Redis has launched Iris, a new platform designed to give AI agents persistent context and memory. The offering targets a recurring pain point in enterprise deployments: the gap between what a large language model knows from its training data and what it needs to know about a specific user, session, or business process in real time.
Bridging the data retrieval gap
Iris works as a middleware layer that stores and retrieves contextual information — conversation history, user preferences, session state — so an AI agent doesn't have to start fresh with every interaction. By keeping that memory outside the model itself, the platform lets enterprises reduce the number of expensive API calls to inference endpoints and cut the infrastructure costs tied to frequent re-processing.
Redis, best known for its in-memory data store, said the platform is built on top of its existing database technology. The company claims Iris can help companies accelerate AI adoption without rebuilding their entire stack.
Why enterprises need a memory layer
Most AI agents today operate statelessly. They process a single request, generate a response, and forget everything. For customer support bots, coding assistants, or internal knowledge-management tools, that means the agent can't remember what the user said two turns ago or what decisions were made in a previous session. Iris provides a dedicated space to store that kind of context, updating it as the conversation evolves.
The platform also handles retrieval — fetching the right pieces of past data when a new query comes in. That avoids the need to feed an entire conversation history into the model prompt, which saves tokens and speeds up responses.
Cost and efficiency gains
Redis framed Iris as a way to lower the total cost of operating AI agents. By reducing the number of model invocations and trimming the size of each prompt, companies can cut down on pay-per-token charges from model providers. The company also highlighted that keeping context in memory rather than in a separate database or caching layer reduces latency.
Exactly how much enterprises can save depends on their usage patterns. Redis did not release specific benchmarks or pricing tiers for Iris in the announcement.
Availability and next steps
Iris is now available through Redis. The company said the platform integrates with existing AI agent frameworks and can be deployed on-premises or in the cloud. Enterprises looking to test it can sign up for early access.
Redis plans to release more technical documentation and case studies in the coming weeks. For now, the company is betting that a specialized memory layer will become as standard for AI agents as a database is for web applications.
