Redis Iris reaches GA with six tools for production AI agents
Redis has released Redis Iris in general availability, bringing six configurable memory, context and messaging capabilities to AI agents intended for production use. The company says the platform is designed to improve reliability while reducing token costs and integration work.
- Redis Iris is now generally available.
- The platform includes six configurable capabilities for AI agents.
- Four capabilities are generally available and two are in beta.
- Redis says Iris supports production reliability and lower token costs.
- Customers have used Iris in production since May 2026.
One engine for memory, context and messaging
Redis Iris combines Intent Router, Agent Playbooks, Agent Memory, Context Gateway, Lens and Agent Mesh. Redis says teams can use the capabilities together or independently, allowing them to adapt the platform to different agent architectures.
The release addresses problems Redis identifies in production deployments, including missing or poorly timed context, incorrectly stored memory, overloaded prompts, redundant calls to large language models and fragile custom integrations. According to the company, these issues can increase token and processing costs beyond the value generated by an agent.
Intent Router interprets the meaning of a request and directs it to an appropriate destination, such as stored frequently asked question responses, a skill, a person or a large language model. It also supports routing to agents and models, safeguards and fuzzy conditional logic.
What the six capabilities do
Agent Playbooks helps teams identify which responses can be reused and which require updated data and another run of a known procedure. Redis says the playbooks become more complete and accurate with each execution, helping reduce resolution times and token spending.
Agent Memory creates memory items from user actions, documents and other semi-structured data through MemoryETL. It organizes those items into hierarchies and provides inspection, auditing and provenance information. The feature can also forget information that should not be retained, including personally identifiable information.
Context Gateway gives agents a governed access point to connected context sources. Teams can use their own MCP servers, cache tools that solve particular problems and create reusable results across sessions.
Lens, which is in beta, tracks agent behavior and provides traceability. Results can be linked to the skill, route or FAQ that produced them, the memory items used by the agent and the originating session. Redis says this provides auditable evidence, task success visibility, token usage per task and latency data.
Agent Mesh, also in beta, enables communication through ephemeral flows and manages buffering and reconnection when connections fail. Persistent SSE Streams maintain response flows through interruptions, while the A2A Comms protocol provides a channel for delegating and transferring tasks between agents.
Availability and early production use
Redis Iris is available from November 6, 2026. Intent Router, Agent Playbooks, Agent Memory and Context Gateway are generally available, while Lens and Agent Mesh remain in beta.
Redis says customers have used Iris in production since its initial launch in May 2026. One example is Paradigm, a San Francisco startup that has built its entire platform on Redis. Paradigm CTO June Lee said the platform provides the memory and orchestration layer for its agents to work in real time.
For users building AI agents, the release offers a single Redis-based platform for functions that are often connected through custom integrations. The company says this can reduce the integration work required for information retrieval, memory, routing and messaging, while giving teams more visibility into agent behavior and the resources individual tasks consume.
Redis will host webinars about Iris with its engineering team on November 12 for the Americas and November 17 for the EMEA region. Registration is available through the resource page listed in the announcement.
