Mem9 adds long-term memory for agents

This is the kind of infrastructure layer agent builders keep reinventing badly, so a dedicated memory service feels overdue. If Mem9 holds up in real workflows, it could turn “remembering” from a brittle app feature into shared agent plumbing.

• –Hybrid keyword plus vector search is the right default for memory: fast recall without forcing teams into a pure embedding workflow.
• –Shared tenant memory means multiple agents can benefit from the same learned context instead of each hoarding its own local notes.
• –Apache-2.0 and self-hosting lower the trust barrier for teams handling sensitive project history or customer context.
• –The real win is not chat memory; it is consistent agent behavior across machines, sessions, and tools.