What is best code editor for local LLM deployment (LM Studio, llama.cpp) as of May 2026?

[-]

Alan_Silva_TI@reddit

RooCode was the best option, but they’ve announced that they’re abandoning the project (check their subreddit).

It originally started as a fork of Cline, and they even recommended that users switch back to it. Cline works quite well and has some useful features.

By the way, both are VS Code extensions, so they should be right up your alley.

[-]

OneSlash137@reddit

These models are ass for coding or development.

It’s probably getting caught in tool call loops because it constantly drops prompt prefill forcing the entire conversation to preprocess from the beginning including all the agents tool parameters etc. then while all that is chugging the tools just keep requiring the same thing over and over until they time out.

If you could manage to get that solved and run it full speed, it won’t give you anything reliable, well built, best practice, scalable, production ready.

LLMs for professional home development is not here yet.

[-]

false79@reddit

^-- I'd take that advice with a grain of salt. As a professional developer, there are gains to be made by offloading repetitive work to an LLM if you provide it with a sufficient amount of context, smaller achievable goals, let its reasoning connect the dots faster than a human would, free up time to do other things.

I find a common pattern among those who are frustrated with local LLMs for coding is not really knowing how to activate the parameters needed to produce the output they need. Cloud models do a lot of the heavy lifting but there are several here who have figured out you can get just as much done if you understand the limits of the model and how it does not read your mind with a zero shot prompt.

[-]

OneSlash137@reddit

Wrong.

[-]

false79@reddit

I dunno if you look over post history but there are fair amount people down voting your lack of expertise in this area.

I am sure if you knew what you were talking about, it would be very different.

[-]

JockY@reddit

Sounds like someone tried coding on an 8GB card with an IQ2_XXS on Cherry Studio and wondered why it doesn’t work.

[-]

OneSlash137@reddit

Try again.

[-]

stormy1one@reddit

Sounds like you have a configuration issue. Qwen3.6-27B is extremely capable when configured correctly. It’s not going to zero shot anything complex, but will absolutely be able to produce reliable code for production use. Tool calling was one of the main things they addressed in 3.6. Performance is excellent in vLLM v19.1 — all works having at least two instances - one for orchestration managing the other for coding.

[-]

OneSlash137@reddit

Wrong. I’m an enterprise architect. You don’t know best practice or production code.. you don’t know enough to know it’s bad. Because LLMs won’t tell you they are bad at something, they’ll just cinfidently say the code works.

And maybe it does, but most likely it’s doing really stupid things like making 5 separate database queries when only 1 is necessary.

Those are the types of things that fly under the radar until 10,000 people are using your app at once and the prod database explodes.

So, yeah. It isn’t a config issue.

[-]