@yangwenyu2

travel-lobster autonomously explores the web on custom intervals, builds persistent memory across sessions via a knowledge graph, and sends illustrated postcards with source links. agent personality adapts to time of day and integrates discoveries across trip history.

agent-memento proposes a tick-driven execution model for long-running llm agents, using physical markdown files as state and git for atomicity. the framework prioritizes preventing context overflow and oom through short-lived worker processes that read plans anew each cycle, but lacks concrete trigger phrases and assumes users can operationalize the architecture without implementation details.