Executes tasks.
Claude Code, Cursor, Devin, and Codex-style agents make execution cheaper and faster.
Elephant Agent
Personal AI that starts from you.
Mother grows to understand who you are, what surrounds you, what is alive now, and what your path has taught you. Then she helps shape the next Paths across work, health, learning, relationships, and the parts of life you are trying to move.
Starts from youPaths after understandingJudgment stays close
Personal agent levels
Where Elephant Agent sits.
The four levels stay central, but L4 is an internal position more than a homepage slogan: L1 executes, L2 carries context, L3 improves procedures, and L4 keeps the person's judgment and growth inside the loop.
Carries context.
OpenClaw publicly emphasizes local agents, persistent memory, full system access, skills, plugins, and integrations.
Improves procedures.
Hermes Agent publicly positions itself around a self-improving learning loop, skill creation, recall, and user modeling.
Grows with the person.
Mother understands the person, shapes Paths, and keeps judgment, evidence, questions, and learning close to the person.
From understanding to Paths
Not just projects. Paths for life.
Once Mother understands enough, she can help turn work, health, learning, habits, relationships, research, and recovery into living Paths with visible Steps and Checkpoints.
Work paths
Repositories, research, writing, launches, and maintenance can still look like projects when that is useful.
Life paths
Fitness, weight loss, habits, learning, relationships, recovery, and personal resets should not be forced into issue language.
Growth paths
Mother can notice lessons, risks, pressure patterns, and stale directions, then help shape the next move.
The coordinating elephant. She starts from the Personal Model, proposes Paths, chooses the next Steps, and knows when to ask.
Any ongoing direction you want to move: a codebase, fitness plan, habit reset, research arc, relationship repair, or life transition.
A concrete action inside a Path. It replaces issue and ticket language when the user's life is broader than work.
The visible board for a Path: later, next, moving, checking, done, stuck, or dropped.
The moment where human judgment returns. Mother can ask first or keep moving inside a trusted boundary.
The completion receipt for a Step: what happened, why, how, what knowledge mattered, and what should be learned.
The baby elephants that can help a Path. Babies do bounded work; Mother keeps the whole shape coherent.
Mother loop
Chat stays simple. The shape stays visible.
You can talk to Mother in chat, answer Checkpoints when judgment matters, or drag Steps directly when Flow is the clearer surface.
Understands
Mother starts from Identity, World, Pulse, and Journey instead of a blank prompt or a task queue.
Shapes Paths
She proposes living Paths and breaks them into Steps only after the direction is clear enough.
Brings the Herd
Baby elephants can take bounded Steps with explicit skills, runtime posture, and visible event trails.
Returns to judgment
Checkpoints and Learning Summaries bring decisions and understanding back to the human.
Learning loop
Do more without thinking less.
A baby does not only mark work as done. It returns a Learning Summary, then asks for the human understanding needed to close the loop.
The baby explains the work
Every completed Step returns what happened, why that approach was chosen, how it was done, and what context mattered.
The human checks understanding
The loop asks for an Understanding Check, so the person can say: I understand this Step before it disappears into Done.
Mother learns what should carry forward
Useful lessons become candidates for Journey, Path history, baby skills, and future Mother planning.
Trust modes
Two simple postures.
The product should feel clear at the chat box: Mother can ask first, or keep moving inside a trusted boundary.
Mother plans, then asks.
She can draft Paths, Steps, and Herd assignments, but waits before applying important changes or taking external action.
Mother keeps moving within your boundaries.
She can create Steps, move Flow state, and assign babies inside trusted limits while risky choices still become Checkpoints.
Product surfaces
Mother in chat. Paths in view.
The macOS app is the default experience. CLI and Dashboard keep Linux, cloud, SSH, and terminal-first workflows supported without turning runtime controls into the product center.
Install
Start where your path is.
Use the macOS app for the full desktop workspace. Use CLI plus Dashboard for Linux, cloud, SSH, and terminal-first setups.
RecommendedmacOS desktop app
Wake, Paths, Personal Model, Herd, skills, messaging, calendar, usage, and settings in one native workspace.
Download macOS appLinux / CloudCLI + Dashboardcurl -fsSL https://elephant.agentic-in.ai/install.sh | bash
Use elephant dashboard --no-open on remote machines.
Open source
Personal AI should be inspectable.
Elephant Agent keeps the personal layer visible: claims, questions, evidence, Paths, Steps, Checkpoints, skills, baby elephants, local agents, and semantic recall.