Zed. They've upped their game in the AI integration and so far it's the best one I've seen (external from work). Cursor and VSCode+Copilot always felt slow and janky, Zed is much less janky feels like pretty mature software, and I can just plug in my Gemini API key and use that for free/cheap instead of paying for the editor's own integration.
Overall Zed is super nice and opposite of janky, but still found a few of defaults were off and Python support still was missing in a few key ways for my daily workflow.
I'll second the zed recommendation, sent from my M4 macbook. I don't know why exactly it's doing this for you but mine is idling with ~500MB RAM (about as little as you can get with a reasonably-sized Rust codebase and a language server) and 0% CPU.
I have also really appreciated something that felt much less janky, had better vim bindings, and wasn't slow to start even on a very fast computer. You can completely botch Cursor if you type really fast. On an older mid-range laptop, I ran into problems with a bunch of its auto-pair stuff of all things.
Why are the Zeds guys so hung up on UI rendering times....? I don't care that the UI can render at 120FPS if it takes 3 seconds to get input from an LLM. I do like the clean UI though.
Oh, then what's the issue? I'm using Zed on Mint and so far I've only had one issue, the window being invisible (which I fixed by updating GPU drivers)
I don't think Zeta is quite up to windsurf's completion quality/speed.
I get that this would go against their business model, but maybe people would pay for this - it could in theory be the fastest completion since it would run locally.
> the fastest completion since it would run locally
We are living in a strange age that local is slower than the cloud. Due to the sheer amount of compute we need to do. Compute takes hundreds of milliseconds (if not seconds) on local hardware, making 100ms of network latency irrelevant.
Even for a 7B model your expensive Mac or 4090 can't beat, for example, a box with 8x A100s running FOSS serving stack (sglang) with TP=8, in latency.
Running models locally is very expensive in terms of memory and scheduling requirements, maybe instead they should host their model on the Cloudflare AI network which is distributed all around the world and can have lower latency
