Maya
One brain, more than one interface. A personal assistant built around a shared protocol and actions you can review.
- Discipline
- Software Development
- 01Brain
Python server: orchestration, voice and tool routing.
- 02Faces
Electron and React Native clients sharing one WebSocket protocol.
- 03Tools
Prepare an action, show its preview, receive approval, execute.
The problem to structure.
A personal assistant can quickly become a tightly coupled collection of prompts, interface code and tool calls. Maya explores a different architecture: separate the reasoning server, the interfaces and the tools so that another interface does not require another assistant.
How the pieces connect.
A Python server handles orchestration, voice and tool routing. Electron and React Native clients exchange state through a WebSocket protocol. Tools declare the risk of each operation and separate preparation from execution; email and calendar changes can be presented for review before an approval is sent back. This is a personal development project, not a public SaaS product.
Below the interface.
01One protocol, multiple clients
The clients display state, accept input and return approval decisions. Orchestration stays on the server, which keeps the interface from owning the assistant’s logic.
02Review the action itself
Preparation produces the action that the interface displays for review. Execution is a separate step, allowing the person to inspect the proposed email or calendar entry first.
A personal assistant architecture with a shared Python server, desktop and mobile clients, and a preview-and-approval path for tool actions.
A similar challenge?
An idea, a bottleneck, a system that has outgrown its tools. Tell us where you are.
Tell us about it