Idlerate docs

Idlerate lets the AI agents and bots you already use run models on the GPU in your own Windows PC: coding agents (Cursor, Claude Code, GitHub Copilot), assistants (Grok Bot, ChatGPT custom connectors), and custom or automation bots (Slack and Discord bots, scripts, n8n-style workflows) that call MCP or the HTTP API.

Your PC only dials out to https://relay.idlerate.dev. Nothing on your network has to accept an inbound connection. Agents and bots talk to the relay over MCP (or plain HTTP). The relay queues jobs; your PC runs them through Ollama and sends the answer back.

Agent or bot

Cursor · Claude Code · Copilot MCP or HTTPS

Relay

relay.idlerate.dev Queues jobs, checks keys

Your PC

Idlerate + Ollama Outbound only · your GPU

Idlerate is models only. The relay cannot open a shell on your PC, read your files, or install new software through a job.

Requirements

Windows 10 or 11 (64-bit). Mac and Linux clients are planned.
An NVIDIA or AMD GPU. Very small cards and CPU-only machines still work; the installer picks a starter model by available memory (see Install on Windows).
An Idlerate account at relay.idlerate.dev.
Ollama, which the installer sets up for you if it is missing.

Quickstart

Install on your PC. The Windows installer is in private preview. Sign up at relay.idlerate.dev and verify your email, and we’ll send you the download link.

Recommended: let an agent with a shell on your PC run it for you. See Install with your agent for the copy-paste prompt.

Fallback: run the installer yourself in PowerShell:
```
.\idlerate.exe install
```
That copies Idlerate into your user profile, starts it at logon, installs Ollama if needed, and pulls one starter model. No administrator prompt. See Install on Windows.
Sign in and pair. Open relay.idlerate.dev, create an account (or sign in), and enter the pairing code the installer printed (your agent shows it to you; it’s also in %LOCALAPPDATA%\Idlerate\pairing.txt). Your machine shows up on the dashboard.
Connect agents and bots. On the dashboard, open Connect a new agent, copy the key (shown once), and paste the config into Cursor, Claude Code, ChatGPT (custom connectors), GitHub Copilot, or Grok Bot, or set it as an environment variable in your bot backend. Details and the paste-able agent prompt are in Connect agents and bots. To add Idlerate as a connector in an assistant bot or another MCP client, see Assistant bots and other MCP clients.
Run a job. Ask your agent to call list_models, then run_model, wait with get_status, and read the answer with get_result. Tool details are in Tools reference.