Redis Iris adds six capabilities for production-ready AI agents
Redis has outlined the next chapter of Redis Iris, a runtime context engine designed to help AI agents work reliably in production. The company says agents need faster access to relevant context, selective memory and more efficient ways to handle repeated tasks.
- Redis says AI agents need relevant context faster and with less repeated work.
- Redis Iris provides six composable capabilities in one runtime context engine.
- Intent Router and Agent Playbooks are listed as generally available.
- Lens and Agent Mesh are listed as beta capabilities.
- Redis says Iris launched in May 2026.
Why AI agents struggle outside demos
Redis says many agents perform well in controlled demonstrations but fail in production because real-world interactions are unpredictable. Customers use multiple channels, ask questions in unexpected ways and move between agents, people and systems. Each handoff can create another point of failure.
The company identifies several recurring problems. Agents may start from scratch during every interaction, remember irrelevant information while forgetting useful preferences, or act without enough context. Redis gives the example of a support agent approving a $350 refund 45 days after a purchase, despite a policy that allows no refunds after 30 days.
Repeated work also raises costs. According to Redis, agents can rebuild context windows, make redundant LLM calls and use bloated prompts to solve problems they have already encountered. The company argues that the central issue is not a lack of reasoning ability, but the absence of the right context at the right time.
Redis Iris combines six capabilities
Redis Iris places a layer between an agent and the data and tools it needs to act. Redis says the system is built around four principles: navigable context, adaptive memory, efficient operation and compounding knowledge that improves through use.
Its six capabilities can be used together or independently. Intent Router, listed as generally available, matches the meaning of a request and routes it to a remembered FAQ, an agent skill, a human or an LLM. Agent Playbooks, also generally available, define which answers can be reused and which require current data and a fresh procedure.
Redis Agent Memory builds memories from documents, knowledge bases and other semi-structured data through MemoryETL. It supports memory hierarchies, inspection, audit and provenance, and can forget information such as personally identifiable information. Context Gateway provides one entry point to connected context sources, supports bringing your own MCP and can cache tool results.
Lens is in beta and traces outcomes to the skill, route or FAQ used, the memory involved and the source session. It also provides visibility into task success, tokens per task and latency. Agent Mesh, also in beta, supports communication over ephemeral streams, automatic reconnection, durable SSE streams and A2A communications for delegation and handoffs.
A production workflow for customer service
Redis illustrates the system with a customer asking, “What’s the charge on my account?” Intent Router classifies the request as a possible dispute, fraud issue or routine payment rather than relying on keyword matching. Agent Playbooks then checks for an approved FAQ or an established skill, while Agent Memory retrieves relevant information about the customer.
Context Gateway supplies additional details from connected systems, such as prior habits and the address on file. In the example, the agent identifies an unfamiliar $89.99 merchant charge and offers to freeze the card and open a dispute. Because the action involves moving money, Lens records the audit trail.
Redis says Iris launched in May 2026 and that customers are already reporting production use. June Lee, CTO of Paradigm, said, “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.”
