Redis debuts the much-needed memory layer for enterprise AI agents

Artificial intelligence agents have a memory problem and now Redis Inc., the database management startup, is trying to fix that with its new, real-time Context Engine.

As the company explains, it’s all about helping enterprise AI agents move beyond simply chatting to users and making them productive workers in their own right. Redis explained that there are three core tools behind the Context Engine, including the Redis Context Retriever, Redis Agent Memory and Redis Data Integration, with the latter made generally available starting today.

The three tools are designed to solve what Redis terms the “context problem” in enterprise AI, which causes autonomous systems to hallucinate and output incorrect information or results, or sometimes even stall from a lack of data. The company argues that the context problem is the result of a lack of memory, which causes problems when AI agents are asked to perform complex tasks. For instance, if an agent is trying to resolve a customer’s issue on the phone, it might need to pull data from the customer relationship management system, a shipping database and a PDF that outlines company policies.

Without a dedicated context engine, the only way to do this is to use brittle, onetime integrations that are slow and difficult to maintain. Redis is aiming to provide a dedicated layer for agents that sits between them and the data, powered by its powerful, in-memory data store. It gives each agent an “agent-readable” view of the environments they operate in, defining business entities and their relationships so that they immediately understand the bigger picture around whatever problem they’re tasked with solving.

The new Context Retriever is perhaps the most critical innovation in the Context Engine. Currently available in preview, it allows developers to create a semantic model of their business data, so agents can map how each customer relates to an opportunity or support ticket. Rather than forcing agents to gamble with “text-to-SQL” queries that often break down, the retriever will automatically generate the tools required by the agent to grab the data they need, using the open-source Model Context Protocol.

Meanwhile, the Agent Memory component in preview from today provides a “dual-layered” approach to the agent’s state. It helps manage the short-term interaction history while also creating a more durable, long-term memory cache that agents can use to remember preferences and previous interactions from past sessions.

The new Data Integration is the final piece of the puzzle, providing the plumbing for the Context Engine. Its job is to continuously synchronize business data from the company’s main relational databases and data warehouses. This means that agents will always act on the most up-to-date data, rather than information that might be several weeks old.

What makes this offering compelling is Redis’ strong enterprise presence. Its flagship open-source in-memory data store, which functions as a database, cache, streaming engine and message broker, already exists in 43% of all enterprise AI agent stacks.

Its evolution from a high-speed cache into a sophisticated context layer suggests that Redis believes it can become a kind of operating system for AI agents. For customers, the big promise is that they’ll be able to create more complex and reliable agents that don’t break down when trying to automate more advanced business tasks.

Image: SiliconANGLE/Gemini