Redis has launched Redis Iris, a context and memory platform for AI agents designed to address failures in production AI systems.
The launch adds two new tools - Redis Context Retriever and Redis Agent Memory - to existing products including Redis Data Integration, Redis Search and LangCache. Together, they form what Redis calls a context engine that sits between an AI agent and the business data it needs to complete tasks.
Redis is targeting a problem that has become more visible as companies move AI systems from pilot projects into live operations. Many agents struggle not because of model quality, it argues, but because they rely on fragmented data sources, outdated information, slow retrieval or limited memory between sessions.
That can affect customer service, fraud detection and other applications where an agent must respond using current operational data. In Redis's view, the context layer has become a critical part of the AI stack because agents need access to structured records, documents, prior interactions and real-time updates while they work.
Context layer
Redis Iris is intended to provide that layer through a single runtime. The Context Retriever component lets developers define business entities, fields, relationships and access rules, then expose them to agents through governed schemas rather than direct database queries or text-to-SQL methods.
The goal is to let agents move across connected data - such as customers, orders, shipments or support tickets - without bespoke integrations for each workflow. Agents authenticate with scoped keys and can discover and use only the tools they are permitted to access, while row-level filters are enforced on the server side.
Agent Memory, the other new element, is designed to preserve both short-term conversational state and longer-term memory for agent-based applications. It stores recent interaction history, user preferences and persistent attributes in Redis so systems can carry context across turns and sessions.
The remaining components support retrieval speed and data freshness. Redis Data Integration synchronises information from source systems such as relational databases, data warehouses and document stores into Redis, where it can be stored in formats optimised for agent access. This creates a continuously updated operational data layer that separates systems of record from the retrieval layer used by agents.
LangCache and Redis Search underpin the platform's retrieval capabilities. LangCache provides semantic caching to reduce response times and lower token use, while Redis Search handles retrieval across vector, structured, unstructured and real-time data.
Runtime focus
Redis is also making a broader argument about where AI system bottlenecks now sit. In production environments, it says, the main challenges increasingly appear at runtime - including stale state, fragmented memory, disconnected tools and slow retrieval - rather than being solved by choosing a different model.
A context engine for production AI must meet four requirements, according to Redis: agents must be able to navigate context, retrieve it quickly, trust that it is current, and improve over time through memory and accumulated interactions. Redis Iris is built around those four points.
The launch also builds on Redis's existing footprint in enterprise AI infrastructure. Redis says 43% of enterprise AI agent stacks already use Redis somewhere in the runtime layer, positioning it to expand from a performance and data-access role into the context infrastructure market.
That installed base may matter as companies try to avoid adding more specialist products to already complex AI architectures. Redis is pitching Iris as a way to combine memory, retrieval, data synchronisation and semantic search in one system, rather than relying on separate vector databases, memory services, streaming pipelines and custom integrations.
Both Redis Context Retriever and Redis Agent Memory are available in preview. The wider Iris offering is built on top of Redis infrastructure already used to store and serve vector, structured, unstructured and real-time data for AI applications.
"Agents don't have an intelligence problem. They have a context problem. They fail because their context layer is scattered, stale, slow, or hard to use," Redis said.