EasyAI
Request access Check your PC
Install the worker

Docs · Worker

Install the worker

Install and enroll the Windows worker app with an invite token.

On this page

The worker app lets a Windows PC contribute spare capacity to the network. It is invite-only: the installer and a one-time enrollment token arrive with your invite. There is no public installer download.

Requirements

  • Windows 10 or 11, 64-bit.
  • An invite with a one-time enrollment token.
  • Ollama and a pulled model for real inference. The app’s built-in Canned runner produces deterministic self-test output and does not run a model.
  • Enough free VRAM for the model. The scheduler adds a 15% safety margin over the plan’s requirement, so a 9,500 MB stage needs at least 10,925 MB of declared or detected VRAM.

The only live model is LFM3-1.7B. Qwen3 models are planned and are not callable.

Install

  1. Run EasyAI-Setup.exe from your invite.
  2. Follow the wizard. The installer is per-user and installs under %LOCALAPPDATA%\EasyAI Worker; it does not require administrator rights or a UAC prompt.
  3. The installer is not code-signed. Windows SmartScreen may show “Windows protected your PC”. After verifying the file came from the project, choose More info → Run anyway.
  4. The current installer does not register the app at Windows sign-in. Autostart is stored as a preference for the signed installer that is planned.

Enroll

  1. Open the app.
  2. Paste the one-time enrollment token from your invite.
  3. Keep or edit the server address. Use the origin from your invite, shown here as https://<your-domain>.
  4. Choose Enroll this computer.

The token is used once and discarded; it is never written to disk. The app stores a rotating device session instead, so later starts do not need the enrollment token.

Run real inference

For real model output, install Ollama, pull the model, and select the Ollama runner in the app:

ollama pull lfm3

The Canned runner needs neither Ollama nor a model, but its output is a fixture. Use it only for self-tests.

Where data lives

The app keeps local state under %APPDATA%\EasyAI:

Path Contents
policy.json The contributor limits written by the Limits screen.
device-key.json Ed25519 device keypair; only the public key is sent.
refresh-token Rotating device session; never the enrollment token.
logs\sidecar-*.log Per-launch sidecar logs used by diagnostics.
model-cache\ Content-addressed GGUF cache for the llama.cpp backend.

The worker connects outbound to the control plane. Contributors do not need to open inbound firewall ports.

Next steps

Docs menu