Redis adds routing, playbooks and messaging to its agent memory layer
Redis Iris grows from a memory and context tool into six composable capabilities, so agent routing, auditing and inter-agent messaging run on the Redis instance you already have.
Redis has expanded Iris, its “real-time context engine” for AI agents, from a memory and retrieval tool into six separate capabilities that can be used together or on their own. Four are generally available: Intent Router, which matches the meaning of a request and sends it to a cached answer, an agent skill, a human or an LLM; Agent Playbooks, which marks some answers as safe to reuse and others as needing a fresh run of a known procedure; Agent Memory, which builds and prunes memories from user actions and documents through what Redis calls MemoryETL; and Context Gateway, a single governed entry point for an agent’s connected tools, including your own MCP servers.
Two capabilities are still in beta: Lens, which traces every agent response back to the memory, route or skill that produced it, alongside token and latency figures; and Agent Mesh, which gives agents a messaging channel over ephemeral streams with automatic reconnection for handoffs between agents. Iris itself launched in May 2026; this is its first major expansion.
Why it matters: if you’re already running Redis and bolting together a vector store, a router and a message queue to get an agent into production, Iris is pitched as a way to do that on infrastructure you already operate, rather than adding three more services to monitor and bill for.
The caveat: two of the six capabilities aren’t GA yet, and this is Redis’s own framing of what agents need, not a neutral standard. Check Lens and Agent Mesh against your own reliability bar before relying on them in production.