elephant-agent 1.0.0.dev20260526090952

Elephant Agent CLI-first persistent agent runtime.

Elephant Agent

Agency-first personal AI.
A correctable Personal Model for Identity, World, Pulse, and Journey.

Website · Blog · Paper

What It Is

Elephant Agent is agency-first personal AI.

It keeps one durable object at the center: a correctable, evidence-backed Personal Model of what should shape future help. Tools, skills, memory, messaging, cron jobs, and background learning orbit that model instead of replacing it.

The product position is L4 personal AI: agents may do more work, but the person should not lose judgment, continuity, or growth in the process.

Personal Agent Levels

Level	Core question	Public examples	Elephant stance
L1 — Do work	Can AI execute tasks?	Claude Code, Cursor, Devin, Codex-style agents.	Useful capability, not the center.
L2 — Carry context	Can the agent stop starting over?	OpenClaw publicly emphasizes local agents, persistent memory, full system access, skills, plugins, and integrations.	Required foundation; memory supports the Personal Model.
L3 — Improve procedures	Can the agent evolve skills and workflows?	Hermes Agent publicly positions itself around a self-improving learning loop, skill creation, recall, and user modeling.	Important downstream layer; skills stay visible and governed.
L4 — Grow the person	Can personal AI return judgment, evidence, questions, and reflection to the user?	Elephant Agent's product position.	The Personal Model should help the person do more without thinking less.

Personal Model

• Identity — who you are, your values, boundaries, decision style, and stable preferences.
• World — the people, projects, tools, places, and relationships around you.
• Pulse — what is alive right now: focus, pressure, constraints, energy, and priorities.
• Journey — what your path has taught: lessons, failures, recovery patterns, and long-running growth.

The goal is not to remember everything. The goal is to understand what matters, show why it matters, and let you change it.

What It Does

Capability	What it means
Correctable claims	Remember, correct, forget, dispute, and inspect why a claim exists.
Evidence-backed recall	Conversation search and embeddings provide support, but retrieved text does not become truth by itself.
User-paced curiosity	Quiet, balanced, or active questions appear only when an answer would improve future help. Silence wins.
Background reflect	Episode-close, diary, dream, and manual jobs turn lived Steps into governed Personal Model updates.
Visible skills and tools	Skills, tools, models, messaging, and cron jobs remain inspectable capabilities around the Personal Model.

If Elephant Agent cannot find reliable Personal Model support, no_match is a feature, not a failure.

Quickstart

Install Elephant Agent, create your first named elephant, then come back through wake whenever you want to continue.

Install

curl -fsSL https://elephant.agentic-in.ai/install.sh | bash

First run

elephant init        # choose identity, provider, and curiosity effort
elephant herd new    # create another named elephant when you need one
elephant wake        # enter the chat TUI
elephant dashboard   # open You, Questions, and Evidence

Read More

README and the homepage stay product-first. The deeper system story lives here:

• Read the docs →
• Read the paper →
• Read the blog →

Contributors

Agentic Intelligence Lab