Yumii
Open-Source AI Companion
- Built a Windows desktop voice companion (v0.13, 200+ commits) on a real-time loop: Silero VAD and Whisper speech-to-text, a LangGraph agent, and streamed text-to-speech that users can interrupt mid-reply.
- Gave the agent permission-gated tools (web search, plus Gmail, Google Calendar and Notion via Composio): every action waits for the user’s approval before it runs.
- Designed long-term memory in local SQLite: full-text search over every past conversation, session summaries, and a fact store the agent writes and corrects itself.
- Private by design, with no account or telemetry and a fully offline mode (local Whisper, Kokoro, Ollama); installs with one PowerShell command.