Redis Agent Memory: Managed Memory for AI Agents
Redis Agent Memory is a memory service for AI agents from Redis, the company behind one of the most widely used databases. It keeps two kinds of memory: session memory, the running record of the current conversation, which it summarizes automatically as it grows; and long-term memory, the facts and preferences it pulls out of those conversations in the background so an agent can recall them weeks later. It’s part of Redis Iris, a set of tools for giving agents memory, context and messaging, and it became generally available on November 4, 2026. Before that, Redis built an open-source Agent Memory Server, which it now keeps as a research project.
The quick answer: Redis Agent Memory is a solid choice for engineering teams that already run Redis and want memory as part of their infrastructure rather than another vendor. It’s straightforward by design: session events in, extracted memories out, with summarization, deduplication and controls over what gets kept. Redis has also been unusually candid about benchmarks, publishing its own comparison and noting where its best research results aren’t in the product yet. A disclosure: we make Carly, an AI assistant with its own memory, so we compete on the memory side. Everything below comes from Redis’s product pages, documentation, press releases and GitHub repository.
Redis Agent Memory at a glance
Checked against Redis’s product page, documentation, press releases and GitHub on November 12, 2026.
| Redis Agent Memory | |
|---|---|
| What it is | A managed memory service for AI agents, with session and long-term memory |
| Part of | Redis Iris, Redis’s real-time context engine for agents |
| Status | Generally available since November 4, 2026; preview since May 2026 |
| Access | Python and TypeScript SDKs, the iris CLI, REST API and an MCP server |
| Deployment | Managed on Redis Cloud, or self-managed on Kubernetes |
| Pricing | Through Redis sales |
| Open source | The original Agent Memory Server, Apache 2.0, no longer actively maintained |
How it works
Redis’s documentation describes two tiers, which it compares to human memory:
- Session memory is working memory. Your app sends each new message as it happens, and Redis stores the ordered events with who sent them, when and any metadata. You set how long sessions are kept. Once a conversation passes a threshold, older messages are summarized while recent ones stay in full, so the conversation doesn’t overflow the model’s context window.
- Long-term memory is what survives after the session ends. In the background, Redis reads the session and extracts durable facts and preferences, storing them as searchable records with embeddings. Its example: a traveler mentions they’re vegetarian while planning a trip to Japan, and “The user is vegetarian” becomes a memory a later session can find, even after the original conversation has expired.
Extraction compares each new memory with existing ones and uses a model to decide whether it’s new or a duplicate. Long-term memories come in built-in types, semantic for general facts and episodic for snapshots of a specific session, and teams can define their own, such as a trip_preference type with fields for destinations, dates and dietary needs. Teams can also steer extraction away from sensitive information and write memories directly from outside data, like a bulk import.
Memories can be scoped by owner, session, namespace and type. A shared namespace can hold something several users need, like a team’s travel policy, while personal namespaces keep individual memories apart.
What the GA release added
Redis’s November 4 announcement says Agent Memory now “builds memories from user actions as well as documents and other semi-structured data” through a feature called MemoryETL, organizes memories in hierarchies, and provides inspection, audit and provenance. “It can also forget what it should not keep, such as personally identifiable information.”
The product page adds a few more points: one memory shared across channels, “from chat to email to voice,” duplicate prevention with content hashing, OAuth2 or JWT authentication, and memories that multiple agents can share so they build on each other’s work.
Part of Redis Iris
Redis launched Iris in May 2026 and expanded it in November. It now has six capabilities that can be used together or separately:
| Capability | What it does | Status |
|---|---|---|
| Agent Memory | Session and long-term memory | Generally available |
| Intent Router | Sends a request to the right place by meaning: a saved answer, a skill, a person or a model | Generally available |
| Agent Playbooks | Reuses stable answers and known procedures | Generally available |
| Context Gateway | One governed entry point to connected data and MCP servers | Generally available |
| Lens | Traces each agent outcome to the route, memory and session behind it | Beta |
| Agent Mesh | Reliable messaging and handoffs between agents | Beta |
Redis’s pitch is consolidation. “Developers shouldn’t need five infrastructure products to make agents work,” said Rowan Trollope, Redis’s CEO, in the announcement. Iris’s original launch also brought in existing Redis products, including LangCache, which caches answers to similar questions to cut model costs, and Redis Search.
Deployment and access
On Redis Cloud, Agent Memory is a managed service you create next to a database. For companies that need to run it themselves, Redis offers a self-managed version deployed on Kubernetes with the redis-iris chart, where you provide the Redis databases that hold memory, background jobs and metadata. Either way, apps reach it through the Python or TypeScript SDK, the iris command-line tool, the REST API or an MCP server, and it works with any agent framework or model provider.
Redis doesn’t publish Iris pricing; its Iris page routes pricing questions to sales.
The open-source Agent Memory Server
Redis’s first take on agent memory was the open-source Agent Memory Server, started in March 2025 and licensed Apache 2.0. It offered a REST API and an MCP server, working memory and long-term memory, configurable extraction strategies, topic and entity extraction, and support for many model providers. Redis says it informed the Iris product.
That code now lives in a V0 folder of the repository, which states that it “is no longer actively maintained and is not the supported production path.” Its last release was in April 2026. Developers who built on it can still run it, but Redis points new projects to the managed service.
Benchmarks, honestly reported
Redis published a LongMemEval study comparing memory approaches on 500 questions about long, multi-session chat histories, using the same answer model and judge across systems. Its strongest research configuration, which combines extracted facts with search over the raw conversation, scored about 86%, the top result in its comparison.
Two caveats come straight from Redis. The repository says plainly that this combined configuration isn’t in Redis Agent Memory yet, so the score isn’t a product score. And it notes that “Some vendors publish higher scores than we report. We were not able to reproduce those numbers under this protocol.” That’s a useful reminder for anyone comparing memory vendors on self-reported numbers.
Who uses it
Redis cites a few early users. June Lee, CTO of Paradigm, said in the November announcement that “Everything you see in Paradigm is stored in Redis.” On the Iris page, a developer at Safe in Home describes using Agent Memory to keep engineering decisions and bug details available across a team’s coding agents, with a monitoring service writing to the same memory. Redis as a company says it serves more than 10,000 customers. We found no independent reviews of Agent Memory itself, which is expected for a service this new.
How Redis Agent Memory compares
Redis Agent Memory competes with memory layers like Supermemory, Zep, Cognee, MemoryLake, Honcho and Mem0, and with libraries like LangMem. Its strengths are speed, a simple two-tier model, and fitting into infrastructure many teams already run. It deliberately does less interpretation than some rivals: there’s no knowledge graph like Zep’s or Cognee’s, no reasoning about who a user is like Honcho’s, and no consumer product like MemoryLake’s Memory Passport.
Redis Agent Memory is plumbing for teams building agents. If you’d rather have an assistant that already has memory, Carly has memory built in: it saves named memory documents that it reads and searches in later conversations, searches a knowledge base of files you upload by meaning, and keeps your to-dos, data tables and CRM in one workspace. There’s no infrastructure to stand up, it works inside ChatGPT, inside Claude or by text, and it also labels and sorts email across all your accounts, handles advanced scheduling, and makes and answers calls and texts.
Who Redis Agent Memory suits
Redis Agent Memory is a good fit if:
- Your team already runs Redis or Redis Cloud and wants memory alongside it.
- You want simple session and long-term memory with automatic summarization and extraction.
- You need to run memory in your own Kubernetes environment.
- You’re also considering other Iris pieces, like routing or semantic caching.
Look elsewhere if:
- You want a knowledge graph or deeper reasoning about users.
- You want published, self-serve pricing.
- You want an open-source memory server that’s actively maintained.
FAQ
What is Redis Agent Memory?
Redis Agent Memory is a managed memory service for AI agents, part of Redis Iris. It stores conversation events in session memory, summarizes them as they grow and extracts durable facts into long-term memory for later sessions.
Is Redis Agent Memory open source?
The managed service isn’t. Redis’s earlier Agent Memory Server is open source under Apache 2.0, but Redis says it’s no longer actively maintained and isn’t the supported production path.
How much does Redis Agent Memory cost?
Redis doesn’t publish pricing for Iris or Agent Memory. Pricing is available through Redis’s sales team.
Does Redis Agent Memory have an MCP server?
Yes. It can be reached through an MCP server, as well as Python and TypeScript SDKs, the iris CLI and a REST API.
What is Redis Iris?
Redis Iris is Redis’s real-time context engine for AI agents. It includes Agent Memory, Intent Router, Agent Playbooks and Context Gateway, with Lens and Agent Mesh in beta.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
