Custom OpenAI-compatible Base URL (Ollama, LM Studio, vLLM, llama.cpp)

[deliciousbob]

Hi!
Thx for creating that tool, it looks like something that I was searching for months already.
I just run the installer on my Windows device, but when i wanted to select the Provider,

I saw that there is no option to select a custom OpenAI-API-Endpoint... Is that planned for future releases ?
It would be great to have an option for an custom Base-URL, the rest may be pretty much the same as the OpenAI Endpoint.
This would enable us to use that tool with local LLMs like Ollama, vLLM or LMStudio.

Thank you! best regards, Robert

[talentsache]

Hi Robert, thanks for the kind words — and good catch.

Short answer: yes, it's planned. A custom OpenAI-compatible Base URL (Ollama, LM Studio, vLLM, llama.cpp) is on the list for the next AIPointer release. You're right that today there is no way to point AIPointer at a local server: it only talks to OpenRouter, Anthropic, OpenAI and Google directly. I've renamed and labelled this issue and I'm keeping it open as the tracking item for that work.

Why it has been quiet here: this repository was re-opened in August after license violations — someone took the published source and rebuilt it into a separate product. The source is no longer published; the repo now carries the documentation and the official signed binaries only. AIPointer itself is very much still maintained, and the next release drops in the coming weeks. It is a bigger step than a Base-URL field: full computer control (the pointer moves the mouse and acts on your screen instead of only answering), voice control, and recording a task once to replay it as a reusable skill.

If you need local models today, the fastest path is Skales — same author, and AIPointer is built into it as a mode you can switch on. It already covers what you're asking for:

• Custom (OpenAI-compatible) as a first-class provider — llama.cpp, vLLM, koboldcpp, anything that speaks /v1, with your own Base URL.
• Ollama and LM Studio as their own providers, no API key needed, pointed at localhost by default.
• Around 25 providers in total (OpenRouter, OpenAI, Anthropic, Google, Groq, Mistral, DeepSeek, xAI, Together, Qwen, GLM, Moonshot, Hugging Face, Nvidia NIM, Cloudflare Workers AI, …), each with its own Base URL field.
• On top of that: models that run fully on your machine with a bundled server, on-device voice (Whisper for speech-to-text, Kokoro for read-aloud), vision, and agentic tool use / multi-step workflows.

Windows builds are signed and available there as well. Thanks for taking the time to write this up — it's exactly the kind of report that decides what gets built next.

[deliciousbob]

Hi Mario, thx for the detailed explanation about putting your software closed source, I fully understand your steps, this is not what opensource was designed for.
I am looking forward to new updates on AIPointer, as mentioned, I am exactly looking for a sleek handy tool like that.
I've tested Skales already and was able to connect my LiteLLM API Endpoint without issues. It is packed full of features i have not even scratched the surface on :D but i like it a lot.
Thx, best regards, Robert