Hello everyone, I've been a local AI enthusiast for a while now, but every time I tried to get someone else running local AI, it was way too difficult. GitHub repos, different runtimes, different model formats. Most people's devices can run at least some of these models, but the setup is way too complex for most people. LM Studio somewhat solves a lot of this for chat and coding models, but I couldn't find a simple answer for on-device image, video, transcription and every other kind of model.
So I built Hearthscale. It's free and needs no account.
You can run almost any model you find on Hugging Face, as long as your device can handle it. I put a lot of work into supporting all kinds of runtimes. They're plug-and-play; I can add more in updates, and Hearthscale Server standardises all of them behind one OpenAI-compatible API. It's really powerful, and I find it hard to describe how seamless the setup is unless you have tried it.
It launches with meeting notes (Granola), system-wide dictation (WhisperFlow), image and music generation, and an agent. If you have a more capable device, run Hearthscale Server on it, and your other devices can use its models, end-to-end encrypted. It all connects without you having to set anything up! Also works on a home server in theory with Docker Containers!
Hearthscale is also a platform, which means anyone can build apps on top of it: there's an SDK and up-to-date docs made so coding agents can easily build apps with local AI. All of the first-party apps are MIT licensed, so you can change them any way you want and they give a great starting point. Want to change something about Granola? - You can't, but you can take meeting notes and make it to your liking with all of the core features already built!
Hearthscale runs on Windows and Linux today. I am somewhat of a perfectionist, and I feel like a lot is still missing, including a macOS release, but I wanted to get it out there now to get some feedback. I'll keep working on it!
I'd really appreciate your feedback, and please reach out about anything. I'll answer everything personally and fix whatever issues you run into.
Hello everyone, I've been a local AI enthusiast for a while now, but every time I tried to get someone else running local AI, it was way too difficult. GitHub repos, different runtimes, different model formats. Most people's devices can run at least some of these models, but the setup is way too complex for most people. LM Studio somewhat solves a lot of this for chat and coding models, but I couldn't find a simple answer for on-device image, video, transcription and every other kind of model.
So I built Hearthscale. It's free and needs no account.
You can run almost any model you find on Hugging Face, as long as your device can handle it. I put a lot of work into supporting all kinds of runtimes. They're plug-and-play; I can add more in updates, and Hearthscale Server standardises all of them behind one OpenAI-compatible API. It's really powerful, and I find it hard to describe how seamless the setup is unless you have tried it.
It launches with meeting notes (Granola), system-wide dictation (WhisperFlow), image and music generation, and an agent. If you have a more capable device, run Hearthscale Server on it, and your other devices can use its models, end-to-end encrypted. It all connects without you having to set anything up! Also works on a home server in theory with Docker Containers!
Hearthscale is also a platform, which means anyone can build apps on top of it: there's an SDK and up-to-date docs made so coding agents can easily build apps with local AI. All of the first-party apps are MIT licensed, so you can change them any way you want and they give a great starting point. Want to change something about Granola? - You can't, but you can take meeting notes and make it to your liking with all of the core features already built!
Hearthscale runs on Windows and Linux today. I am somewhat of a perfectionist, and I feel like a lot is still missing, including a macOS release, but I wanted to get it out there now to get some feedback. I'll keep working on it!
I'd really appreciate your feedback, and please reach out about anything. I'll answer everything personally and fix whatever issues you run into.