AI 模型和平台
Redis Broadens Iris 带有 Agent 路线、可观察性和信使
在 Google 上的首选来源中添加 Unite.AI
2026年11月4日Redis 宣布，其Iris 实时上下文引擎现在为AI agents提供6个可堆肥能力，4个一般可用，2个在β. 扩展的新增请求向以前以上下文和内存为中心的产品提供路由，可重复使用的程序，可观察性和弹性信息。
客户可以在一次运行中独立或任意组合地一起使用能力。 Rowan Trollope的理由是，开发者不应需要组装单独的基础设施产品来给agents内存，相关上下文，实时数据访问，路由和通信。 Redis开始提供5000条免费信息。
由Redis的软件工程总监Simba Khadder(Simba Khadder)共同进行的一项技术深度潜水，描述了设计要解决的生产问题：agents开始没有以前的历史，保留敏感的细节，同时失去有用的偏好，使用不完整的信息工作，并反复重建他们已经处理过的请求的背景。 这些故障模式将释放的特性与连续性、再利用和控制联系起来。
四种能力一般可用
意向路由器与请求的含义相匹配，而不是其确切的措辞，然后将它指向一个被记住的FAQ,agent技能，一个人类或LLM. 它还支持agent和模型的路由，护栏和模糊的有条件决定，允许将同一请求的不同措辞版本作为相同的根本问题处理。
Agent 电话 Playbook将可重复使用的稳定答案与需要当前数据和新程序运行的请求区分开来。 Redis 称游戏本随着反复使用，剪切解析时间和象征性支出而变得更加彻底和准确。 这使得Playbooks的重复使用对应的Intentent路由器：一个识别请求的类型，而另一个则决定一个现有的答案或程序是否能够满足它。
Agent Memory使用记忆ETL从用户动作，文档，知识基础和其他半结构数据中构建记忆。 它分级地组织这些记忆来进行精细的刻度控制，并提供检查，审计和出处的特征，以便各小组能够看到每个记忆的来源。 它还可以放弃它不应保留的信息，包括个人可识别的信息。
上下文网关提供了一个连接上下文源的受控切入点。 客户可以带来自己的MCP服务器，缓存用于解决特定问题，并构建可重复使用的结果，以便各会话保持一致。 Redis将工具缓存作为减少象征性使用的一种方式，将发行的重点扩大到避免重复的模型工作从答案和程序到数据访问。
Beta 工具覆盖审计和连接失败
伦斯将agent结果追溯到为它服务的技巧，路线或FAQ，所使用的内存和起源会话。 它还揭示了任务的成功、每项任务的迹象和耐久性，将审计线索与业务衡量结果挂钩。 其结果旨在让agent行为在团队需要诊断错误答案，调查象征性消费或确定为何发生动作时可以查询。
Agent Mesh地址为通信中断。 Agents在ephemeral 溪流上通信，在连接下降时自动缓冲和重联。 持久的SSE Streams旨在让响应通过中断流畅，而A2A Commons则为agent代表团和交接提供通道。 因此，Lens和Mesh涵盖不同的生产风险：重建决定和保持对话。
从5月发布到客户使用
2026年5月18日Redis 推出 Iris作为agent与其需要的数据之间的上下文和内存层。 该发布结合了上下文 Retriever,Agent内存，Redis数据集成，LangCache和Redis搜索，预览了上下文 Retriever和Agent内存。 Agent 电话内存从此转移到了一般可用性，而新的线条则将原始的检索和记忆重点扩展到了路由，可重复的程序，决策追踪和流回。
Safe in Home 资深软件开发商Nick Thompson描述了Redis Iris产品页面上的具体部署：“Redis Agent Memory帮助我们实时存储和重复使用我们编码的agents的上下文。” 他的团队在agents工作时用它来捕捉工程决策，bug细节和开发背景，保留整个团队现有的信息，而不是在会话或工具之间丢失。 一个服务在监控公司的API时也会读取和写到这个内存，然后将bug报告实时存入其售票系统中。
因此，这一发布扩大了Iris的作用，从提供上下文到协调agent如何处理请求、重新使用先前的工作、记忆选定的信息并到达受管制的数据。 不过，其成熟程度参差不齐：这些业务基础一般都有，而决策-审计和连接-恢复层仍然在β。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
AI Models & Platforms
Redis Broadens Iris With Agent Routing, Observability and Messaging
Add Unite.AI to your preferred sources on Google
Redis announced on November 4, 2026 that its Iris real-time context engine now offers six composable capabilities for AI agents, with four generally available and two in beta. The expansion adds request routing, reusable procedures, observability and resilient messaging to a product previously centered on context and memory.
Customers can use the capabilities together, independently or in any combination within one runtime. Rowan Trollope’s rationale is that developers should not have to assemble separate infrastructure products to give agents memory, relevant context, live data access, routing and communication. Redis’s getting-started offer includes 5,000 free messages.
A companion technical deep dive by Simba Khadder, Redis’s director of software engineering, describes the production problems that design is meant to address: agents starting without prior history, retaining sensitive details while losing useful preferences, working with incomplete information and repeatedly rebuilding context for requests they have already handled. Those failure modes connect the release’s otherwise separate features around continuity, reuse and control.
Four Capabilities Are Generally Available
Intent Router matches a request’s meaning rather than its exact wording, then directs it to a remembered FAQ, an agent skill, a human or an LLM. It also supports agent and model routing, guardrails and fuzzy conditional decisions, allowing differently phrased versions of the same request to be handled as the same underlying problem.
Agent Playbooks separates stable answers that can be reused from requests that need current data and a fresh run of a known procedure. Redis said the playbooks become more thorough and accurate with repeated use, cutting resolution times and token spending. That makes Playbooks the reuse counterpart to Intent Router: one identifies the kind of request, while the other determines whether an existing answer or procedure can satisfy it.
Agent Memory uses MemoryETL to build memories from user actions, documents, knowledge bases and other semi-structured data. It organizes those memories hierarchically for finer-grained control and provides inspection, audit and provenance features so teams can see where each memory came from. It can also discard information it should not retain, including personally identifiable information.
Context Gateway provides one governed entry point to connected context sources. Customers can bring their own MCP servers, cache which tools solve particular problems and build reusable results for consistency across sessions. Redis presents tool caching as a way to reduce token use, extending the release’s emphasis on avoiding repeated model work from answers and procedures to data access.
Beta Tools Cover Auditing and Connection Failures
Lens traces an agent outcome to the skill, route or FAQ that served it, the memory used and the originating session. It also exposes task success, tokens per task and latency, pairing an audit trail with operational measurements. The result is intended to make agent behavior queryable when teams need to diagnose a wrong answer, investigate token consumption or establish why an action occurred.
Agent Mesh addresses communication interruptions. Agents communicate over ephemeral streams, with automatic buffering and reconnection when connections drop. Durable SSE Streams are designed to keep responses flowing through interruptions, while A2A Comms supplies a channel for agent delegation and handoffs. Lens and Mesh therefore cover distinct production risks: reconstructing a decision and preserving a conversation.
From May Launch to Customer Use
Redis introduced Iris on May 18, 2026 as a context and memory layer between an agent and the data it needs. That launch combined Context Retriever, Agent Memory, Redis Data Integration, LangCache and Redis Search, with Context Retriever and Agent Memory in preview. Agent Memory has since moved to general availability, while the new lineup extends the original retrieval-and-memory emphasis into routing, repeatable procedures, decision tracing and stream recovery.
Safe in Home senior software developer Nick Thompson described a concrete deployment on the Redis Iris product page: “Redis Agent Memory helps us store and reuse context across our coding agents in real time.” His team uses it to capture engineering decisions, bug details and development context as agents work, keeping the information available across the team instead of losing it between sessions or tools. A service also reads from and writes to that memory while monitoring the company’s API, then files bug reports in its ticketing system in real time.
The release consequently broadens Iris’s role from supplying context to coordinating how an agent handles a request, reuses prior work, remembers selected information and reaches governed data. Its maturity is uneven, though: those operational foundations are generally available, while the decision-auditing and connection-recovery layers remain in beta.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.