Built for SRE / DevOps: human-in-the-loop by design

[Cznorth]

WinkTerm is intentionally not an autonomous agent that runs destructive commands while you are away.

Design principles:

1. Shared PTY — same session you see is what the agent reads
2. Suggest, do not execute — write_command puts text on your input line; Enter is yours
3. SSH first-class — remote tabs, SFTP, one-shot agent API for automation you trigger
4. Self-hosted — your keys, your network, BYO API key

If you have tried other AI terminal products and bounced off because they felt unsafe or fiddly, what would need to change?

https://github.com/Cznorth/winkterm · MIT