Local LLM apps: GUIs and clients for Ollama and LM Studio
Updated · 2 min read
A local LLM is a language model that runs on your own computer instead of in someone else's cloud. You download the model once; after that there is no API key, no per-token bill, and nothing leaves your machine. Two tools do the heavy lifting, and many apps connect to them.
The two runtimes
Ollama
Ollama downloads and runs open models (Llama, Qwen, Gemma, Mistral, DeepSeek, gpt-oss and more) and serves them on your computer at http://localhost:11434. It speaks its own API and an OpenAI-compatible API at http://localhost:11434/v1, which is why so many apps support it. If you sign in, Ollama can also run larger cloud models for you; those are convenient but not local.
LM Studio
LM Studio is a desktop app for finding, downloading and chatting with local models. It can also start a local server with OpenAI-compatible endpoints (chat, responses, embeddings), so other apps can use the model you loaded; its documentation uses http://localhost:1234 in examples.
Get started in five minutes
- Install Ollama or LM Studio.
- Download a model. In Ollama:
ollama pull llama3.2(a small, fast model) orollama pull qwen3for a stronger one; in LM Studio, search and download from the app. - Make sure the server is running (Ollama runs in the background; in LM Studio, start the local server).
- In your app, choose Ollama or LM Studio as the provider, or a custom OpenAI-compatible provider with the address above. Leave the API key empty (or type anything if the field is required).
- Pick the model by name and start chatting.
What your computer can run
A rough guide for quantized (4-bit) models, the size these tools usually download:
| Model size | Memory needed (roughly) | Runs well on |
|---|---|---|
| 1–4B parameters | 2–4 GB | Almost any recent laptop |
| 7–9B | 5–8 GB | 16 GB laptops, Apple Silicon Macs |
| 12–14B | 8–12 GB | 16–32 GB machines or a mid-range GPU |
| 27–32B | 18–24 GB | 32 GB+ Macs or a 24 GB GPU |
| 70B | 40 GB+ | High-memory Macs or multi-GPU PCs |
Speed depends on memory bandwidth more than raw processor speed: Apple Silicon Macs and dedicated GPUs are much faster than laptop CPUs. Start small, then move up if answers are too weak and speed is still fine.
Local, cloud, or both
Local models are the best choice for private documents, offline work and unlimited cheap drafting. For hard questions or big coding tasks, the frontier cloud models are still much stronger. Many apps below let you keep both: Ollama for everyday use and a cloud key (Claude, OpenAI, Gemini or OpenRouter) when you need more.
Apps that work with Ollama or LM Studio
VisitAnythingLLM
Free · macOS, Windows, Linux · Open source
Private AI app for chatting with your documents. Runs local models or connects to OpenAI, Anthropic, Gemini and others with your key.
Works with:OpenAIAzure OpenAIAnthropicGoogle Gemini+9 more
Questions
Do I need a GPU to run a local LLM?
No. Small models (a few billion parameters) run on an ordinary laptop's CPU, slowly but usably. A Mac with Apple Silicon or a PC with a recent graphics card makes them much faster and lets you run bigger models.
Is a local model as good as GPT or Claude?
Not usually. Models that fit on a laptop are smaller than the frontier cloud models, so they are weaker at hard reasoning and long tasks. They are good for drafting, summarizing, private notes and offline use, and many apps let you switch to a cloud key when you need more.
What address do I enter in the app?
For Ollama, http://localhost:11434 (or http://localhost:11434/v1 for apps that expect an OpenAI-compatible endpoint). For LM Studio, start its local server and use the address it shows; its documentation uses http://localhost:1234 in examples. Leave the API key empty or enter any text if the field is required.
Are my prompts private?
With a model running locally, yes: prompts and answers stay on your computer. Ollama also offers cloud models when you sign in; those are not local.
Sources
Checked against the providers' own documentation on 2026-10-23. Prices and limits change; the links below have the current figures.