H.O.T-Jarvis

What makes it different

The jarvis space is full of voice clones that read the weather. H.O.T-Jarvis has one sharp identity: an assistant you can actually trust, that gets more capable the more you use it — with nothing to pay and nothing leaving your machine.

Grows its own skills

Ask for a new ability and it writes the code and a test, proves the test passes, and refines on failure. Skills that can’t prove themselves are flagged and refuse to run.

Remembers how it reasons

It re-reads its own action log and keeps short lessons about what worked and what failed — then applies them to future work.

Tells you when it's unsure

Every answer carries a confidence score. Below a threshold it asks a clarifying question instead of guessing.

Undo anything

Every action is recorded and reversible, with a replay audit that proves the log reproduces memory exactly.

Free, local, private — by design

Inference runs on your machine through Ollama by default, with free cloud tiers (Groq, OpenRouter) only as a fallback. No paid API is ever required, and your conversations, skills, and memory stay in a local folder you control — exportable and wipeable at any time.

three commands to a running assistant

ollama pull llama3.2       # 1. a free local model
cp .env.example .env       # 2. (optional) add a free cloud key instead
npm run tauri dev          # 3. launch the desktop app

QuickstartInstall, run, and have your first conversation that survives a restart.

Author your first skillAsk Jarvis to build a skill and watch it self-test.

ArchitectureHow the router, memory, and skill engine fit together.

Tauri command referenceEvery IPC command, with arguments and return shapes.

Built with

Tauri v2 · Rust core · React + TypeScript · SQLite · Rhai · Ollama · Apache-2.0. No telemetry, no accounts, no lock-in.