Skip to content

FAQ

No. You run one short command, sign in with Google, and your coding agent is linked automatically. You don’t need any networking knowledge. See Install.

Yes. Your prompts and replies travel straight between your own machines over an end-to-end encrypted link. They never reach our servers, and even the relay can’t read them. See Privacy.

A computer to run the model — a GPU helps, but a recent CPU runs a 7B model fine — plus your everyday laptop. The client machine you type from needs no special hardware.

The software is free, and the model runs on hardware you already own, so there’s no per-message bill and no subscription.

Any local model you can run with Ollama, plus larger models on vLLM for NVIDIA/AMD GPU servers. Waired bundles a coding model by default and you can swap in another anytime — see the model catalog and Switch the bundled model.

Claude Code and OpenCode are set up automatically by waired link. Any other OpenAI- or Anthropic-compatible client works too, by pointing it at the Local Gateway.

The Waired client is open source — you can read exactly what runs on your machines on GitHub. The coordination service that introduces your devices is the part hosted for you.

waired init signs you in with Google and enrolls the device. Enroll every device with the same Google account — that shared identity is what puts them on the same private network and lets one use another’s model.

What if my network has a strict firewall or NAT?

Section titled “What if my network has a strict firewall or NAT?”

Devices try a direct UDP link first. When a strict NAT or firewall blocks that, they fall back automatically to a relay that forwards the encrypted traffic. You don’t configure anything; connectivity works either way. (Waired does not use UPnP or NAT-PMP to open ports.)

Running a model on the same machine you’re typing on works locally. Using a remote device’s model needs network connectivity, because the control plane is what discovers peers and keeps the signed network map current. The control plane never sees your prompts — it only hands out keys and endpoints.

The Auto-Selector chooses the highest-quality model that fits your hardware’s memory, favoring efficient Mixture-of-Experts models on shared-memory machines. Preview a routing decision with waired infer --explain "say hi".

For Claude Code, the managed-settings integration is fail-open: if local serving is paused, disabled, or unavailable, requests go to the real Anthropic API so Claude keeps working — and the routing state is always visible via waired claude status and waired doctor. Because no credential is written into managed settings, your claude.ai subscription stays in use throughout. Waired never silently breaks your coding agent.