How Redis is Solving the AI Memory Problem

As AI agents become more capable, one challenge becomes harder to ignore: memory. How much an agent can retain, how efficiently it can recall past interactions, and how that memory evolves over time are emerging as critical questions for the industry.

In an exclusive conversation with AIM, Rowan Trollope, CEO of Redis, explains how the company is addressing the memory bottleneck in AI systems, why databases also matter alongside chips, and how Redis is positioning itself for the next phase of AI adoption.

The company is already seeing strong momentum, having crossed $300 million in annual recurring revenue (ARR), driven by rising demand for infrastructure supporting AI and agentic systems.

Trollope says the industry often simplifies the problem by focusing only on memory size. In reality, according to him, AI agents face two distinct challenges.

“One is memory, and that’s remembering things that they have seen or done that are relevant for the future,” he says. The second challenge involves bringing relevant enterprise data to AI agents in a structured and meaningful way so they can use that information effectively when making decisions.

Redis’ Agent Memory Server

To address this, Redis has built an open-source project called Agent Memory Server. The system uses a large language model to review past interactions, compress them, and extract relevant facts or preferences that are worth retaining.

“We take the memories from a day’s activities, scan them, compress them, and store only what’s useful,” Trollope says.

The harder problem, Trollope notes, is contradiction—a user might change their preference at the last minute—a problem he says the Memory Server is addressing.

Beyond memory, agents need access to enterprise data, but not unrestricted access. Redis is building what it calls a Context Engine, which allows developers to decide in advance what enterprise data should be available to an agent.

“We don’t think it’s a good idea to let agents talk directly to systems like Workday,” Trollope says. “Instead, you materialise the relevant data into Redis and expose only what the agent needs.”

He explains that enterprise data is routed through Redis Data Integration, loaded into Redis, and then exposed to the agent through semantically defined tools. This approach normalises information into a flat structure, allowing agents to access and interpret enterprise data with ease. The data is then delivered through tools such as MCP, ensuring agents receive what they need without unnecessary complexity.

The second part of the system is retrieval. When a user returns weeks later, the agent should automatically surface relevant past information, such as previously stated preferences, without having to process entire transcripts.

Optimisation is Essential 

As the demand for memory chips surges, prices are skyrocketing. In this environment, hardware alone is not the answer, and software optimisation becomes essential. According to Trollope, the key to maximising memory efficiency lies in configuring databases to handle specific workloads more intelligently.

This is where Redis finds its core value. Rather than just storing data, Redis focuses on workload optimisation to alleviate the pressure on data centres.

One of the most effective tools for this is Redis LangCache, a fully managed semantic caching service. It reduces the high costs of AI by eliminating redundant computations in chatbots, agents, and retrieval-augmented generation applications. Redis claims up to 90% API cost savings and up to 15 times faster RAG performance through its vector database.

Redis Business Model 

Trollope says Redis has made several strategic changes to position itself for the AI transition. One of the most significant moves has been its partnership with Microsoft, through which Redis now powers the Microsoft Redis offering on Azure. The company also offers its products through the Amazon and Google marketplaces.

At the same time, Redis has been increasing its investments in AI. “All the incremental resources and headcount that we’ve been adding are really going into our AI teams,” Trollope says. He adds that the company’s focus has shifted away from traditional enterprise sales towards deeper integration with emerging technology ecosystems.

The company has more than 50 customers spending over $1 million annually, representing a year-on-year increase of over 20%. Its total paying customer base stands at around 12,000, including one-third of the Fortune 100.

The company serves a range of prominent sectors, including financial services, telco, media, consumer internet, and healthcare. 

On competition from vector-native databases, Trollope dismissed the idea of vectors as a standalone category. “Every database now supports vectors,” he says. “Vector search will just be part of every data platform.” Redis’ differentiation, he says, comes down to performance and accuracy.

When it comes to balancing open source with enterprise monetisation, Trollope describes the approach as straightforward. Redis continues to invest heavily in its open-source offering, aiming to make it “the best real-time data platform in the world.” At the same time, advanced capabilities, manageability, and support are packaged into Redis Enterprise.

“We primarily monetise when customers move to the cloud and want a hosted solution,” he says. “We also monetise when they need additional capabilities and the benefits that come with the paid product.” 

India Presence 

While Redis has discussed expanding its presence in India, Trollope said no timelines have been finalised for opening a physical office in the country. The company has doubled its India headcount year over year to 129 employees and plans to continue hiring across engineering, AI, and go-to-market roles.

As AI infrastructure matures, markets like India are expected to gain importance. For now, Redis is focused on building the systems that will support that growth when it arrives.