Skip to main content

Overview

Most AI starts from the task in front of it. Elephant Agent starts from the person it is helping.

The bet is simple: personal AI should not win by keeping the longest transcript or collecting the largest tool shelf. It should grow a correctable understanding of the paths, people, risks, rhythms, and decisions that keep shaping your work. That understanding should stay durable between sessions, remain visible enough to inspect, and get better through gentle curiosity at the pace you choose.

That is why Elephant Agent is personal-model-first AI. Memory is the beginning, not the product. The center is a Personal Model: what Elephant Agent currently understands about your Identity, World, Pulse, and Journey, with source evidence and open questions close enough to correct.

Start here

If you only have a few minutes, install Elephant Agent, run elephant init, then return through elephant wake. The same elephant should be able to pick up the right thread because it is learning the path, not just replaying the transcript.

Choose your path

If you want to...Read this firstWhat you will understand
Start using Elephant AgentQuickstartThe shortest supported path from install to wake.
Understand the thesisWhy Elephant AgentWhy elephant-inspired memory means judgment from the right experiences, not more storage.
Learn the daily surfacesCLI / Chat TUIHow elephant, init, wake, and slash commands fit together.
Inspect what it knowsDashboardHow You, Diary, Herd, Curiosity, Skills, Tools, Messaging, and Jobs map to the product model.
Extend what it can doSkills and ToolsHow visible capabilities orbit the Personal Model.
Understand learningProactive curiosityHow Elephant Agent asks gently, learns in the background, and stays correctable.

The core idea

Product betWhat it meansWhere to go deeper
Personal Model firstElephant Agent keeps an explicit, inspectable model of what it understands, rather than treating every retrieved memory as truth.Personal Model first
Curious by designIt does not wait for you to explain everything forever. It may ask when a missing or stale answer would change future help.Proactive curiosity
Correctable understandingClaims can be remembered, corrected, forgotten, disputed, and traced back to evidence.Correctable understanding
Continuity across surfacesCLI, Chat TUI, Dashboard, messaging, jobs, skills, tools, and recall all orbit the same local understanding system.Continuity

The docs map

How Elephant Agent is organized

AreaProduct questionMain docs
Personal ModelWhat does Elephant Agent understand about the person?Personal Model first, Memory
Memory architectureWhat becomes durable truth, what stays evidence, and what is recalled only for the current turn?System model, Memory, Embeddings
Daily surfacesWhere do I talk to it, inspect it, and correct it?CLI / Chat TUI, Dashboard
Visible capabilitiesWhat can it reach for without becoming a feature shelf?Skills, Tools, Messaging
Learning loopsHow does understanding deepen over time?Proactive curiosity, Background learning, Correctable understanding

Design source of truth

These docs are the public operator guide. The deeper repository architecture truth lives in the system-design docs and the paper: