Show HN: Coni – Trust-first Claude Cowork-style agent with permission prompts (github.com)

Coni is an open-source, local-first AI workmate for fast, trustworthy delivery — with verifiable outputs on disk.

If you’ve tried Claude Cowork (or tools like OpenWork), Coni is in the same “AI workmate” category, but optimized for trustworthy delivery:

- Permissioned execution (allow once / always / deny)

- Observable runs (see what happened, when, and why)

- Reviewable artifacts as real files you can diff and verify

Feedback welcome:

1) What workflow should we nail first (bugfix/refactor/PR review/report/etc.)?

2) What would make you trust an agent tool enough to use it daily?

Quick start (macOS):

- brew tap coni-ai/coni

- brew install --cask coni

Demos:

- Search latest news → generate a website: https://youtu.be/94HyUKrR1nA

- Book an event on calendar: https://youtu.be/nWBmBheGRqQ