Local AI
Private AI that runs on your Mac.
43 local models. One click to download. No terminal.
Pandore includes a library of open models that run on your computer through llama.cpp. Pick one, download it, and use it in your projects. Your questions and documents never leave the machine.
Choose where Pandore runs AI.
Automatic routing is on. Pandore selects local or remote AI for each request based on privacy, tools, complexity and cost.
Why run a model locally
A local model answers without sending anything to a provider. That matters for client documents, personal notes and anything you would rather keep to yourself. It also works offline, and it costs nothing per request once the model is downloaded.
The trade-off is capacity. A model that fits in your Mac's memory is smaller than the large models available in the cloud, so it is best for focused tasks: summarizing, rewriting, extracting, answering questions about a text. Pandore lets you combine both, so you do not have to choose once and for all.
A library, not a command line
Running a model locally usually means installing a runtime, finding the right file on Hugging Face and typing commands. In Pandore, it works like an app store.
- 01Pandore checks your Mac
The library shows your processor and memory, and recommends the models that fit.
- 02Download a model
Each model is a single file. Pandore downloads it and installs the runtime it needs.
- 03Use it right away
The model loads in the background and stays ready while Pandore is open, so the first answer comes quickly.
43 models from 10 families
| Family | Made by | Good for |
|---|---|---|
| Qwen 2.5, Qwen 3, Qwen 2.5 Coder | Qwen | Multilingual work, reasoning, code |
| Gemma 3 | Everyday writing and analysis | |
| Ministral 3, Mistral Nemo | Mistral AI | French and European languages, reasoning |
| Llama 3.2 | Meta | Fast general use |
| DeepSeek R1 Distill | DeepSeek | Step-by-step reasoning |
| Phi-3, Phi-4 | Microsoft | Compact reasoning |
| SmolLM2, SmolLM3 | Hugging Face | Very small, very fast |
| LFM2.5 | Liquid AI | Efficient general use |
| Granite 3.3 | IBM | Business writing |
| OLMo 2 | AI2 | Fully open research models |
Sizes range from 360 million to 30 billion parameters. Small models run on any Apple Silicon Mac; the largest need more memory, and the library tells you which ones fit.
Local, remote or automatic
In Settings, choose where Pandore runs AI:
- Local: every compatible request stays on your computer.
- Remote: Claude Code, Codex or Gemini CLI handle requests with your own account.
- Both: automatic routing. Pandore keeps simple or sensitive requests local and uses a remote agent when a task needs tools or more capacity.
Local models answer in the chat and handle text work. Tasks that need tools, such as working inside a project folder or running a Mission, use a remote agent.
See what stayed local
The AI usage dashboard shows how many tokens local models handled, next to remote usage. When you also use a remote agent that reports its cost, Pandore estimates what the local work would have cost at your own observed rate.
Questions
Before you start.
Which Macs can run local models?
Any Apple Silicon Mac with macOS 11 or later. The more memory your Mac has, the larger the models you can run. Pandore also runs local models on Windows 10 and 11.
Do I need to install llama.cpp or Ollama?
No. Pandore installs the runtime it needs when you download your first model.
Does anything leave my computer?
Not when you use a local model. The model, your prompts and its answers stay on your computer.
How much disk space does a model take?
From about 400 MB for the smallest models to about 19 GB for the largest. The size is shown before you download.
Keep reading
Keep your AI on your computer.
Free for Mac and Windows. No account, no subscription required.
Version 0.9.0 · free · macOS 11+ Apple Silicon · Windows 10/11 x64