New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
AI agent. Which, Why and How you use?
Hi,
i know most of you dont like ai. I want to know which ai agent you use for coding your project
why you use specifically that ai agent?
How do you use it at cheapest or free?
i never use any agent. so i tried googling but dont get exact starting knowledge. suppose i download a free ai agent like cursor and want to use gpt-5 or kimi k3 or deepseek v4 or Claude Sonnet 5.5 model on that agent, how to do that?
which one is cheapest but best for coding software ?
Thanked by 1DrNutella
Do you like to use ai agent or web chat?
- Do you like to use ai agent or web chat?59 votes
- AI agent66.10%
- Web chat15.25%
- none, even no ai18.64%
Comments
i know for a fact that @MikePT is a Claude CUCK
If you are willing to pay, try Claude Code or Cursor.
Cursor is not free / free plan doesn't really let ou access LLMs. You can try Freebuff (if you are willing for your data to be shared aka. CharityHost of AI) or OpenCode Free (ZDR iirc) - they have experimental free models.
If you are willing to pay $10/mo OpenCode Go sub is pretty good.
$20/mo get ChatGPT Plus and use the Codex Web Connector to get basically unlimited usage. Or Claude if you are doing frontend and use Haiku.
GPT-5 is quite old, use GPT-6.1 Sol instead
Kimi K3 is overpriced from most vendors, Deepseek V4 again is old - use V4.1 Flash. Sonnet 5.5 is pretty good, you can use that in Claude Code too.
Most of these are separate agents/IDEs, not something you use inside another IDE (though some do offer this as a feature).
I use lowendgirlfriend. She doesn't code very well, but she tells me that I'm pretty.
This weekend I'll be finetuning LowEndGirlfriend on my proprietary internal dataset of only the highest-quality agent traces, ethically sourced
@wadhah
Sounds like a sexy weekend.
Damn, she will be well trained with such large proprietary internal dataset.
I believe you must've downloaded the entire internet at least twice by now.
i am looking for something cheapest but can code some erp or similar php, mysql based simple software. for complex code, i find cloude is best. i saw some news that people create full game with just few line commands with ai agent/IDEs. But i never used any IDE so i am totally blind in this point. tried to figure out while googling but didn't find any proper tutorial or explanation.
where i can download one for me? i am 38 and single.
Brother, @wadhah has got you covered. https://huggingface.co/huggingwadhah/lowendgirlfriend
I'd recommend you try out OpenCode free first
When I see ERP and php in the same sentence, a shiver goes down my spine. Please don't do that.
I work for a company that makes ERP software. Our devs use Claude. Some use Cline as the IDE. Others use Visual Studio and/or VS Code.
Personally, I have Codex and Cursor. I use Codex CLI most of the time and launch that from within Cursor because I like the terminal. Also use Claude at work, but generally not for coding.
Maybe if you want.
Check out Muse, your personal AI agent. Redeem my code in Settings within 48 hours of joining and we'll both get 1 billion Muse tokens.
Code: FTNKG1
https://muse.ai/join
Thank you.
I use Claude Code, Codex, and Agy.
I'm using one made by myself after I didn't liked any other (mainly vendor-locking):
https://github.com/BradPerbs/acestes-agent
https://tryacestes.com/ (a bit of what it can do)
it just plugs to whatever model or other harness you are using already (claude code, muse code etc..)
It's a fork of cloudterm with 260 commits on top for agentic use, coding etc..
AI Agent or AI Harness? I think the term you're looking for is harness which is the environment you operate the AI models in.
For best bang for your buck in terms of performance would always be if you go with AI providers directly as their subscription plans give the most subsidizations. 3rd parties aren't able to afford to subsidize your tokens and usage as much. So Anthropic Claude AI, OpenAI ChatGPT, Google AI subscription plans.
I currently use the following:
Below is my desktop-statusline Claude Code Mod plugin new inline usage metrics https://ai.georgeliu.com/p/claude-code-mod-desktop-statusline with my /consult-panel multi-AI monitor Mod - leveraging Codex GPT-6.1 Sol + ZAI GLM 5.3 Flash + DeepSeek V4.1 Flash & up to 8 different AI provider models in same Claude Code session.
I have similar /consult-panel mult-AI skill for OpenAI Codex too just Claude Opus 5.5 + ZAI GLM 5.3 Flash + DeepSeek V4.1 Flash
I share some of my AI project journeys and adventures on my AI Substack at https://ai.georgeliu.com/ if folks want to subscribe and follow
I don't use any agents never have. Almost everything I do is via VSCode with Claude Fable, except for dumb shit like questions that just go in the web UI or Gemini since that's free.
I’m using @inaba agent botnaba 👌🏽💯
For a simple PHP/MySQL app, I'd start with Gemini CLI since the free tier is enough to learn how an agent actually works in a terminal: reading your files, running commands, proposing diffs. Once you hit its limits or want more reliable multi-file edits, Claude Code on the Pro plan is the usual next step. Either way, ask it to write a short plan before touching code and keep the project in git so you can roll back when it goes off the rails.
Nowadays I just tell my Muse to “find the bugs overnight and fix them by morning”.
Muse has their own computer so they’ll clone the repository, install the toolchain, audit the code, write the fixes.
When Muse gets the patches ready, my Puppeteer clicks the approve button, and then the Pull Requests are up for review.
That’s how we work as a team.
i don't build any software, But using omp.sh for long running work especially when for edit verifying tests plans etc you can use many sub/api for now deepseek is enough, make sure don't listen to @Rubben
that last part is especially important
Librechat & Kilo CLI >>> OpenRouter
Claude code + Codex for coding (one develop and one review)
hermes + DS v4.1 API for daily
Qwen3.8-27B-uncensored @ local 5090 for XXX
I use Hermes Agent for everything. It's simply the most powerful harness/agent I've ever used. As main model I use DeepSeek v4.1 Flash (they just released a new version today). It's an extremely good model and costs a fraction of GPT/Claude models. I use that one for everything, from coding to hacking to research to automating business stuff. I still keep a 100/mo ChatGPT subscription for code reviews with Sol 6.1.
I forgot to mention that currently OpenDesign offers the best deal in the market: $120/mo of DeepSeek v4.1 Flash for just $10. It's amazing and the speed is great at 200-250 tps.
Fact
you are the real programer in this thread
>
Same setup here, but honestly, Hermes Agent + DeepSeek v4.1 Flash only works for the simplest things. Once the complexity goes up, the results degrade fast. For actual coding, it's exclusively Codex or Claude Code.
monthly 100$ is huge for me. i did complete an erp using free cloude, deepseek, chatgpt webchat but thats has limitation and need to describe file structure to new chat each time. thats why i am looking for free agent with free model.