Yumii

The open-source AI companion that lives on your desktop. Voice-first, private, remembers you — and asks before it acts. Runs on a normal CPU.

Install & Talk

One command on Windows. One API key. Tap the orb and start talking.

Using Yumii

Voice conversation, six personalities, the dashboard, and memory.

Tools & Permissions

Web search, Gmail, and more — Yumii always asks before acting.

Providers

Swap the mind, ears, and voice: Ollama, Groq, OpenAI, Anthropic, Whisper, Kokoro.

What is Yumii?

Yumii is an open-source AI companion that runs on your own machine. She listens through your microphone, thinks with the LLM provider you choose, answers out loud in a local voice, and shows up as a small floating orb — always there, never in the way.

Three promises:

  • Voice-first. Real conversation, not a chat box. Interrupt her mid-sentence and she stops and listens.
  • Private by design. Conversations, memory, and API keys are local files on your disk. The only network traffic goes to the providers you configure.
  • She asks before she acts. Every tool that touches the outside world goes through an Approve / Deny prompt first.

What she can do today

Talk in real timeBarge-in interruption, ~1s reply latency on an ordinary CPU
Six personalitiescaring · tsundere · genki · kuudere · yandere · dandere — switch by asking
Remember youLearns facts, searches every past conversation, knows when things happened
Do thingsWeb search built in; Gmail, Calendar, Notion via Composio — all permission-gated
Stay inspectableA dashboard shows every conversation, every fact, every connected tool
Speak locallyKokoro-82M synthesizes her voice on your CPU — free, offline, no key
Experimental preview

Yumii is pre-1.0 and moving fast. The core experience works end-to-end; expect rough edges. Found one? Open an issue.

Where to go next