I made a local AI agent named Piwi

Hi everyone,

I now have a fully functional version of Piwi, a local AI agent for Windows
that executes real tasks using scripts generated via the OpenAI API.

Current state:

  • Production-ready execution pipeline on Windows (WSL sandbox).
  • OpenAI used strictly for understanding + script generation.
  • Local execution with explicit I/O boundaries.
  • macOS and mobile variants planned (restricted execution).

Before deciding whether to make the project public, I’d appreciate feedback from
this community on:

  • risks or pitfalls of exposing such an agent to end users,
  • API usage patterns that scale safely in local-execution contexts,
  • lessons learned from similar tools.

I’m currently discussing with a few tech companies, so I’m carefully evaluating
what level of public access makes sense.

Any feedback or experience sharing would be very helpful.
Thanks.