The DingTalk AI response release: authorized messages can drive bound roles and receive safe, streaming Markdown card replies.
Added
- Streaming DingTalk AI cards. Model text deltas are throttled and serialized into one card before the authoritative final response is applied.
- Immediate processing feedback. DingTalk users see a processing response while longer agent and tool workflows run.
Changed
- Authorized sender intent handling. Messages from allowlisted DingTalk senders now drive the bound role as trusted intent; non-allowlisted senders remain fail-closed.
- Markdown fallback. If an AI card cannot be created, processing and final responses use DingTalk Markdown messages.
Fixed
- Internal reasoning and raw tool events are excluded from streamed card content.
- Streaming output is scrubbed across delta boundaries before being sent to DingTalk.