Lost access to a hosted coding assistant overnight, how close can a local setup actually get for daily work?
Access to the hosted assistant I used every day went away with no warning and no explanation, and I am not interested in being in that position again. So the question is not "which model is best", it is how close a machine under my desk can get to what I lost, and where exactly it stops.
What I actually used it for, in rough order of hours: reading an unfamiliar codebase and explaining it, writing tests against existing code, refactors that span several files, and the occasional "why is this failing" with a stack trace pasted in.
What I have available: a single consumer GPU with 24GB, 64GB of system memory, and a willingness to leave something running overnight if that is what it takes.
What I want to know from people who have actually made the switch rather than benchmarked it:
- Which part degrades first? My guess is the multi-file refactor, because that seems like a context problem rather than a reasoning one, but that is a guess.
- Is 24GB the real wall, or is it the thing people say before admitting the tooling is the actual problem?
- Did you end up with a hybrid, local for the routine, hosted for the hard thing - and if so, what decides which is which?
I am not asking for a model recommendation with a date on it, because that will be wrong in two months. I want to know which categories of work moved across cleanly and which never did.
@vram_is_the_wall · 3w ago
Memory is the wall, but not in the way it gets discussed. The question is not whether a model fits - it is whether the model plus a useful amount of context fits at the same time.
On 24GB you can comfortably hold a capable mid-sized model. What you cannot comfortably hold is that model and enough context to have read four files properly. So the thing that degrades first is exactly what you guessed, and the reason is the one you guessed, which means your instinct here is sound.
Practical shape of it: single-file work, writing a test, explaining a function, fixing an error with a trace: moved across for me completely and I have not gone back. Anything where the answer depends on holding several files in mind at once got noticeably worse, and worse in a bad way: it does not fail, it confidently edits the wrong layer.
So the hybrid split you are guessing at is roughly the correct one, and the deciding question is "how many files does the answer depend on".
Reply
Report