November 2026 Summaries

Redis has released @redis/vercel-ai-sdk, an MIT-licensed open-source package that integrates Redis Agent Memory with Vercel AI SDK agents, providing managed short-term session context and long-term user memories across sessions, channels, and agents. The package can automatically restore, recall, summarize, and record conversation information through a model wrapper, or expose memory tools that let the model selectively search, save, update, and forget memories. Redis Agent Memory uses semantic search to retrieve durable user preferences and automatically compacts lengthy session histories, while scoping data by authenticated user and session identifiers. The integration includes bounded context limits, one-second read and write deadlines, optional diagnostics that exclude sensitive content, and safeguards that treat recalled memory as potentially stale rather than authoritative. It is intended for server-side ESM applications using Node.js 22 or later, AI SDK 7.x, and Zod 4.x, with Redis Cloud credentials required to provision and connect to a memory store.

Broadcom and Redis have introduced a joint reference architecture for deploying Redis Software on VMware Cloud Foundation 9.1 with vSphere Kubernetes Service, validated on Dell PowerEdge R770 servers using Intel Xeon processors. Designed for real-time workloads such as fraud detection, streaming analytics, AI agents, and feature stores, the architecture provides benchmark results, sizing recommendations, and Kubernetes- and GitOps-based deployment guidance through the Redis Enterprise Operator. Testing over a physical 25 GbE network found sub-millisecond combined P99 read/write latency in a three-node configuration and roughly 2 million operations per second at 1.0–1.5 ms combined P99 latency in a five-node deployment. The validation also demonstrated online expansion from five to ten primary shards in 56 seconds without recorded connection errors, automatic recovery during pod and worker-node disruptions, and balanced performance for two concurrent databases sharing a cluster. Key recommendations include using 16-vCPU workers to avoid NUMA-related throughput losses, tuning networking with Antrea Hybrid Encapsulation, selecting persistence options based on recovery requirements, and aligning shard counts, replicas, and placement with cluster nodes and physical failure domains.

Redis introduces Redis Iris as a context and orchestration layer intended to help AI agents operate more reliably in real-world production environments, where ambiguous requests, fragmented data, repeated interactions, security requirements, and system handoffs often cause failures. The platform aims to provide agents with selective, governed memory; reusable procedures; efficient retrieval; and continuous learning so they do not repeatedly reconstruct context or consume unnecessary model tokens. Iris combines six capabilities: Intent Router for classifying requests and directing workflows, Agent Playbooks for reusing and improving validated procedures, Agent Memory for managing persistent and auditable information while excluding sensitive data, Context Gateway for secure access to distributed enterprise sources, Lens for tracing decisions and monitoring performance, and Agent Mesh for durable communications among agents and users. In a customer-service example involving a suspicious account charge, these components classify the issue, retrieve customer and policy context, execute an approved workflow, record an audit trail, and preserve the interaction through disruptions. Redis positions Iris as a way to make agents increasingly useful, explainable, cost-efficient, and dependable over time.

Redis has introduced the next chapter of Redis Iris, its real-time context engine for enterprise AI agents, designed to reduce the complexity, cost, and unreliability associated with assembling separate memory, retrieval, routing, and messaging systems. Built alongside live Redis application data, Iris provides six customizable capabilities: Intent Router for semantic request routing and guardrails, Agent Playbooks for reusable procedures and answers, Agent Memory for governed hierarchical memory creation, Context Gateway for controlled access to context sources and tools, Lens for tracing agent decisions and performance, and Agent Mesh for resilient inter-agent communication. Redis positions the platform around navigable, adaptive, efficient, and compounding context, aiming to help agents retrieve relevant information, retain or forget data appropriately, reuse work, and improve over time. First launched in May 2026, Iris is presented as a production-oriented system for operating reliable, explainable, and scalable agents, with free messaging, technical resources, demonstrations, and webinars available for prospective users.