v1.3.0 — Multimodal Tool Outputs

Features

• Support multimodal tool outputs, bump the Agents SDK to 0.4.1 by @bonk1t in #383
• Use agency name as a default trace name by @ArtemShatokhin in #404
• Group runs in a single trace by @ArtemShatokhin in #411
• get_response_stream returns run response wrapper by @bonk1t in #400. See "Streaming API update" below for more details.
• Expanded imports from Agents SDK into the framework by @ArtemShatokhin in #402

Improvements & Fixes

• Improved mention parsing in the terminal demo, update show_reasoning default value by @ArtemShatokhin in #403
• fix: streamline vector store ingestion waits by @bonk1t in #423
• fix: propagate run_config trace ids by @ArtemShatokhin in #413
• fix: unblock streaming final futures by @bonk1t in #414
• fix: set callerAgent for hosted tool system messages during agent→agent runs by @ArtemShatokhin in #420

Docs

• docs: update API reference to match Agency and Agent APIs by @ArtemShatokhin #416
• Revise README for clarity and updates by @bonk1t in #422
• Fixed formatting issue in validation doc by @ArtemShatokhin in #412

Streaming API update

• Agency.get_response_stream and Agent.get_response_stream now return a StreamingRunResponse immediately instead of an awaitable coroutine.
• Existing async for iteration keeps working, but calling await on these methods now raises TypeError because they are synchronous.

Updated usage:

stream = agent.get_response_stream(...)
async for event in stream:
...
final_result = await stream.wait_final_result()

Drop the leading await when obtaining the stream. Callers that previously wrote await agent.get_response_stream(...) must adopt the pattern above.

Full Changelog: v1.2.1...v1.3.0