XARIV

XARIV Hearth

Own your chat.
Keep it on your Mac.

A ChatGPT-like experience that runs entirely locally — small 1–3B open models through llama.cpp and Metal. Your prompts never leave your machine. Inspired by the local-first idea behind products like Osaurus, but wired into the XARIV control plane.

Open Hearth (local)Manage models

Private by default

Inference runs on localhost via llama-server. No telemetry, no training on your data.

Small models first

Llama 3.2 1B–3B GGUFs — fast enough on Apple Silicon for daily chat and demos.

Relay underneath

Download, deploy, and compare models with XARIV Relay. Fine-tune and publish later.

  • Today: browser chat at /app/hearth while Pulse + llama.cpp run locally.
  • Next: packaged XARIV Hearth.app (.dmg) — same stack, no terminal.
  • Later: fine-tune with Relay, host and publish your own models.

← XARIV Relay (serving engine)