weiqingy commented on issue #912: URL: https://github.com/apache/flink-agents/issues/912#issuecomment-5708185187
Thanks, that settles all three. On the round trip: the implementation takes the simpler path, an extra tool-free call only when the effective strategy resolves to `NATIVE`, with `PROMPT` keeping today's behavior unchanged. On the coexistence nuance, that distinction is already expressible here, which I had not expected. Gemini is the concrete case that cannot coexist: it drops a native schema whenever tools are bound, which is the reason the finalization call is issued with no tools at all. The feasibility query each connection now answers is asked about a specific request, tools included, so a connection that can carry a schema alongside tools is able to say so. The loop could then pass the schema and parse the first tool-free response, exactly as you describe. I would still rather land the uniform finalization call first and add that optimization once there is a caller to measure it against, since the measurement is what tells us which providers actually benefit. On the parse failure: agreed, and that is what the current code does, so nothing changes. The retry repeats the chat attempt rather than rerunning completed tool-call rounds. On the directive: I will use your wording verbatim in both languages. -- 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]
