Userouter Code

Your agent,
on our models.

Code, edit and ship from the terminal or your editor on every model Userouter hosts on its own GPUs. One command sets it up, and ROUTER stakers get a free daily allowance.

curl -fsSL https://userouter.xyz/install.sh | sh

Set up by hand

$ curl -fsSL userouter.xyz/install.sh | sh
Userouter API key: ••••••••••••
✓ key ok
✓ 6 hosted models
✓ wrote ~/.config/userouter/opencode.json
✓ installed userouter-opencode
$ userouter-opencode

Wherever you work.Same models, same key, same credits.

Terminal

OpenCode and Aider read and edit your project, run commands and commit, on Userouter models.

Editor

Continue, Cline and Roo Code live in VS Code and JetBrains. Chat, edit and apply across files.

Your own agent

One OpenAI-compatible endpoint with tool calling. Bring any SDK or framework.

Pick your tool.The config below lists the models that are live right now.

01Install OpenCode: curl -fsSL https://opencode.ai/install | bash
02Export your key: export ROUTER_API_KEY=sk-userouter-…
03Save this file, run opencode, and pick a Userouter model with /models.

Model

Loading the models Userouter hosts…

Need a key? Create one in your account · Tool docs at opencode.ai

Userouter Code app (beta).Our own terminal app, built for hosted and confidential models. It installs as userouter-code and updates itself with userouter-code update.

macOS and Linux

curl -fsSL https://userouter.xyz/install-cli.sh | sh

Windows (PowerShell)

irm https://userouter.xyz/install-cli.ps1 | iex

Native app

One binary for macOS, Linux and Windows, checked against its sha256. Signs you in through the browser.

Confidential models

Pick a tee/ model (marked 🔒) and every message is encrypted to an attested enclave. Userouter relays ciphertext.

Proof on every step

Each answer is shown only after the enclave's signed receipt matches the exact request sent and bytes received.

Questions.

Which models can I use?

Models launched on the Userouter launchpad and live on our own GPUs (ids starting with launch/) that passed our tool-call check. The list on this page and the installer both read GET /api/v1/models and keep the ones with "tools" in supported_parameters, so a model that graduates and passes shows up on its own.

What does it cost?

The launch model's price per 1M tokens, paid from your Userouter credits. ROUTER stakers get a free daily allowance on hosted models, and a launch token's stakers use that launch's model for free.

What does the installer do?

It asks for your API key, checks it, lists the hosted models, and writes an OpenCode config plus a userouter-opencode command into ~/.config/userouter and ~/.local/bin. It never touches your existing configs or shell profile. Read it first at userouter.xyz/install.sh.

Do tools and file edits work?

Yes. Every listed model emitted a real tool call when it went live, and tools, tool_choice and parallel_tool_calls go straight to it. Smaller models follow long agent prompts less reliably, so start with the biggest one for multi-step work.

Why does my config stop working after an upgrade?

A model id pins the exact weights (the part after @). When a builder upgrades their model the id changes. Run the installer again, or copy the fresh config from this page.

Your agent,on our models.