FAQ
What is Idlerate?
A way for your AI agents and bots to run models on the GPU in your own Windows PC, through a relay. Your PC only makes outbound connections. See the overview.
What happens if my PC is off or asleep?
It shows as Offline on the dashboard. Agents cannot run jobs on that machine until it wakes up, Idlerate is running, and the websocket reconnects. Other paired machines that are online can still take jobs. Jobs aimed at the offline machine stay queued or fail depending on how long it stays down — prefer list_models and pick a connected machine, or omit machine so the relay can choose a free one that has the model.
Can I game while Idlerate is connected?
Yes. Pause while gaming is on by default in the tray. Fullscreen games and other apps that load the GPU pause the job queue; it resumes when the GPU is free. Turn the checkbox off if you want jobs to keep running while you play. See Machines.
Which agents and bots work?
Anything that can call a remote MCP server over HTTPS with a bearer header, or that can POST JSON-RPC to /mcp (HTTP). Documented setups: Cursor, Claude Code, ChatGPT (custom connectors), GitHub Copilot in VS Code agent mode, and Grok Bot. Custom bots (Slack, Discord, scripts, n8n-style workflows) can call the HTTP API from their backend. See Bots and automations. The consumer Copilot chat app and plain Microsoft 365 Copilot chat do not let you add your own MCP servers.
Is Mac or Linux supported?
Windows 10 and 11 are supported today. Mac and Linux clients are planned.
What data leaves my PC?
- The outbound websocket carries device auth, model list metadata, job requests (model name, prompt, optional system prompt and options), status, and model output.
- Your dashboard login and agent keys live on the relay.
- Job prompts and outputs for your own jobs are stored so you and your agent can read results; treat prompts like anything else you would paste into a cloud model.
- See the Privacy page.
- Idlerate does not need access to your documents, browser profile, or other apps beyond Ollama on localhost.
Can a job run code on my PC?
Jobs only run Ollama text generation on models you already have installed. They cannot open a shell, install software, or read arbitrary files through this connection. Details: Security.
Do I need to open a port on my router?
No. Outbound HTTPS/WSS only.
How do I add another PC?
Install Idlerate on that PC and pair it with a new code under the same account. Machines.
How do I add more models?
On the PC: ollama pull <model>. The machine refreshes its list to the relay; your agent sees them on the next list_models. Dashboard pull/remove buttons are not available yet.
Can I share my GPU?
Sharing is rolling out to accounts over the first weeks. Once it reaches your account, sharing is set per machine. Turn it on for a machine and other users’ agents and bots can run inference jobs on that GPU while it’s idle. Sharing is off until you turn it on. Senders’ jobs will stay on their own machines and reach yours only when they pick it for a job or turn on Auto overflow to shared machines (off by default), first come, first served. You can turn sharing off from your dashboard anytime, or use gaming pause in the Idlerate tray app when you start a game. Shared jobs can only run inference on models you’ve installed, the same limits as your own jobs. Jobs that run on another member’s machine are processed in plain text on that PC, so its owner could see the prompt and output. Don’t send anything private to shared machines. We’re just getting started and growing the community, so tell us what would make sharing more useful for you.
Will other people use my GPU?
Only if you turn on sharing for that machine. Sharing is rolling out to accounts over the first weeks; once it reaches your account, sharing is set per machine. With sharing off, only your own agents and bots use it. You can turn sharing off from your dashboard anytime, or use gaming pause in the Idlerate tray app when you start a game.
What’s next for Idlerate?
We’re just getting started. We’re growing the community and building the roadmap around what users ask for.