A model on your own machine when you're offline. The full open network when you're connected. Same one-line API — and no off-switch.
# grab the single-file runtime curl -fsSL https://opentau.ai/opentau-local.js -o opentau-local.js # run it — set your key to reach the network when online OPENTAU_API_KEY=otau_... node opentau-local.js # then point any OpenAI client at http://localhost:11435/v1
No signal? A small open model runs locally — on a plane, in a bunker, behind an air-gap. Nothing leaves the machine.
Connected, or need a frontier-scale model? It transparently routes to the full OpenTau decentralized network.
You don't choose. It detects connectivity and routes each request — and every response says where it ran via a header.
The strongest form of censorship-resistance is inference that runs on hardware you control: no off-switch, no logging, no permission required. OpenTau Local gives you that floor — and the moment you reconnect, it upgrades to the whole open network, without changing a line of your code. BYO engine: auto-detects Ollama, or point it at LM Studio / llama.cpp / vLLM.
Honest note: on-device models are smaller and less capable than the frontier. Quality degrades offline and upgrades online — by design.
Part of the open frontier — read why decentralized AI matters.