Observational Memory

Most AI agents wake up with amnesia between sessions — and even during a long project, context can fall out of the active window as conversations grow (token limits are real). Observational Memory fixes that with a simple, reliable Observer → Reflector loop that keeps your agent’s recall sharp without any databases or external services.

• Observer: captures what happened today (decisions, tasks, context) into lightweight daily notes.
• Reflector: periodically reviews those notes and distills the durable bits into a curated MEMORY.md (preferences, project state, lessons learned).

In practice, this means your agent can quickly re-anchor itself: it searches memory first, pulls the exact snippet it needs, and keeps moving — even when the current chat history is too long to load or you start a fresh session. This setup has saved us more than once when we were deep in a build and needed to recover “what we decided last week” without re-litigating the whole thread.

Everything is markdown-first (human-readable, git-friendly), zero-dependency, and designed with privacy boundaries so long-term memory doesn’t leak into shared contexts.