Hi everyone, I’m pretty new to the whole AI assistant plugin scene and I’m feeling a bit lost. 😅
I’ve been trying to set up Tabnine in IntelliJ IDEA for my team’s project (we’re a small SaaS shop working on a CRM with some marketing automation bits). The suggestions are honestly amazing when they work, but I’ve hit a wall with memory usage. After a few hours, my IDE gets so sluggish it’s almost unusable, and I see Java process memory climbing steadily. I’m worried about rolling this out to the rest of the team if it’s going to eat all our RAM.
I’m on macOS Ventura, using IntelliJ IDEA 2023.2. My plugin list isn’t huge: the usual Kotlin, GitToolBox, and a couple for database navigation. Is Tabnine known to be a memory hog, or is it likely conflicting with something? I’ve seen some older forum posts about similar issues, but nothing recent.
Has anyone here successfully run Tabnine in a real production IntelliJ environment without these problems? If so, did you have to tweak any specific settings, like the heap size for the plugin itself or limit its scope? I’d love to hear what your setup looks like—maybe I’m just configuring it wrong. Any advice on how to narrow this down would be a lifesaver.
Oh yeah, memory issues with Tabnine are practically a rite of passage. It's brilliant but thirsty.
The trick that worked for my team was restricting its indexing scope. By default, it wants to eat your whole project and every dependency. In the plugin settings, try adding `node_modules`, `build`, `.gradle`, and any other generated directories to the excluded paths. Made a night-and-day difference for us.
Also, check if you're on the "Deep Completions" tier. That's the real RAM killer. Sometimes the standard model is plenty fast and keeps your IDE from feeling like it's running in molasses.
The "Deep Completions" tier is the real gotcha. It sounds fancy, but you're basically paying more to have a larger model idle in RAM, waiting for you to type. I'd bet most teams never see a tangible ROI on that upgrade over the standard suggestions.
All this indexing advice feels like we're back to tuning Java heap size for an app server. We're just accepting that the tool consumes resources unpredictably, and the burden is on us to constrain it.
-- cost first