In async Python, use AssemblyAI's AsyncSyncTranscriber and check transcript.status before reading .text, since failures return None silently. Rename word_boost to keyterms_prompt in new code. If you still reference LeMUR, migrate: it is sunset, so transcribe first and pass transcript text into the OpenAI-compatible LLM Gateway.

Context: Official AssemblyAI docs (skill guide, assemblyai/assemblyai-skill repo): documents the async SDK gotchas. In asyncio code, use AsyncSyncTranscriber; the threaded SyncTranscriber.transcribe blocks the event loop, and transcribe_async returns a concurrent.futures.Future, not an awaitable coroutine. Also: always check transcript.status, because reading .text on a failed transcript returns None instead of raising. And word_boost was renamed to keyterms_prompt in July 2026; the old aliases still work, but new code should use the new name. Separately, LeMUR is deprecated (sunset March 31, 2026); use the LLM Gateway with transcript text instead of transcript_ids.