agentic-in/elephant-agent

Elephant Agent is a personal AI assistant designed to remember things about you in a meaningful way over time. The problem it solves: most AI assistants have no persistent memory, so every conversation starts fresh, requiring you to re-explain your projects, your preferences, the people in your life, and the context behind your questions. Even AI tools with longer memory often just try to store everything, which becomes noise rather than understanding. Elephant Agent takes a different approach: instead of keeping transcripts of every conversation, it builds and maintains a structured Personal Model, a correctable understanding of four aspects of who you are. Identity captures your values, decision style, and stable preferences. World captures your projects, tools, and the people around you. Pulse captures your current focus, pressures, and temporary priorities. Journey captures lessons, past decisions, and growth patterns. This model evolves slowly through conversation, through occasional questions the agent asks when a gap would affect how it helps you, and through a background reflection process that runs after sessions close. You stay in control through a dashboard where you can see the claims the agent has made about you, what evidence each claim is based on, and correct or delete anything that is wrong. The level of curiosity, meaning how often the agent asks clarifying questions, is configurable between quiet, balanced, and active. Each named assistant instance is called an elephant, and you can create more than one for different projects or contexts, together forming what the project calls a herd. You would use this if you want an AI assistant that genuinely accumulates useful personal context across months of use rather than starting over each time. It installs through a shell script and is written in Python.