Cline

“(if you paste one failing Cline prompt's token count (or roughly how big the context is when it hits 2 tok/s.............) It's usually instant, the prefill bogs and then it's over. I'll carry on with it, try some more debugging, and report back.”

“it’s not on clinepass, you can just pay for it with credits”

“The Cline team bundled both their old/new extensions into a single release, then used a PostHog feature flag to safely toggle between them.”

“I tried cline but its too hands on with the act/plan thing.”

“I suggest you **first take an idea and develop it enough to make a solid methodological base** ( you can use AI to soundboard here! just make sure you aren't in some delusional loop!), and then use your AI ( via Cursor,Cline,Copilot etc ( I personally use Copilot + ZooCode for debugging, which is RooCode's newer community version.)) to make that base and tweak until you think it fits your project and what you basically have qualitatively in your head.”

“Yeah I would suggest trying a different harness for example pi.dev. check clines system prompt it could be using a lot of context. Also others documented that Q4 is kinda the lowest you can go.”

“I got it working with VS Code + Cline, but right after prompting, it keeps reading the same 2 files over and over again. I'm using 64k context window and medium reasoning setting, Q5 XL model.”

“Get OpenCode for example- or Cline or Roo extension- they usually have some free models- which can be used.”

“Which AI coding tool do you open first?”

“Cline was the second most popular coding agent at a friend's company (after Claude Code).”

“Fits 24gb vram and works well with cline as a coding agent. I rarely have to use frontier models now, qwen can do the mundane work for hours.”

“That's basically the Cursor setup without the subscription.”

“Cline pass ($5 even tho it says $10) or command code go”

“Not true of other AIs, I burn through my work tokens quite easily if I am not careful, and I tried Cline first... can burn through the $10 clinepass limits in 2 weeks.”

“The only weirdness here is the amount of overhead work the client (in your case a laptop) has to do to continuously render the streaming response from cline. you can play with these settings too but that bottleneck will actually slow the harness down even though most of what is happening is on the LLM/code server”

“AIHubMix gives you free access + an API key you can use with Cline in VS Code”

“Cline pass lo sto provando e per 5€ ne vale la pena, a 10€ vince opencode. Ti consiglio Opencode, appena scade Cline tornerò. Prova anche Commandcode Goat, anche lì non è male”

“Which one is better opencode go or cline pass both are $10.”

“And for the API question: paying API prices will be always more expensive than paying a subscription for the model provider directly. You will get a lot more usage per dollar paying 200 dollars for Claude Max (or you can choose a cheaper tier, like the Pro 20 dollars a month) then API.”

“I use Cline in VScode with a free model that claims to have a 1.3M token context window and free input/output. But why am I hitting a daily limit after using only 178k? (i forgot to take screenshot for 2nd image).”

“Stack: LM Studio and then Cline plugin for VS Code. You can change LM Studio to llama.cpp, and cline for roo code. Good luck.”

“so many options but what multi-model AI platform is good to use? OpenCode Cline Kilo Code Cursor Copilot Command Code RooCode”

“Cline showed thinking when I used it before, maybe it's a setup issue? I don't think I did anything particularly special with the setup either.”

“Next I tried Cline, which is almost perfect. It looks nice, shows code difs, shows the task list and all the visual gubbins, but it won't let me expand the "thinking" block to see what the model is doing, and with Qwen it can be "thinking" for a loooong time!”

“- They don't get the best performance compared to Coding agents like Cline/ Claude code on most coding agent benchmarks”