LLM Providers
Local vs Cloud LLMs
Choose where the model runs based on your hardware, data handling needs, and access to a model server. MLJAR Studio can connect to a model on your computer, an internal server, or a hosted API.
Available choices
- Local desktop servers: Ollama, LM Studio, and Jan.
- Servers you manage: vLLM, llama.cpp, or Custom OpenAI-compatible. These can run locally or remotely.
- Hosted providers: MLJAR AI, OpenAI, OpenRouter, and Ollama Cloud.
Comparison
| Factor | On your computer | On your own remote server | Hosted API |
|---|---|---|---|
| Model execution | Uses your machine | Uses the server you connect to | Uses the provider infrastructure |
| Prompts and notebook context | Stay with local inference when the model is served locally | Sent to your configured server | Sent to the hosted provider; OpenRouter routes to a model provider |
| Setup | Install server and download model | Operate a model server and configure network access | Account access or API key and model |
| Hardware | Limited by your CPU, GPU, and memory | Limited by server capacity | No local inference hardware needed |
| Costs | Local compute and Studio license for configuration | Server infrastructure and Studio license for configuration | Provider usage or subscription; Studio license for configuration |
| Performance | Depends on model and hardware | Depends on server capacity and network | Depends on selected model, service limits, and network |
Choose a model that fits the task
A server can expose many model types. Select a chat or instruction model for notebook assistance. Agent workflows also need appropriate tool-calling support. Start with a short request, then try a representative analysis task before using larger datasets or longer conversations.
What “local” means
A loopback address such as 127.0.0.1 points to the machine running the notebook backend. If the backend runs elsewhere, the same address refers to that machine. A local-server preset does not make a remote endpoint local; check where the server actually executes the model.
Getting started
Choose a setup guide from the provider overview. AI Provider Settings require a lifetime MLJAR Studio license. Configure the endpoint and model, test the connection, save the provider, and try a notebook request.
Frequently asked questions
Can local models keep prompts on my computer?
Yes, when the model is served and executed locally. Check the endpoint and model setup; a remote server sends inference context to that server.
Does OpenAI-compatible mean OpenAI cloud?
No. It describes the API format. LM Studio, Jan, vLLM, llama.cpp, and other servers can expose this API while running models on your own hardware.
Can I switch between local and cloud providers?
Yes. Configure the desired provider in AI Provider Settings, test the connection, and save. The active endpoint determines where subsequent model requests are sent.