H.O.T-Jarvis 0.1.0: an assistant that grows its own skills

Today marks the first release of H.O.T-Jarvis 0.1.0 — an open-source, local-first personal AI assistant that runs on your machine, for free, and gets more capable the more you use it.

Why another “Jarvis”

The name is everywhere, attached mostly to voice clones that open websites and read the weather. We wanted something with a sharper identity: not a demo, but an assistant you can actually trust with real context, run forever without a bill, and inspect completely.

What’s in 0.1.0

All four hero features are real and tested, not roadmap items:

• Self-evolving skill library. Ask for an ability; it writes the code and a test, proves the test passes, and refines on failure. Untested skills are flagged and refused.
• Reflective reasoning-memory. It re-reads its own action log and keeps lessons about what worked and what didn’t — then applies them.
• Calibrated confidence. Every answer carries a self-rating; below a threshold it asks instead of guessing.
• Replay & undo. Every action is recorded and reversible, with an audit that proves the log reproduces memory exactly.

Underneath: a Tauri v2 desktop app, a Tauri-independent Rust core, a local-first model router (Ollama → free cloud fallback), persistent SQLite memory, spoken replies through your OS voices, and a Jarvis-style HUD. Around 90 tests, CI-gated, Apache-2.0.

Free and private, on purpose

Inference runs locally by default; no paid API is ever required, and your data stays in a folder you control — exportable and wipeable at any time. See Local-first and free for the reasoning.

Try it

The quickstart takes about ten minutes: install a free local model, run the app, and confirm it remembers you across a restart. Then author your first skill and watch it test itself.

Feedback and contributions are welcome on GitHub.