Show HN: Sentient, Your Coding Agent Shouldn't Be a Black Box

after working quite rigorously for a few months with claude code, I found out that I was blindly approving permission requests for the sake of progressing my projects, this led me to realize that coding agents generate hundreds of commands underneath the covers which is necessary, but quite scary if you ask me. you've probably seen news recently about agents deleting databases or environments in production by accident. so i wanted to create something to gain visibility into what's actually happening and potentially block things before they occur.

coding agents often log everything that they're doing, but even still, i want the ground level truth of what's happening. that's why i created sentient, written using claude code. it's open source (mit license), runs completely locally, no cloud syncing, read only, and watches actions straight from the kernel. all of this is summarized with a hopefully simple web ui.

i run claude code on linux, so kernel tracking relies on tetragon, but, at the same time, i am trying to get approval to use security endpoint on mac, and will eventually roll out on windows. if you want to run sentient today against just the agent logs without kernel logs, that's also available now, you just need golang and sqlite3 to build. and today, i'm only supporting claude code, but will eventually get to codex and the like. please have a try, ping me if you have any questions or comments: hello@nosaka.ai

https://github.com/nosaka4i/sentient

https://nosaka.ai/blog/sentient