Aether
Not a black box, a devbox
82 followers
Not a black box, a devbox
82 followers
Autonomous coding agents in a cloud devbox you can watch, steer, and verify, running Claude Code, Codex, or OpenCode on your own subscription.
Free Options
Launch Team / Built With
running on your own subscription instead of taking a margin on inference is the right call, most cloud agent products quietly become an inference reseller. practical question though - if I spin up 3-4 devboxes in parallel all hitting Claude Code on the same personal subscription, do they just compete for the same rate limit/quota pool and throttle each other, or is there any queuing so one greedy agent doesn't starve the others?
@galdayan currently they do all draw from the same pool, but it's unlikely that a greedy agent will starve others because agents stop running while they're waiting for additional input (CI failures, PR review, GitHub comments). To mitigate this you can provide an API key or enable extra usage in your Claude Code or ChatGPT subscription.
@pranav_sharan1 that makes sense, thanks for the detail. good to know it degrades gracefully instead of silently throttling. will try running 3 in parallel this week and see how it feels in practice
@galdayan should be able to handle a lot more than 3! Looking forward to hearing your feedback, please let me know if there's anything I can do to help you or if there are any questions I can answer🙂
honestly the ability to pause and rewind a session would be huge, like if an agent makes a bad code change halfway through you could just scrub back instead of restarting the whole thing. would save so much time on longer runs
@glerh561 This is a great idea! Currently the agent checkpoints with git commits, but recovering state in between commits is currently not supported. In my experience the commits are usually small enough that I'm able to revert unwanted changes easily, but I can definitely see the use for more granular checkpointing. Would support for this feature make you interested in using Aether?
finally got a chance to test this out and being able to watch the agent work in real time and step in when it goes sideways is genuinely useful, the verification step catches stuff i'd normally have to debug later.
Congratulations on the Launch this looks like a great product. I have personally run into the issue of running out of ram on the computer due to multiple agents running and slowing the whole system down. My mac hard reboot a once as well after it ran out of storage and the ram didn't have room on the drive to swap to.
Sweet!! What is the standard right now? Claude code and keep your computer open?
@bullmeza yup exactly! Somehow walking down the street with my laptop open while connected to my phone's hotspot doesn't feel like AGI
the fact that you can actually watch the agent work and step in when it goes off track is such a thoughtful design choice, makes it feel way more like collaborating than just hitting run