Getting started · yeti docs

yeti runs your own AI agents on computers you own. Each agent is a yeti: one yeti, three yeti, your yeti. They run on cubes, which are this computer, other machines you own, or a cloud machine in your own AWS account. They think with Claude or OpenAI models, using your own API keys. There is no yeti server in the middle.

Install

Get the app from the download section and open it like any other app.

What happens on first launch

The launch screen stays up while yeti sets up this computer's cube. This happens once:

yeti creates a private cluster with its own certificate authority. Nothing leaves this computer.
It installs the cube as a background service, so your yeti keep working after you close the window: a launchd agent on macOS, a systemd user unit on Linux, or a Windows service (Windows asks for permission once).
The cube listens on this computer only (127.0.0.1, ports 7443 and 7444). The app talks to it over mutual TLS, the same way any other machine would.

Later launches just connect. If the app can't reach the cube, pages say "Can't reach your cube" and offer Try again.

Find your way around

In the sidebar	What it's for
New chat	Start a conversation with your default yeti
Agents	Each yeti's persona, model and preferred cube, plus your model keys
Studio	How each yeti looks
Skills & tools	Accounts your yeti can use: Gmail and Google Calendar
Cubes	The machines your yeti run on, your trusted devices, and cloud accounts
Your yeti	Your roster. Pick one to chat with it

Meet Tilly

Every new cluster starts with one yeti: Tilly, the "Everyday helper". Tilly is Warm and Bubbly, keeps replies Short & sweet, and always checks with you before sending messages, deleting things or spending money. Tilly is your default yeti, so new chats start there.

You can change all of it. See Your yeti.

Add a model key

Your yeti think with Claude or OpenAI models using your own API key.

Open Agents and scroll to Model access.
Paste your Anthropic key (for Claude) or your OpenAI key, then choose Save key.
The row now shows set and when it was last updated. The key itself is never shown again, not even to you.

The key is sealed with your cluster's own encryption key and stored on your cubes. Models and keys has the details.

Make it yours

Persona. On Agents, pick a yeti and set its name, role, personality, tone of voice, custom instructions and the things it always checks with you about first.
Look. In Studio, choose horns, face, accessories, fur and tile, then Save look. The new look shows up everywhere.
More yeti. Choose New agent, give it a name and a role, and choose Create yeti.

Chat

Choose New chat at the top of the sidebar, or pick a yeti under Your yeti. Your yeti reads its custom instructions before every chat, answers in the tone you picked, and stops to ask you before anything on its "Always check with me before…" list.

Next steps

Cubes: add another machine you own, so your yeti have somewhere to go when this one sleeps.
Cloud cubes: launch a cube in your own AWS account that keeps running with this computer closed.
Skills & tools: connect Gmail and Google Calendar with your own Google OAuth client.
Security: exactly what is stored where, who yeti talks to, and the known limits.