Repository navigation
The system supports a complete voice loop all running locally with no required external APIs.
- This release prioritizes correctness, transparency, and debuggability over optimization or polish. Memory is session-scoped, and there is no long-term learning beyond the indexed document store
- Known limitations: audio I/O behavior may vary by device, vector storage is optimized for small-to-medium document sets, and the agent does not yet perform multi-step tool reasoning or background tasks.
- UI under construction and meant to be minimal
This release establishes the foundation for future work on persistent memory, improved audio mixing, agent tooling, and multi-modal inputs.