Skip to content

Releases: 0pen-Sourcer/Hearth

Hearth v0.7.0-preview

Choose a tag to compare

@0pen-Sourcer 0pen-Sourcer released this 16 Jul 14:04

Local-first personal AI for Windows. Point it at any OpenAI-compatible model (LM Studio, Ollama, vLLM, llama.cpp) and it actually operates your PC: files, shell, apps, a real browser, your screen, and voice. No account, no telemetry, nothing leaves your machine.

Download

Two editions, pick one:

  • Hearth-Setup-v0.7.0-preview.exe (Full, ~1 GB) — includes the built-in GPU model server (bundled CUDA llama.cpp). Pick this if you want Hearth to load and run models for you out of the box on an NVIDIA GPU.
  • Hearth-Setup-Lite-v0.7.0-preview.exe (Lite, ~500 MB) — no bundled model server. Point it at LM Studio, Ollama, or any OpenAI-compatible endpoint you already run.

Installs to your user folder, no admin needed.

What it does

  • Read and edit files, run shell commands
  • Open and control your apps
  • Drive a real browser tab
  • See your screen from screenshots
  • Talk and listen with a wake word, TTS and STT, fully offline
  • 100+ tools plus a skills system that builds PDFs, decks and diagrams
  • Runs as a desktop app, a CLI, a headless bridge, or an MCP server

Notes

Windows-first. Reliability tracks the model you run; on 8 GB VRAM a 9B like Qwythos handles tool calls cleanly. Preview release, MIT licensed.