wenjin272 commented on code in PR #1362:
URL: https://github.com/apache/flink-agents/pull/1362#discussion_r4510015388
##########
docs/content/docs/development/calling-llms/model_routing.md:
##########
@@ -424,3 +425,58 @@ for the persistence and replay guarantees.
- {{< gh_link
file="examples/src/main/java/org/apache/flink/agents/examples/openai/OpenAiDecisionRoutingExample.java"
name="`OpenAiDecisionRoutingExample`" >}}: the same two OpenAI models routed
by a decision model (the built-in Python OpenAI decision model, used through
the cross-language API), with an agent that reports the selected model,
decision source and score.
All of them register their resources on the execution environment. The first
three use the {{< gh_link
file="examples/src/main/java/org/apache/flink/agents/examples/agents/ModelRoutingAgent.java"
name="`ModelRoutingAgent`" >}} to send requests and handle responses.
+
+
+## Python routers
Review Comment:
Could we integrate the Python documentation into the existing sections
instead of adding a separate “Python routers” section? Registration,
invocation, and custom executor examples could use Java/Python tabs, while
shared behavior such as fallback, tool-call handling, and recovery could be
described once. Python-specific limitations and regex differences could stay in
the relevant sections.
It would also help to keep the documentation focused on user-facing
behavior. For example, the internal persistence keys (`route:<router>` and
`chat:<router>:<candidate>`) seem unnecessary; explaining that completed
routing decisions and model calls are reused during recovery should be
sufficient. For regex compatibility, we can state the constraints without
describing the internal Java validation path.
Please also update the existing Java-only wording where it no longer applies.
--
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]
