Claude Code, a tool used by the author for product management tasks, defaults to a built-in memory file, MEMORY.md, that holds about 200 lines of stable context. This approach, while efficient, leads to a lack of depth in recalling the reasoning and nuances behind decisions. To address this memory gap, the author integrated Engram, a memory product built on Weaviate’s vector search technology, intending to improve Claude’s contextual recall during sessions. Early experiments with Engram showed promise, making sessions feel more like ongoing conversations rather than starting from scratch each time. However, Claude often ignored Engram, favoring the existing MEMORY.md. The author realized Engram needed explicit triggers for activation, so they restructured their workflow to include specific moments for saving and recalling information. This involved creating categories for memory, like communication style and workflow preferences, and determining when to save context during sessions. The author found that using shorter, focused saves improved retrieval speed and efficiency. After two weeks of testing Engram, the author compared sessions with and without it. The results indicated that Engram significantly enhanced decision archaeology, allowing for faster access to reasoning and context during discussions. However, it fell short during planning sessions, where Claude didn’t utilize prior context as expected. The integration of Engram added some overhead, slowing down sessions by about 10%, but the potential to improve contextual memory made it a valuable tool despite these challenges.