ThinkRail itself is free and open source (Apache-2.0, from JetBrains' InnovationHub). There's nothing to pay JetBrains and no ThinkRail account. The model is on you: until you connect one, the app says plainly, "No model provider connected, the agent can't run."
That connection can be a subscription you already have: the provider screen offers sign-in for existing Claude, ChatGPT, Copilot, and other plans with no API key. Or an API key from one of 32-plus providers. Or JetBrains AI, where usage is billed as credits at public API rates. The README says each chat shows a live token and cost readout; we didn't connect a provider, so we never saw an amount.
A computer on its platform list: the desktop app runs on macOS Apple Silicon, Windows x64, or Linux (Ubuntu 24.04 or newer). There's no Intel Mac desktop build, and no prebuilt CLI for Intel Macs either. Every platform can also run the thinkrail CLI, which opens the same app in your browser. For the local-project flow you need git installed and a git repository to open.
Hands-on status: we built ThinkRail from source and ran it on October 25. The UI loaded, the telemetry prompt appeared, the provider screen listed subscription sign-ins and 32-plus API-key providers, and cutting a workspace really created a git worktree on a new branch. We didn't connect a provider or run the agent, so agent chats are vendor-documented, not tested by us.