Own your
intelligence.

Aspen is private AI that runs 100% locally on your own device. You download a free app for Mac, Windows, or iPhone, open it, and start chatting, coding, or analyzing images. There is no cloud, no server in the middle, no subscription, and no account. The model runs on your hardware, so your conversations never leave your machine.

Yes. The app is free forever on the devices you already own: Mac, Windows, and iPhone. No subscription, no account. The Aspen device is a separate, optional product you can preorder, but you never need it to use Aspen.

Download the free app for Mac or Windows from runonaspen.com, or run one command in your terminal — curl -fsSL https://runonaspen.com/install.sh | sh — which installs everything and adds Aspen to your apps menu. After that you never need the terminal. On iPhone, install "Aspen Local AI" from the App Store.

No. Aspen is a normal chat app: download, open, and ask. There is no terminal or configuration required for everyday use. Developers can optionally use the built-in OpenAI-compatible API, but most people never touch it.

Yes. Because the model runs on your machine, core chat and coding work with no internet at all. You only need a connection for optional tools like live web search, or to download a new model the first time.

The latest open models, including Llama, Qwen, DeepSeek, Mistral, and Gemma. Aspen detects your hardware and recommends a model that fits it, and can update to a newer, better model automatically when one is released.

Yes. "Aspen Local AI" is free on the App Store. It connects to the AI running on your own computer, so you can chat with your private models from your phone, anywhere.

It is a dedicated, always-on machine for people who want to run the largest models around the clock without using their own computer. Think of the free app as the product for almost everyone, and the device as the optional high end for power users. It delivers about 1 petaflop of AI performance, 128GB of unified memory, runs models up to roughly 200B parameters, and is silent. You never need it to use Aspen.

Yes. Aspen exposes an OpenAI-compatible API. Change the base_url and api_key — two lines of code — and your existing tools run against your own private AI. It works with the ChatGPT and Claude SDKs, plus tools like LangChain, Cursor, and Continue.dev.

It is normal and safe. Windows shows this for any new app it has not seen many times yet, because Aspen is from an independent developer. Click More info, then Run anyway. If Windows Defender blocked the download, open your Downloads folder, right-click the Aspen file, choose Properties, check Unblock at the bottom, then run it. Windows code signing is rolling out to remove the warning entirely.

A local LLM is a large language model — the kind of AI that powers chat assistants — that runs directly on your own computer instead of on a company's servers. Your prompts are processed on your hardware and never sent to the cloud, which makes local LLMs private by default and usable offline.

You download the model's weights (a large file) once, and an inference engine on your machine uses your CPU or GPU to generate responses token by token. Tools like Aspen package the model, the engine, and a friendly interface together so you do not have to set any of it up yourself.

For most everyday tasks — writing, summarizing, answering questions, coding help, analyzing a document or image — modern open models running locally are genuinely good and often indistinguishable from cloud AI. The largest cloud models still lead on the hardest reasoning tasks, but the gap narrows every few months, and local models win decisively on privacy, cost, and offline use.

Chat and brainstorm, write and edit text, generate and debug code, summarize documents, analyze images with a vision model, search the web and cite sources, and power your own apps through an API — all without sending anything to a third party.

No, not for core use. Once the model is downloaded, chat and coding run fully offline. You only need a connection to download a new model or to use optional online tools such as live web search.

A modern laptop or desktop is enough for capable models. Apple Silicon Macs (M1 and newer) are excellent because memory is shared with the GPU. On Windows or Linux, more RAM and a recent GPU let you run larger, faster models. Aspen checks your hardware and picks a model that fits, so it works on a wide range of machines.

As a rough guide: 8GB runs small models (around 3B parameters), 16GB comfortably runs 7–8B models, 32GB runs 13–14B models, and 64GB or more runs 30B+ models. Quantized models need less. Aspen estimates the requirement for each model and flags ones that may be too large for your machine.

Yes. Many models run on CPU alone, just more slowly. Apple Silicon is a sweet spot because its unified memory acts like fast GPU memory. A dedicated GPU mainly helps with speed and lets you run larger models.

A cloud LLM runs on a company's servers: it can be very powerful but it sees your data, usually costs a subscription or per-token fee, and needs internet. A local LLM runs on your machine: it is private, free to run, and works offline, with capability bounded by your hardware.

Yes. The open models Aspen uses are released under licenses that permit local use, and in most cases commercial use as well. Running them on your own hardware is entirely legal; you simply follow each model's license, which Aspen surfaces for you.

Yes. There is a large ecosystem of high-quality open-weight models — Llama, Qwen, DeepSeek, Mistral, Gemma and others — that are free to download and run. Aspen bundles these so you can use them with no cost and no account.

It depends on your hardware and task. For most machines, a mid-sized Qwen or Llama model is the best all-round choice for chat, tools, and coding. Smaller Gemma or Qwen models are great on lighter laptops, while 30B+ models shine on high-memory machines. Aspen recommends a good default for your specific computer.

Yes — modern open models are strong. For writing, summarizing, coding assistance, and everyday questions they perform at a level that surprises most first-time users. The very largest paid cloud models still lead on the hardest tasks, but free local models are more than capable for the vast majority of real work.

It is genuinely free. The Aspen app is free forever, and the models are open weights. The "cost" is that the AI runs on your own electricity and hardware instead of someone else's servers — which is exactly what keeps it private. The only paid product is the optional Aspen hardware device, which you never need.

Paid cloud LLMs charge a subscription or per-token fee and run on remote servers that process your data. Free local LLMs run on your machine at no cost, keep your data private, and work offline. Many people use free local models for daily work and reserve a paid cloud model only for occasional heavy reasoning.

Usually yes. Most popular open models permit commercial use under their licenses, though terms vary by model and some have conditions at very large scale. Aspen shows each model's license so you can confirm before using it in a business.

When you run them locally with Aspen, no. The model has no network connection of its own; your prompts are processed on your device and are never uploaded or used for training. "Free" cloud chatbots, by contrast, often do log and train on your conversations.

Yes. With Aspen, the model runs on your own hardware and there is no server in the middle. Your prompts, files, and conversations are never transmitted anywhere and are never used to train any model.

No. Everything runs on your own hardware. The only time anything touches the network is if you explicitly use an online tool like web search, and even then the request goes out from your own machine and IP, not through Aspen's servers.

Never. Because nothing is uploaded, there is nothing for anyone to train on. Aspen's memory of you — its "World Model" — is a plain file on your own computer that you can view, edit, or delete at any time.

With a local LLM the conversation stays on your machine, so it is far safer than pasting secrets into a cloud chatbot. As always, treat any credential carefully, but there is no third-party server receiving it.

On a modern Apple Silicon Mac or a recent GPU, a well-sized model streams text about as fast as you can read it. Speed depends on the model size relative to your hardware: smaller models are faster, larger models are slower but more capable. Aspen picks a size that runs smoothly on your machine.

Usually the model is large relative to your memory, forcing the system to swap. Choosing a smaller or more heavily quantized model, closing memory-hungry apps, or using a machine with more RAM all help. Aspen flags models that may be too big for your hardware.

Parameters are the internal values a model learned during training; "7B" means seven billion of them. More parameters generally means more capable but also larger and slower. The right number depends on your hardware — many people run 7–14B models happily on a laptop.

Quantization compresses a model's numbers to use less memory and run faster, with a small and usually unnoticeable quality cost. It is what lets large models fit on consumer hardware. Aspen uses sensible quantized versions by default.

No. The free app runs well on a modern Mac or PC, and the better your machine, the better it runs. The Aspen device is simply the top of that range — a silent, always-on machine for running the largest models continuously. It is completely optional.

Yes. Aspen runs a local gateway that speaks the OpenAI API format. Point any OpenAI-style client at http://localhost:4000/v1 with an Aspen API key and it works unchanged.

Set the base URL to your Aspen endpoint and use an Aspen API key. For example, with the OpenAI Python SDK: OpenAI(base_url="http://localhost:4000/v1", api_key="YOUR-ASPEN-KEY"). That is the whole change — two lines.

Anything that speaks the OpenAI API: the official OpenAI and Anthropic (Claude) SDKs, LangChain, Cursor, Continue.dev, n8n, Zapier, and similar tools. You just swap the base_url and api_key.

Yes. Aspen can expose a private, secure HTTPS URL so your own apps and your phone can reach the AI on your computer from anywhere, while the model and data stay on your machine.

Three. An Owner key has full access including computer use and shared memory — only for devices that are you. A Family/member key gets its own private memory plus safe tools, with no computer use. An Anonymous guest key allows chat and safe tools only, is ephemeral, and is safe to share widely.