xintongsong commented on code in PR #809: URL: https://github.com/apache/flink-web/pull/809#discussion_r2412949183
########## docs/content/posts/2025-10-01-release-flink-agents-0.1.0.md: ########## @@ -0,0 +1,57 @@ +--- +title: "Apache Flink Agents 0.1.0 Release Announcement" +date: "2025-10-01T08:00:00.000Z" +authors: +- xtsong: + name: "Xintong Song" +aliases: +- /news/2025/10/01/release-flink-agents-0.1.0.html +--- +\<todo: update release date in: filename, header-date, header-aliases\> + +The Apache Flink Community is excited to announce the first preview release of Apache Flink Agents (0.1.0). + +## What is Apache Flink Agents + +Apache Flink Agents is a brand-new sub-project from the Apache Flink community. It's an open-source framework for building event-driven streaming agents that can operate with scalability, reliability, and real-time responsiveness. + +### Why Apache Flink Agents Matters + +While AI agents have made rapid progress in interactive applications like chatbots, most still operate outside the high-throughput, low-latency world of real-time data processing. Yet in industrial settings -- from e-commerce and finance to IoT and logistics -- critical decisions must be made instantly in response to live events: a payment failure, a sensor anomaly, a user click. These workloads demand more than just intelligence -- they require massive scale, millisecond latency, fault tolerance, and stateful coordination, all of which are strengths of Apache Flink. But until now, there’s been no unified framework to bring agentic AI patterns into this proven streaming ecosystem. Apache Flink Agents bridges this gap. + +### Key Features + +Building on Flink's battle-tested streaming engine, Apache Flink Agents inherits distributed, at-scale, fault-tolerant structured data processing and mature state management, and adds first-class abstractions for Agentic AI building blocks and functionalities -- large language models (LLMs), prompts, tools memory, dynamic orchestration, observability, and more. + +The key features of Apache Flink Agents include: Review Comment: By `side effects`, I meant things like sending an email or making a payment, which should not be repeated. On the other hand, non-deterministic LLM outputs also have an side-effect, but is a bit different. It's not something shouldn't be repeated, but may lead to in-consistency. E.g., we already paid vendor A for purchasing a product, but during the replay the LLM decides to purchase from vendor B. So my intention was not to cover the non-deterministic LLM outputs with the `agent actions and their side effects` here. And I think it might be a bit too complicated to explain this difference and the cause of the in-consistency in the announcement. And I'm not sure how use cases like regression testing and debugging can be satisfied by this feature. My understanding for the regression testing and debugging is like, we upgraded to a new model version or adjusted a prompt, and replay the agent on the same input dataset, and check whether the outcomes are still correct or improved. But with per-action consistency, the LLM invocations are actually skipped, so we won't be able to see impact of the new model or prompt. Or maybe I've not understand the use cases correctly? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
