GitHub user wenjin272 edited a discussion: Planning Flink Agents 0.4

# Brainstorm
from @xintongsong @wenjin272 @joeyutong @pltbkd

## Must

- **Highlight Features**
  - Conduct a comprehensive review of the existing API and design a 
compatibility mechanism
  - Support asynchronous execution for Java durable execution on JDK versions 
below 21 
  - Support Sub Agent primitives #660 
  - Flink Agents Dev Skills
- **Requirements from Production**
  - API
    - Support invoking multimodal models 
    - Support injecting parameters into tool calls #853 
    - Support structured output from ChatModel #280
  - Runtime
    - Support dynamic updates to the AgentPlan, covering events, actions, and 
resources 
- **Important Experience Improvements**
  - Observability
    - Support observing Memory through the EventLog
    - Support Per-Run Trace to visualize the execution path of a single record 
through the Agent #710
    - Metrics improvements 
      - #857 
      - #858 
      - #859 
  - CI & Infra
    - Support Flink 2.3 #833
    - AI-assisted code review mechanism (explored within the release cycle; not 
tied to a release)

## If Possible

- **API**
  - Support to-C sources such as Slack / GitHub / Gmail / Jira
  - Support conditional triggering of Actions 
  - Provide syntactic sugar for common workflow patterns to reduce the effort 
of building Workflows 
    - fan-out / fan-in API
- **Runtime**
  - MCP dynamic loading and refreshing #458 
  - Parallel tool call execution #855 
  - Agent Harness (governance layer) 
    - `WorkflowRunPolicy`: max events / max actions / max tokens / timeout
    - Support safety guardrails and hooks
- **CI & Infra**
  - Support automated validation of the quick start example via GitHub Actions 
#642 
- **SQL Integration**
  - Support invoking Agents from SQL jobs 
  - Support Invoking Agents from CDC YAML 
  - Agent-specific SQL syntax (Table / View as VectorStore / KeyedContext), 
Inject other Tables / Views as context into the Agent
  - Support SQL Function / Operator as Action, reusing existing SQL / DataFrame 
operator capabilities within Flink Agents 

GitHub link: https://github.com/apache/flink-agents/discussions/862

----
This is an automatically sent email for [email protected].
To unsubscribe, please send an email to: [email protected]

Reply via email to