I got tired of YouTube: half my feed is clickbait and rehashes.
So I built Crux. It marks what's worth watching right in your feed and gives you any video's gist in 5 timestamped points.
Beta, 20 tester spots. Reply "in" and I'll DM you a link.
#buildinpublic#ai
Let's connect π€
AI dev, building tools for agents:
β’ sushiAI β open-source macOS workspace for Claude Code & Codex
β’ agents that run tasks on their own in sandboxes
β’ Crux β filters empty AI content on YouTube
Looking for people who build.
@RobIW_dev@neoromantic Agreed, though with your own solution it's easy to drown and keep building it forever. But yes, better to sketch something for yourself than to depend on someone else's. These days it's easy to build and deploy on a VPS. And forks are always an option.
@PawelJLisowski Yes, since I juggle a lot of projects at work, sushiAI is my main focus. I built it for myself, to work comfortably and efficiently across many projects at once. What about you?
@NovaOrbiting AI tools + devs here π
Working on:
β’ sushiAI β open-source macOS workspace for Claude Code & Codex
β’ sushiAI Tasks (in the works) β agents that run tasks as autonomously as possible, in sandboxes
β’ Crux β filters junk content on YouTube
@aetherfrontier AI tools + devs here π
Working on:
β’ sushiAI β open-source macOS workspace for Claude Code & Codex
β’ sushiAI Tasks (in the works) β agents that run tasks as autonomously as possible, in sandboxes
β’ Crux β filters junk content on YouTube
@tazqen Yes, once I'm confident it's really stable and useful, I plan to fold it into sushiAI. Enjoyed this thread too! Feel free to DM me anytime, always happy to chat tech π
Mine so far: launched yesterday with 11 followers. One "let's connect" post brought 13 followers and a few real conversations. Product posts? Crickets so far π
Love that anyone can now build their own agent workspace or a tool that actually saves people time.
But how do you get it in front of people? What would you do starting with zero audience?
Share your stories π
@tazqen Each sandbox has a CPU limit and the engine caps how many tasks run at once. Heavy checks have their own cap, so a couple run side by side and the rest queue.
@Jazzen_Chen Totally agree. I don't want to be pulled in for every little thing, so inside the sandbox the agent has full permissions. It only asks me about critical stuff: plan approval, PR sign-off and so on. The simple questions the engine answers on its own.
@tazqen Pulled real numbers from last week, 320 engine checks: median 38s. Focused checks finish in seconds to a couple of minutes. Full pytest suites on Python backends take 15-35 min, worst one ~50. That's exactly why the full run happens once, at the end.
@di_decoded Thanks, followed back! Interesting angle: I built it as a viewer, but for creators it's a quick way to scan what competitors already covered. If you want to test it, just let me know!