Hi, I have been using AntiG for about a month and since Opus 4.6 has been added I’ve notice an incredible increase in the Language_server_windows_x64.exe of VRAM. even with a single agent it climbs to over 75%+ of total memory within a few minutes and have to continuously restart AntiG every 20 mins. I have 16GB and litte to no other apps open beyond a browser for testing. I was about to go out and buy a 64GB PC but I’m concerned this would just continue even with that amount of RAM. Has anyone else come across this ?
Same here, my vps crash (24gb ram).
This memory leak was already there, i thought it was solved in one of the last updates since i havent’ had this problem anymore, but today we’re back to this…
Hi Herian, I see you are on linux and have a history in another thread.
I actually struggled to try and solve the issue and in the end, actually asked AntiG to diagnose its own settings and OS settings. It found two issues, the .antigravityignore was missing in a project so it was indexing node modules etc. this reduce it a bit..
The second was that a conversation was too long and the indexer process was just climbing and climbing. It was probably 2 days old but there was a LOT of content in the output
I’ve just taken action to limit the context and conversation in each of the agents and this has solved the issue.
Another ‘gotme’ is that on Windows, when you launch AntiG, it launches the last active agent which launches a language_server_x64.exe, then you select the Agent manager to load. this actually kicks off a second language_server_x64.exe which kick off an indexing and consumer more memory too. I have stopped using the agent manager while working on projects until I get a more powerful machine (looking..).
I know its a different experience on Linux but the rule seems to be to do good conversation hygiene regularly and what I am doing is building a .md file to track and inform the next fresh conversation of where I am along with a project doc, and this resets the counter for the memory. I wasn’t getting past 20 mins and now the agent is happy running at 25% of resources with series workload all day today.
If you are reading this and having the same issue, ask the Agent itself to self diagnose with your issues. It saved me 20 hours and it resolved itself.
I’m not a trad dev background, but heavy all other aspects of Tech stack but its been a great learning lesson too.
1 Like