Local-first and free

H.O.T-Jarvis has a constitution, and two clauses shape almost every design decision: free forever at runtime and local by default.

Free forever

The assistant must cost you nothing to run. Inference goes through a model-agnostic router with this priority: local models via Ollama first (unlimited, private), then free cloud tiers (Groq, OpenRouter :free) only as a fallback. No paid API is ever required, and the app never nudges you toward paying. The router respects free-tier limits in code — caching, backoff — so the project never pressures anyone to upgrade.

This is why, for example, the router exempts the local model from rate-limit penalties, and why cache hits are labeled rather than silently served: the cheapest, most private path stays the default and the behavior stays honest.

Local by default

Your conversations, skills, insights, and event log live in a folder on your machine. Memory persists across restarts and upgrades, and it’s fully exportable, importable, and wipeable — your data, your control. Nothing has to leave the room. The one exception is explicit: if you add a free cloud key, requests to that provider necessarily leave your device — the app tells you which provider answered every time.

What this rules out

• No telemetry, no accounts, no lock-in.
• No feature that only works behind a paywall.
• No silent data exfiltration — side-effecting actions are gated and logged.

What it enables

A tool you can run forever without a bill, inspect completely, and trust with private context — because the private, free path isn’t a downgrade, it’s the default. Everything else in these docs is downstream of these two clauses.