Remembra launches open-source memory for AI agents

Agent memory is quickly becoming core infrastructure, and Remembra is betting that developers want a full stack answer instead of another thin vector wrapper.

• –The strongest idea here is the split between memory and classic RAG: Remembra focuses on evolving preferences, decisions, and relationships rather than just document retrieval
• –MCP support is the real distribution wedge, because a memory layer becomes much more useful when Claude Code, Cursor, and other clients can plug in without custom glue code
• –The self-hosted story matters for this category since memory can contain sensitive user history, coding context, and long-lived behavioral data
• –Its positioning is aggressive: the docs explicitly compare Remembra against Mem0, Zep, and LangMem on graph features, self-hosting, and pricing
• –Early benchmark and feature claims are ambitious, so the next question for developers will be operational reliability and how well the memory quality holds up in real agent workflows