Install with your agent
This is the main way to install Idlerate. You ask an agent that has a shell on your PC to run the installer. It reads the installer’s progress lines, shows you the pairing code, and you confirm the machine on your signed-in dashboard. The agent never needs your password.
Prefer to do it yourself? Use Install on Windows. The steps are the same.
Before you start
- Windows 10 or 11 (64-bit). No administrator account is needed.
- An Idlerate account at relay.idlerate.dev.
- The installer link. 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. The email gives you two links:
idlerate.exeand itschecksums.txt. - An agent that can run commands on this PC, for example Cursor or Claude Code in a terminal on the PC, or Grok Bot on a connected computer. A cloud agent without a shell on the PC can’t run the installer.
1. Tell your agent
Paste this into your agent and replace the two links with the ones from your email:
Install Idlerate on this Windows PC. Use PowerShell. Don't use an administrator prompt. Before each download or command, show me exactly what you will run and wait for me to say yes. Stop if I say no.
1. Download these two files to a new temp folder:
idlerate.exe: PASTE_EXE_LINK_FROM_EMAIL
checksums.txt: PASTE_CHECKSUM_LINK_FROM_EMAIL
2. Check the sha256 of idlerate.exe against the idlerate.exe line in checksums.txt. Stop if it doesn't match.
3. Ask me before installing. Tell me that the installer will also install Ollama and download a starter model of several GB if they are missing. After I say yes, run: .\idlerate.exe install
4. Follow the IDLERATE_PROGRESS lines and tell me each step in one short line. If Ollama or a model download is in progress, tell me the percentage now and then.
5. When you see the IDLERATE_INSTALL_OK line, read its JSON. Show me pairing_code and the model. Don't sign in to anything for me and don't ask for my password.
6. Don't run any other installer, change any other settings, or retry with different options without asking me first.
If the PC already has Ollama and the model you want, add --no-ollama to step 3. To choose the starter model yourself, add --model qwen2.5:7b (or another Ollama tag).
2. What the agent runs
idlerate install does the following, all as your normal Windows user:
- Copies itself to
%LOCALAPPDATA%\Idlerate\idlerate.exe. - Registers a per-user logon task named Idlerate and starts it. Nothing is installed as a system service.
- Points the PC at
https://relay.idlerate.dev. - Checks for Ollama on
http://127.0.0.1:11434. If it’s missing, it downloads a pinnedOllamaSetup.exefrom Ollama’s GitHub releases, checks its sha256, and runs it silently. If that fails, it falls back towinget install Ollama.Ollamafor the current user. - Pulls one starter model, picked by GPU memory:
| Available GPU memory | Model pulled |
|---|---|
| Under 4 GiB (or CPU only) | llama3.2:1b |
| 4 GiB up to under 8 GiB | llama3.2:3b |
| 8 GiB up to under 14 GiB | qwen2.5:7b (Q4) |
| 14 GiB and above | qwen2.5:14b (Q4) |
If the GPU memory can’t be read, it picks llama3.2:1b. --model overrides the table. --no-ollama skips the Ollama check and the model pull.
3. How it reports progress
Each step prints one line that starts with IDLERATE_PROGRESS followed by JSON. Agents can parse these lines:
IDLERATE_PROGRESS {"step":"ollama","status":"checking","detail":"http://127.0.0.1:11434"}
IDLERATE_PROGRESS {"step":"ollama","status":"downloading","completed":52428800,"total":734003200}
IDLERATE_PROGRESS {"step":"ollama","status":"verified"}
IDLERATE_PROGRESS {"step":"model","status":"chosen","model":"qwen2.5:7b"}
IDLERATE_PROGRESS {"step":"model","status":"pulling","model":"qwen2.5:7b","completed":1048576,"total":4683087332}
IDLERATE_PROGRESS {"step":"install","status":"task","detail":"Idlerate"}
IDLERATE_PROGRESS {"step":"pairing","status":"waiting"}
| Field | Meaning |
|---|---|
step |
ollama, model, install, or pairing |
status |
For example checking, missing, downloading, verified, installing, fallback, ready, chosen, pulling, copying, task, starting, waiting, skipped, error |
detail |
Extra text, such as a path, or an error message when status is error |
completed, total |
Download progress in bytes |
The last line is IDLERATE_INSTALL_OK followed by a JSON summary. It includes pairing_code, paired (true when this PC was already paired), pairing_pending, model, ollama_installed, exe, and relay. It never includes the machine’s secret.
If the installer fails, it prints the error and exits with code 1. You won’t see IDLERATE_INSTALL_OK. Ask the agent to show you the error text.
4. Pair the machine
- The agent shows you the pairing code from the install output. It looks like
ABCD-EFGH. The output also has aPairing URL:line that opens the pairing page. - Sign in at relay.idlerate.dev yourself, enter the code under Pair machine, and submit.
- The machine appears as a card on your dashboard. The dashboard shows a key for it once. Copy it if you want to use it, or create one later under Connect a new agent.
The code is short-lived. If it expires, or the install output says the code wasn’t ready yet, ask the agent to read the newest code from %LOCALAPPDATA%\Idlerate\pairing.txt. Idlerate writes the current code there. Running .\idlerate.exe install again also works.
You confirm the pairing in your own browser session, so the agent never needs your password or dashboard login.
Planned improvements (in a pending update, not live yet):
- The pairing code will also show in the Idlerate tray menu.
- The pairing code won’t be stored on disk. Today it is also written to
%LOCALAPPDATA%\Idlerate\pairing.txt.
5. Check that it works
Create a key under Connect a new agent and add the relay to your agent as described in Connect agents and bots. Then paste:
Use the Idlerate MCP server. Call list_models and show me each machine with its status and models. Then call run_model with the smallest model on this PC and the prompt "Reply with one word: ready". Wait with get_status (wait_ms 25000) until it finishes, then show me the get_result output. Do not print the API key.
You should see this PC listed as connected and a one-word answer. If run_model returns subscription_required, finish signup and email verification to start your trial. Your account needs an active trial or plan to run jobs.
Safety notes
- The prompt above makes the agent show you each command and wait for your yes before it runs it. Don’t use an agent mode that runs commands without asking.
- The agent only runs the Idlerate installer, plus the Ollama installer when Ollama is missing.
- Nothing needs administrator rights. Ollama’s own installer runs for your user.
- The installed app only makes outbound connections and only runs inference on installed models. See Security.
- Don’t paste API keys into chat. Ask the agent to keep them in its MCP config or an environment variable.
Uninstall
& "$env:LOCALAPPDATA\Idlerate\idlerate.exe" uninstall
This removes the logon task, the program, and the pairing and log files. It keeps config.json, so a reinstall reconnects as the same machine. It doesn’t remove Ollama or your models. To drop the machine from your account, remove its card on the dashboard.
Not supported yet
- No public download link. We’ll email your download link as soon as you verify your email.
- Pairing needs you. No step lets an agent confirm a machine on your account without your signed-in dashboard.
- Windows only in the preview. Mac and Linux clients are planned. There is no preview download for them yet.