Hearthscale

Hearthscale is a local AI platform with the line "Own your intelligence": it is built for everyone, not just developers, and runs text, voice, image, video and music models on your machine so data stays on the device. The site pitches it as free, private and free of heavy setup, and it ships apps rather than one giant chat box — install a meeting notetaker, a music or video editor, or an AI-aware mail client from the Marketplace, or build your own on top of local models.

Hearthscale Server can run on your most capable machine so every other device with Hearthscale can use the models there. Changelog 0.3.0 (8 November 2026) adds per-device model recommendations (memory, GPU and runtimes, no account required), video generation with Wan, LTX, HunyuanVideo, MiniMax-H3 and LingBot, Linux dictation hotkeys, and a sandbox for agent commands (Microsoft MXC containers on Windows; bubblewrap on Linux). Debian packages and a Windows-on-Arm beta shipped in the same release.

Highlights:

• Site: https://hearthscale.com , with docs, a changelog and a Marketplace;
• Maker @TR3x3r promoted a Product Hunt launch on 2026-11-09, framed as making local AI straightforward for everyone;
• Music studio mentions models such as MiniMax Music 3, ACE-Step 1.5 and Stable Audio 3;
• @hearthscale/app and @hearthscale/app-ui are on npm for building local apps on the platform.

Why it matters

Most "local AI" products are either a runtime for developers or a single chat client. Hearthscale puts recommended models, a command sandbox, Marketplace apps and an optional household/studio server in one place — closer to a local app store plus a model backend. Recommending only what the device can run well, and wrapping agent commands in MXC or bubblewrap, shows it is trying both to lower the beginner bar and to isolate what an agent may execute.

Limits

Video and large music/image models still hit local VRAM and CPU/NPU limits. The recommended list is only what the machine runs well; searching all of Hugging Face does not mean every weight will be smooth. The Windows sandbox needs MXC, Linux needs evdev/uinput and the shipped AppArmor profile, which some managed fleets will have to allow. Cloud sync and multi-user collaboration are not the pitch — on-device privacy is the point, so backups are yours.

1. Go to https://hearthscale.com and download the client for your OS (Server, Debian packages and a Windows-on-Arm beta are also offered);
2. Let it recommend models for the device, or search Hugging Face; install apps such as meeting notes, Music studio or Image studio from the Marketplace;
3. If one machine is stronger, run Hearthscale Server there and point other Hearthscale devices at it;
4. If agents will run shell commands, read the MXC / bubblewrap notes in the 0.3.0 changelog and confirm the sandbox matches your threat model.