This will be a quick one—I wanted to document some of the AI tools I’m using. Most are free, though I do have a ChatGPT Plus subscription. I might switch to Anthropic’s offering one day, but for now, I like having ChatGPT as a reliable assistant, especially for coding tasks.
OpenRouter
The first thing to mention is OpenRouter. This fantastic service routes your API requests to providers of nearly every major model you can think of. The best part? Their free tier is really, really good. If you upload $10 (just once), you get 1,000 requests per day to any free models. Some notable ones:
- Qwen3 Coder – a 480B model
- Kimi K2 – a 1 trillion parameter model
- Deepseek V3 0324
If you want access to premium models, OpenRouter’s pricing is also quite reasonable.
Ollama
Next is Ollama, a simple CLI that works on Linux, macOS, and Windows, allowing you to run a wide range of models locally. You probably have a machine capable of running some model.
I run Ollama on two systems:
-
Proxmox server
- Nvidia 2070 Super (8GB VRAM)
- 16GB DDR4 RAM
- Intel i7-9700K
-
Gaming PC
- Nvidia 5080
- 32GB DDR5 RAM
- Ryzen 7 9800X3D
I can fit a 4B model with a 16k context on the Proxmox server. It’s okay for some tasks, but limited. I get decent speed running deepseek-r1:8b with small context sizes (50+ tokens/sec). On the gaming PC, I can comfortably run coding models in the 7B–14B range. Ollama automatically handles CPU offloading when VRAM isn’t sufficient.
So what’s the point?
The key reason I use these locally is because they’re absolutely free. I’ve probably burned through $100 or more in tokens on experiments that didn’t go anywhere—but they helped me learn how the tools work. For example, when testing OpenCode, instead of spending tokens or burning through OpenRouter’s free credits, I connected it to Ollama and tested things locally.
Open WebUI
I use Open WebUI as the interface for all my local and remote models. It’s a self-hostable, ChatGPT-style front end that lets you aggregate your LLMs in one place.
I was initially mesmerized by chatting with local models in such a polished interface. It supports agents, MCP servers, RAG, and much more. But the real power is this: you can integrate nearly all your AI tools into this one interface.
- Add your OpenRouter API key and use any of their free models.
- Connect Ollama and chat with local models.
- Use the built-in OpenAI-compatible REST API to integrate other tools that expect a standard OpenAI endpoint.
Coding Tools
Finally, here are some tools I use for coding:
- I primarily use Aider because it fits perfectly into my terminal-focused workflow. It supports multiple backends and is easy to use.
- I’ve also tried Cline, RooCode, and OpenCode—solid options, but I prefer Neovim and the terminal.
- And of course, there’s gemini-cli, Google’s agentic CLI tool (a direct competitor to Claude Code). Its free tier is impressive.
That wraps up the AI tools I use—most of which are free. I’ll admit, we get access to some excellent models at Oracle with near-unlimited usage, although the tooling is still catching up to state-of-the-art things in the open-source world.