This computer
yeti sets up a cube here on its own. It runs quietly in the background as a service.
launchd · systemd · Windows service
Private by default
Coming soon
Are you yeti for this?
Your own AI agents, on your own compute.
yeti runs AI agents on machines you own: this computer, your other computers, or a server in your own cloud. Your chats, memories and keys stay there.
macOS · Windows · Linux
01 · Your yeti
Agents with a personality. And manners.
Give each yeti a name, a job and a way of talking. Then decide what it has to check with you first.
- Personality. Pick up to three traits: calm and precise, or bubbly and a little dramatic.
- Tone. Short & sweet, just right, or tell me everything.
- Ask first. Sending, deleting, spending and sharing wait for your okay.
- Its own model. Each yeti can think with a different model.
- Studio. Design how your yeti looks: horns, face, accessories and fur.
Tilly
Everyday helper
Personality
- Warm
- Bubbly
- Calm
- Witty
- Sassy
- Precise
- Curious
- Dramatic
Tone of voice
Short & sweetJust rightTell me everything
Always check with me before
- Sending messages
- Deleting things
- Spending money
- Sharing files
02 · Cubes
Your yeti live on your machines.
Cubes are the computers your yeti run on. Start with this one, add more, and keep going while your laptop sleeps.
Your other machines
Add a desktop, a home server or a spare laptop with the cube command-line tool. With three or more cubes, work moves on if one goes away.
raft · mutual TLS 1.3
A cloud cube
A machine in your own AWS account that keeps your yeti working while this computer sleeps. Set its hours and it starts and stops on its own.
EC2 · Fargate · Linux, Windows, macOS
yeti never needs admin access to your cloud. It works through a role that can only touch what yeti made.
03 · Keys & models
Your keys. Your models.
Your yeti think with Claude or OpenAI, using API keys you bring. Your cubes call the provider directly, and the provider bills you directly.
Save a key once. The app can store it but has no way to read it back.
Bring-your-own and self-hosted models are on the way.
Your model key
- Providers
- Claude · OpenAI
- Stored as
- Encrypted cluster secret
- Cipher
- AES-256-GCM
- Lives on
- Your cubes, nowhere else
- Sent to
- The provider, only
- Shown again
- Never, not even to you
04 · Security
Security you can check.
No vague promises. Here's exactly what yeti does with your data and your keys.
Nothing phones home
No telemetry, no analytics, no yeti server in the middle. Watch the traffic yourself.
Mutual TLS, everywhere
Every connection is TLS 1.3 with a certificate on both ends, even app to cube on one computer. There's no plaintext listener.
Certificates, not passwords
Every device and cube brings its own certificate. Revoke one and it's out at the next handshake.
Cloud keys in your keychain
AWS access keys stay in your computer's keychain, never in your cluster.
Least privilege in your cloud
Cloud cubes work through a role fenced in by a permissions boundary. Root credentials are refused.
Secrets stay masked
Tokens, keys and passwords show up as •••• in the app's Logs view.