Bringing memory and context to AI agents
Artificial intelligence agents are becoming increasingly capable of performing complex tasks, but a fundamental challenge continues to limit their usefulness in real-world applications: remembering what matters, accessing the right information, and maintaining context across interactions.
Now, Redis is expanding its Redis Iris platform with six capabilities designed to address these shortcomings, bringing memory, context management, intelligent routing, and communication together within a single runtime environment.
Announced on November 4, 2026, the expansion represents an effort to tackle one of the most persistent obstacles facing enterprise AI adoption: transforming impressive demonstrations of autonomous agents into reliable, scalable systems that organizations can actually depend on.
Why AI agents struggle in production
Large language models have become remarkably proficient at interpreting instructions, generating content, and reasoning through complex problems.
However, deploying these capabilities as autonomous agents introduces challenges that are less about the intelligence of the underlying models and more about the infrastructure supporting them.
An AI agent might successfully resolve a customer service issue during one interaction, only to forget important details when the customer returns. Another might repeatedly retrieve the same information or execute expensive model calls to solve problems it has already encountered.
These limitations create inefficiencies, increase operating costs, and undermine confidence in automated systems.
The challenge becomes particularly significant when multiple agents must coordinate tasks, access sensitive enterprise information, or maintain consistent behavior across different applications.
As explored in Tech Trends’ recent coverage of AI agent monitoring and the importance of maintaining control over autonomous systems, ensuring that agents behave predictably and remain accountable is becoming an essential consideration for organizations adopting AI.
Redis believes addressing these problems requires a fundamentally different approach to managing the information that agents use.
“Agents need the same foundations as apps and microservices: live app data, runtime state and service communication,” said Rowan Trollope, CEO of Redis.
Rather than requiring developers to assemble separate infrastructure components for memory, retrieval, routing, and communication, Redis Iris aims to consolidate these capabilities into a unified environment.
Six capabilities designed for enterprise AI agents
The expanded Redis Iris platform introduces six customizable capabilities that developers can deploy independently or combine according to their requirements.
Four are generally available, while two remain in beta.
1. Intent Router: Understanding what users actually mean
Traditional software frequently relies on predefined commands or exact keyword matches to determine how requests should be handled.
Redis Intent Router instead uses semantic understanding to identify the meaning behind a request and direct it to an appropriate destination.
Depending on the situation, this could involve retrieving an existing answer, activating a specialized AI agent, selecting a language model, or transferring the request to a human operator.
The system also supports configurable guardrails and more flexible conditional routing.
For organizations managing multiple AI services, this approach could help reduce unnecessary processing while improving the consistency of responses.
2. Agent Playbooks: Learning from previous solutions
One of the inefficiencies of current AI agents is their tendency to repeatedly solve problems that have already been addressed.
Agent Playbooks attempts to overcome this by allowing developers to define reusable answers and procedures.
When an agent encounters a familiar situation, it can draw on an established response or workflow rather than generating an entirely new solution.
For tasks requiring updated information, the system can execute a known procedure using current data.
Redis says these playbooks can become more comprehensive and accurate through repeated use, potentially reducing resolution times and token consumption.
This could be particularly valuable for customer support, IT operations, and other environments where similar requests occur frequently.
3. Agent Memory: Remembering the right information
Perhaps the most significant capability is Redis Agent Memory, which addresses the difficulty of maintaining useful information across interactions.
The system can construct memories from user actions, documents, knowledge bases, and other semi-structured information.
Using a component called MemoryETL, it organizes this information into hierarchical structures that provide greater control over how memories are stored and retrieved.
Importantly, Redis also emphasizes selective forgetting.
Rather than indiscriminately retaining everything an agent encounters, the platform includes mechanisms for identifying information that should not be preserved, including personally identifiable information.
This distinction is critical for enterprise deployments, where maintaining useful context must be balanced against privacy, security, and data governance requirements.
4. Context Gateway: Connecting agents to enterprise information
AI agents frequently need access to information distributed across databases, business applications, and external services.
Redis Context Gateway provides a centralized, governed entry point through which agents can access these resources.
The capability supports existing Model Context Protocol (MCP) servers, enabling organizations to connect agents with tools and information sources without rebuilding every integration.
It can also cache information about which tools successfully resolve particular problems, potentially reducing repeated lookups and unnecessary model processing.
However, connecting autonomous systems to enterprise resources introduces important security considerations.
Tech Trends previously examined these challenges in its coverage of NordLayer AI Scout and the growing security risks associated with autonomous AI agents.
As agents gain access to more sensitive information and business functions, managing permissions and maintaining visibility into their activities will become increasingly important.
5. Lens: Making AI decisions traceable
One of the major barriers to enterprise adoption of autonomous AI is the difficulty of understanding precisely how an agent arrived at a particular decision.
Redis Lens, currently in beta, is designed to make agent behavior more transparent.
The capability allows organizations to trace outcomes back to the underlying skills, routing decisions, stored memories, and source interactions that contributed to them.
It also provides operational metrics, including task success, token consumption, and latency.
For organizations operating in regulated environments, such visibility could help support auditing, troubleshooting, and accountability.
However, tracing an agent’s decisions does not necessarily guarantee that those decisions are correct or that unauthorized actions can be prevented.
Organizations will still need independent security controls, appropriate authorization mechanisms, and effective human oversight.
6. Agent Mesh: Keeping multiple agents connected
As AI systems become more sophisticated, individual agents are increasingly expected to collaborate with other specialized agents.
This creates new challenges around communication, task delegation, and recovering from interrupted connections.
Redis Agent Mesh, also in beta, introduces messaging capabilities intended to make these interactions more resilient.
The system supports communication over temporary streams, automatically buffering information and reconnecting when connections are interrupted.
It also incorporates durable Server-Sent Events (SSE) streams and agent-to-agent communication capabilities.
These features are designed to help maintain continuity when agents delegate tasks or exchange information across distributed systems.
For businesses relying on interconnected AI workflows, communication reliability could become as important as the capabilities of the individual models themselves.
Moving beyond increasingly expensive prompts
A central theme of the Redis Iris expansion is the need to make AI agents more efficient.
One common approach to improving agent performance involves supplying language models with increasingly large amounts of contextual information.
Although this can improve responses in some circumstances, it also increases token consumption, processing costs, and the potential for irrelevant information to interfere with decision-making.
Redis is proposing a different approach: making context more selective, reusable, and accessible.
By retaining useful memories, reusing established procedures, and retrieving relevant information when needed, agents could potentially perform more work without repeatedly processing the same material.
This reflects a broader shift in enterprise AI development.
Rather than focusing exclusively on building larger or more capable models, developers are increasingly considering how to construct the supporting systems that allow those models to operate efficiently.
The practical value of this approach will depend on how effectively Redis Iris can deliver these improvements across different workloads.
The announcement does not provide independently verified benchmarks demonstrating reductions in operating costs or improvements in task completion rates.
Consequently, organizations evaluating the platform will need to assess its performance against their own requirements.
Early adoption and real-world applications
Redis Iris originally launched in May 2026, and the company says customers are already using its technology in production environments.
One example is Paradigm, which uses Redis as part of the infrastructure supporting its AI agents.
“Everything you see in Paradigm is stored in Redis. It is the memory and orchestration layer that lets our agents work in real time, at a scale that wasn’t possible before,” said June Lee, CTO of Paradigm.
The example illustrates how persistent memory and orchestration can support real-time AI applications.
However, the broader expansion of Iris is still relatively new, and two of its six capabilities remain in beta.
This means the full potential of the integrated platform has yet to be demonstrated across a wide range of enterprise deployments.
The next challenge for agentic AI
The expansion of Redis Iris highlights an increasingly important development in artificial intelligence.
As language models become more capable, the infrastructure surrounding them is emerging as a critical factor in determining whether autonomous agents can deliver consistent business value.
Memory, context management, communication, and observability are not necessarily the most visible aspects of AI innovation, but they are essential for systems expected to operate reliably over extended periods.
Redis is positioning Iris as a way to bring these capabilities together, reducing the complexity of building and maintaining production AI applications.
The approach is promising, particularly for organizations already using Redis infrastructure.
Nevertheless, consolidating multiple capabilities into one platform also introduces questions around interoperability, vendor dependence, and the practical effectiveness of individual components.
Organizations will need to consider these factors alongside performance, security, and cost.
Ultimately, the success of enterprise AI agents may depend less on how intelligently they respond to individual prompts and more on whether they can remember previous interactions, access trustworthy information, coordinate actions, and operate consistently over time.
With Redis Iris, the company is betting that solving those infrastructure challenges will be essential to making autonomous AI genuinely useful in the real world.