OrionPod
Run AI Locally
A desktop app and a command line for local models.
No cloud. No account. Just your machine.
Run AI Locally
A desktop app and a command line for local models.
No cloud. No account. Just your machine.
Everything you need to run AI locally
AI that lives on your machine and works for you. Offline and always available, in a chat window, a terminal or behind an API for your other tools. No subscriptions, no rate limits, no one watching.
The app, the command line and the llama.cpp engine in one small download. Nothing else to install.
Native Rust and Tauri, with no Electron shell and no Python runtime. Each model runs in its own engine process, so a crash never takes the app down, and the engine restarts on its own.
Try it outNo telemetry, no account. A network ledger lists every outbound connection the app makes, so you don't have to take our word for it.
Metal on Mac. Vulkan on Windows and Linux, so one build covers NVIDIA, AMD and Intel. MLX models on Apple Silicon.
Browse and download GGUF models from Hugging Face inside the app, split files and vision projectors included. Already have models from another app? Link them without copying a byte. Every model shows how it fits your memory before you load it.
Streaming chat with saved conversations and collapsible reasoning. Give the model tools from MCP servers, install plugins and skills, and approve what runs.
Run as many models as fit, side by side. The Instances page shows each one's health, speed and memory, and they can keep serving after you close the app.
See whether a model fits your GPU and memory, with the settings that make it fit and the reasons. Then measure its real speed on your machine.
Tokens per second, latency, memory and GPU, live as your model runs, on a dashboard of glass charts.
A clean, native desktop experience
Every model is also a local API server. Drive it from a terminal, run your coding agent on it and see exactly what it did.
Developer toolsUp and running in a minute
Free, for macOS, Windows and Linux. Open it and you're in. No account to make.
Pick one in the app's Hugging Face browser, which shows what fits your machine. Already have models from LM Studio or Ollama? Link them in place.
Load the model and talk to it. Everything runs on your machine, online or off. Want it in your other tools too? See the developer tools.
8 GB of RAM is a comfortable minimum for 7B models. The fit advisor tells you how any model fits before you load it.
Common questions & troubleshooting
Because OrionPod isn't signed with an Apple Developer certificate (yet), macOS Gatekeeper blocks it on first launch. This is normal for indie apps. Here's how to install it:
OrionPod_0.6.0_universal.dmg
OPTION A Easiest
OPTION B If Option A doesn't work
Any GGUF model that fits your machine: Llama, Qwen, Mistral, DeepSeek, Gemma and whatever llama.cpp supports next, since each model runs with its own chat template. Browse and download them from Hugging Face inside the app or with orionpod pull, or link the ones LM Studio, Ollama, Jan, GPT4All or the Hugging Face cache already downloaded, without copying them. On Apple Silicon, MLX models run too with the optional MLX engine.
Yes. Models run on your machine, and there is no account, no telemetry and no analytics in the app. OrionPod only goes online for things you ask for, like downloading a model or checking for an update (which you can turn off), and a network ledger in Settings lists every outbound connection it made, so you can check. Read our privacy policy. It's short, because there's nothing to collect.
A Mac on macOS 10.15 or later (Apple Silicon recommended, for Metal), a Windows 10 or 11 PC (x64), or Linux (x64, glibc 2.35 or newer). On Windows and Linux, GPU acceleration uses Vulkan, so one build covers NVIDIA, AMD and Intel GPUs. Without one, models run on the CPU. For 7B models, 8 GB of RAM is a comfortable minimum, and the fit advisor tells you how any model fits before you load it.
Yes. OrionPod is free, an individual project built with love: no subscriptions, no paywalls, no premium tiers. Its agent engine, orion-core, is open source under the MIT license.