Bi-temporal AI Agent Memory (Memharness-style)

AI agents need to understand context across different time periods and maintain consistent state through multiple interactions. A lightweight, file-based bi-temporal memory system lets agents track how information changed over time without complex databases. Target users are AI engineers building stateful agent systems.

Agent memory is a genuinely active problem right now — LangChain, LlamaIndex, and the wave of agent frameworks have all exposed how poorly current context management handles state that changes over time, and bi-temporal data modeling (tracking both valid time and transaction time) is a well-understood database concept that almost no agent tooling has adopted yet. No clear incumbent owns this specific niche, though Zep comes closest as a managed memory layer for agents, and MemGPT handles some long-term memory problems differently. The unknown revenue band is a real concern — AI engineers will adopt a devtool like this as open-source infrastructure and resist paying unless there's a hosted or managed layer with clear ROI, which means monetization requires a deliberate wedge beyond the library itself. The biggest risk is timing out: the major agent frameworks are actively building native memory solutions, and if LangGraph or a similar dominant abstraction ships something good enough in the next 6–12 months, the addressable market for a standalone tool collapses fast.