Thanks to Xintong, Sean, and Chris for initiating this proposal.
I believe this is one of the most exciting FLIPs in the Flink community —
especially in the era of GenAI. It opens up great opportunities for Flink
to
evolve into a more intelligent and autonomous data processing system.
+1 for the FLIP and for creating the "flink-agents" sub-project!
Looking forward to seeing this project grow.
Best regards,
Jark
On Wed, 21 May 2025 at 19:45, Guowei Ma <guowei....@gmail.com> wrote:
> Great to see this discussion getting started — thanks Xintong, Sean, and
> Chris!
> +1 to initial Flink Agents sub project.
>
> Best,
> Guowei
>
>
> On Wed, May 21, 2025 at 5:52 PM Jing Ge <j...@ververica.com.invalid>
> wrote:
>
> > Thanks Xintong, Sean, and Chris for starting this discussion!
> >
> > +1 for the initiative!
> >
> > After reading the FLIP, I have some questions and would love to
> understand
> > your thoughts, thanks!
> >
> > 1. Glad to see that Flink Agent will support MCP natively. In this case,
> > the Flink Agent should be acting as a MCP Host with multiple MCP Clients.
> > Great choice! Afaiu, one thing MCP tries to do is to standardize the way
> > how LLMs interact with prompts, tools, etc, via Context and let LLM make
> > the decision, i.e. the true agentic way. The code example shows
> > explicitly prompt("..."), tools("...") methods called alongside
> > mcpServer("...") registration call, which shows additional prompts and
> > tools context bypasses MCP. May I know the thoughts behind this design?
> >
> > 2. Will A2A protocol also be supported natively like MCP?
> >
> > 3. It is a great idea to encourage asynchronous agent-to-agent
> > communication, because it will indeed solve a lot of issues. In addition
> to
> > that, will synchronous agent-to-agent communication also be considered? I
> > think there will still be many cases that prefer synchronous agent calls.
> >
> > 4. Currently, common agent frameworks contain two major blocks: 1. Data
> > pipeline to build embedding(RAG), e.g. LlamaIndex, 2. Agent workflow,
> e.g.
> > LlamaIndex Workflow or LangGraph. I can see that Flink can naturally
> offer
> > a data pipeline solution in real-time and still need a lot of effort to
> > build a next level abstraction for it, like parsering, extracting,
> > embedding, querying, retrieving, etc.. While Agent workflow is more or
> > less new for Flink. My question is what is the major focus of Flink Agent
> > as the next step? Data pipeline for embedding vs. workflow, or both?
> >
> > 5. Did you consider Memory for Agent, especially with the real-time
> > streaming data powered natively by Flink? Some people might think, as
> LLMs
> > will support much larger tokens, there will be less and less requirement
> of
> > using memory. It is correct for some use cases. For others, considering
> > cost, hallucination, context switch(like behaviours of human beings),
> > memory might still be a very important component for Agents.
> >
> > Best regards,
> > Jing
> >
> > On Wed, May 21, 2025 at 5:56 AM Jingsong Li <jingsongl...@gmail.com>
> > wrote:
> >
> > > Thanks Xintong, Sean, and Chris for starting this discussion.
> > >
> > > +1 to initial Flink Agents sub project.
> > >
> > > Look forward to its first release!
> > >
> > > Best,
> > > Jingsong
> > >
> > > On Wed, May 21, 2025 at 11:21 AM Xintong Song <tonysong...@gmail.com>
> > > wrote:
> > > >
> > > > Hi devs,
> > > >
> > > > Sean, Chris and I would like to start a discussion on FLIP-531 [1],
> > about
> > > > introducing a new sub-project, Flink Agents.
> > > >
> > > > With the rise of agentic AI, we have identified great new
> opportunities
> > > for
> > > > Flink, particularly in the system-triggered agent scenarios. We
> believe
> > > the
> > > > future of AI agent applications is industrialized, where agents will
> > not
> > > > only be triggered by users, but increasingly by systems as well.
> > Flink's
> > > > event capabilities in real-time distributed event processing, state
> > > > management and exact-once consistency fault tolerance make it
> > well-suited
> > > > as a framework for building such system-triggered agents.
> Furthermore,
> > > > system-triggered agents are often tightly coupled with data
> processing.
> > > > Flink's outstanding data processing capabilities allows seamless
> > > > integration between data and agentic processing. These capabilities
> > > > differentiate Flink from other agent frameworks with unique
> advantages
> > in
> > > > the context of system-triggered agents.
> > > >
> > > > We propose this effort as a sub-project of Apache Flink, with a
> > separate
> > > > code repository and lightweight developing process, for rapid
> iteration
> > > > during the early stage.
> > > >
> > > > Please note that this FLIP is focused on the high-level plans,
> > including
> > > > motivation, positioning, goals, roadmap, and operating model of the
> > > > project. Detailed technical design is out of the scope and will be
> > > > discussed during the rapid prototyping and iterations.
> > > >
> > > > For more details, please check the FLIP [1]. Looking forward to your
> > > > feedback.
> > > >
> > > > Best,
> > > >
> > > > Xintong
> > > >
> > > >
> > > > [1]
> > > >
> > >
> >
> https://cwiki.apache.org/confluence/display/FLINK/FLIP-531%3A+Initiate+Flink+Agents+as+a+new+Sub-Peoject
> > >
> >
>
