Hey PH — Mars here, one of the builders behind MarsNMe. The frustration that started this: I was using Claude and Cursor daily for real work, and every session started from zero. I'd re-explain the same project context, the same constraints, a dozen times. It felt like working with a colleague who had perfect skills but permanent amnesia. System prompts help, but they're static. They don't capture what actually happened last Tuesday, or that I changed my mind about the architecture. So we built MarsNMe — a memory backend that any MCP-compatible agent can talk to. Two Supabase tables, a local HTTP gateway, and Jina v3 embeddings for semantic retrieval. pgvector keeps recall fast as memory grows. The multi-profile design came from a real need — we run several agents and didn't want them sharing memory. Schema-level separation, not just a filter. Apache 2.0, fully self-hosted. We're a small team from Macau. Happy to answer any questions!

MarsNMe — Not Just Context. Continuity.

Claude.md is for context. MarsNMe is for continuity.

Claude.md is for context. MarsNMe is for continuity.