What have you vibe coded lately, /g/oys?
>>106408362I've started vibe lifeing at this point.>Hey ChatGPT, can you add a reverse TLS proxy for my SOAP API in Kubernetes, update my stock portfolio in accordance with the latest tariff policies, and give me the top 5 places I can go to find hot singles in my area? And make it snappy.
https://rentry.org/simple4chanencoder
That is a child!
I've only ever vibe coded bash scripts because it actually puts the notes in there for me, and I feel that I always forget what shit does in it while being too lazy to note.
Spoonfeed me. What can I use for vibe coding? Just asking online services like chatGPT? Shouldn't I be running a local LLM that can access the project directory?
>>106410938You sound like you don't want to pay sohttps://github.com/QwenLM/qwen-code
im vibe coding a web browser alike system using litehtml to replace generic linux tools like polybar and rofi
>>106408362Kill yourself
>>106411253Cool, thanks. That means I'm still feeding data into an online model somewhere as I'm making something, but it's better than nothing.
>>106408362Vibe coded a scripting parser, mainly because I couldn’t be bothered to go through and write a parser myself. Probably ended up rewriting half of it myself because the model kept making dumb assumptions about what I wanted.
My 19000 line local media collection browser is starting to become unmaintainable bros..
>>106410938google ai studio has the best model and it's completely free with very lax usage limits.
Here is a list of agentic (vibecoding) tools available that I'm aware of. If I've forgotten something, please let me know. My knowledge of local stuff is pretty limited, so that section is sparse. We should probably get a rentry or something going.Softwares>Terminal/CLI-based:Crush (by Charm)OpenCode Gemini CLIQwen CodeClaude CodeGoose (by Block)>IDECursorWindsurfTrae (made by ByteDance aka tiktok)ClineRoo CodeAgentic Models>APIClaudeQwenGeminiDeepseek V3.1Kimi K2Z.ai GLM 4.5>LocalQwen distillsOpenAI gpt-oss-20b>Special mentionGroq and Cerebras serve Kimi K2 and Qwen3 Coder at extremely high inference speed, 500-2000 tok/s specifically for agentic workflows.Gemini CLI and Qwen Code have a pretty generous free tier of 1000 and 2000 requests (not tokens) per day. Yes you need to make an account.I haven't used Elon Musk's Grok or ChatGPT for vibecoding, so I don't know if they wire into these tools properly.
>>106412420Oh, I guess there's Copilot and I know Jetbrains and Zed have some AI stuff built in. I don't know if it's vibecoding or just fancy autocomplete. Also there's like half a dozen VSCode forks that I probably missed.
>>106412420I guess Gemini CLI is what I want for Google AI Studio in VSCode? Thanks anon.
>>106412485You can run Gemini CLI in a terminal even without VSCode. I'm guessing there's probably a VSCode extension that hooks it into the IDE similar to Claude Code's vscode extension? You might have to do some digging. This area moves so fast I can barely keep up with half of the stuff people are using lol
>>106412226That's what happens when you don't ask for modularity and client-server design
>>106412932It's annoying how these things still aren't smart enough to do this on their own. I don't think any amount of prompting can keep them from taking shortcuts in every situation.
>>106413399You just need to keep in context the project tree and a short explanation of what is doing what when asking for a new feature.
>>106413399I’ve noticed this happens when using Claude Sonnet, who will straight up lie to you about what it does. The other models tend to be a lot better in this regard. If you must use Claude it’s better to get a bigger thinking model in a brainstorm session to plan out architecture before using Claude to start implementing.Thinking about it, I’m actually starting to prefer most models over Claude even in implementation.
Any tips on preventing the models from getting trapped in hallucinations?I've tried vibe coding an app and it worked for a while, but then it stopped. I don't know why, supposedly some weird Android inter-app limitation but I don't know for sure. Either way, when I ask GPT-5 or Gemini to solve it they just get stuck in a certain hallucination. For example, GPT-5 was sure a code change could fix it. The code didn't compile. It then reverted the changes. It then tried to implement the exact same fix again. Repeat ad infinitum.
I’m vibe coding a script that logs the IP of coomers posting their anime kiddie porn and forwards it to the FBI to be put on a watchlist
>>106413625>SonnetOpus or nothing, they botched their last Sonnet
>>106414049Just tell it what has been tried and how it didn't work
>>106414049Context management. The models need enough context to solve the problem, but not so much that it creates a problem space they get lost in. If things are starting to feel like they are going off the rails, start a fresh session. Also models have different things they are good at, sometimes you need to make the call at which one you need for the job. They are tools after all.
>>106414093Already made https://github.com/Cyberes/automated-fbi-reporter
>>106412355used to be the best, you mean. it popularized and got nerfed, now there are only mid models
>>1064142192.5 Pro is still very good
I don't really understand how CLIs are supposed to help? I've just been copy pasting back and forth from the chat
>>106414260Gpt5/opus4.1 are better, but it's still very useful for common stuff with the high limits
>>106414362They can perform operations directly on you system like creating folders, files, initializing git repos, and interacting with other processes
>>106414382>gpt5I assume the Pro version?
>>106414528Yes
>>106414362It has access to your code base and can even run commands. Like grep. Or start/stop services. Run tests and analyze results. It’s not comparable to copy pasting stuff to/from chat.
>>106414520>>106414542You better run that shit in a sandbox and not in production
>>106408362Stupid sexy sword
>>106414557Literally no one does dev work on production servers. Let me guess, you are a typical /g/ larper who has very strong opinions about technology while having zero professional experience.
>>106414611Shush retard https://x.com/jasonlk/status/1946066422477529487 there are a dozen of you out there
>>106408362I'm asking grok/gemini to make me a bot for my mmo game. rotation bot for dps/healing and finding health of party members. Its creating a pretty sophisticated script for me. 500+ lines of code, objects, classes, caching mechanisms, performance optimizations, multithreading, etc. With many neat tricks. Very happy with it. My own basic coding would take me months/weeks of refining and even then it would only amount to 1/10th the functionality
>>106414362I don't understand why someone would choose a CLI tool over an LLM powered editor.
>>106414520That sounds like a very bad idea. Yeah I don't think I need it yet.I was wondering where all the horror stories were coming from.
>>106414130Bullshit. It's pretty good
>>106414715The CLI tools and their models have good tooling and can perform like virtual agents that do work for you. It's very neat when you know how to handle it, no need for UI in most cases.
>>106414663That's actually cool as hell. Are you using Cline or something?
>>106414557>He's never let Claude go buckwild with sudo permissions on system foldersYou truly haven't lived yet
>>106408362>>106414566who dat?
>>106418511Loli sword from Tsuyokute New Saga
>>106408666
>>106412420>>106412485I have compared gemini cli and web ai studio. The latter is much smarter and does doesn't make as many silly mistakes. Cli wastes a ton of requests and fucks up files.ai studio is more hands on but with a good workflow (like a script to automate copying files and applying edits) it is superior to gemini cli.
>>106421165How do you give it your code to look at for context when using web AI studio?
>>106422177I use a tool, see >>106251751It's fairly frictionless for small projects as you can just copy the entire project. For large projects you have to select the files yourself, but that also improves llm performance.
>>106408362>manager vibe coding with AI for 6 weeks >basically just querying databases>often tells me about issues he's having>timeouts during queries>variables not populating?>eventually takes down the prod db>get curious about what's going on>do code review>AI slop, as one would expect from a nocoder>misuse of sqlalchemy engine>misuse of exceptions>misuse of threadpoolexecutor variable access>suspect python 3.8 being used in 2025>.format() for queriesAll that and the AI couldn't figure out what was wrong with the code.Good luck /vcg/!
>>106410399upload the code to github so we can laugh it
>>106418992Does it have a pussy?
>>106408362>vibetrooning general
aren't you concerned about your vibecode wrecking your filesystem or introducing vulnerabilities to your application
>>106423234We aren't making anything super complicated that's gonna run in a FAGMAN company, and even if we did, it would be no different from the recent incidents FAGMAN companies have had
