Astra's context handling is absolutely trash😭

A user building agents on GPT finds Astra highly capable but criticizes its memory handling. While Astra performs well mid-session, agents begin losing context after 15-20 turns, even within the same session. The user resorted to integrating third-party memory SDKs for entity resolution and typed extraction, arguing that such additional infrastructure shouldn't be necessary for Astra-based agents to retain user data.

This user report is one of the first to detail specific context-loss thresholds for Astra, noting issues after 15-20 turns within a single session, unlike general complaints about memory.