Two weeks of dogfooding Engram, Weaviate's memory product, in daily Claude Code sessions. This surfaced where a dedicated memory product adds value, and the specific mechanics that prevent integration with coding assistants from working well.

Weaviate's blog offers insights into artificial intelligence, machine learning, and natural language processing. Readers can explore articles on topics such as deep learning, neural networks, and knowledge graph technology. Additionally, they can learn about use cases for AI applications, industry trends, and ethical considerations in AI development.

Weaviate

A product manager at Weaviate spent two weeks dogfooding Engram, Weaviate's memory product for AI coding assistants, integrated with Claude Code via an MCP server. The key finding: Claude ignored Engram entirely when given discretion, defaulting to its built-in MEMORY.md file. The post details why this happened, how structured memory topics and deterministic retrieval hooks solve it, where Engram genuinely added value (decision archaeology, preventing hallucinations), and where it failed (forward-looking planning tasks). It also covers session overhead issues, the case for infrastructure-level memory injection rather than LLM-decided recall, and open problems around team collaboration scope and cold-start bootstrapping.

Oh Memories, Where'd You Go