+ 1bKjuR^RIHt^RIHt^RIHt^RIHtHtRRlt RRlt R #) )LLMGenerationError)reply_system_prompt) get_provider) ChatMessageSuggestResponseRequestc\V^8dQhR\\,R\\,/#)rhistoryreturn)listrdict)formats":C:\xampp\htdocs\meridian\meridian-ai\app\services\reply.py __annotate__rs#d;&7DJcf.pVFpVP'dRVP RVP 2M VPpV'dEVR,R,VP8Xd&VR,R;;,RV,, uu&KVPRVPRV/4K V#)aCollapse consecutive same-role turns into one message. OpenAI tolerates back-to-back "user" messages in the list; Anthropic's Messages API rejects them outright (it requires strict user/assistant alternation). Since conversation_history is built from an external source (an inbox thread) rather than turn-by-turn by this service, it can't be assumed to already alternate - merging here keeps every provider implementation equally usable instead of only working by accident with whichever one tolerates the input. [z] rolecontent ) timestamprrappend)r mergedmessagers& r_to_alternating_messagesrsF@G@Q@Q@QAg''(7??*;r.s+),9( Tr