or queries regarding this site to [email protected].