ThinkRail is an open-source (Apache 2.0) "agentic IDE" from JetBrains InnovationHub, the company's incubator. It is not a plugin for IntelliJ IDEA or other JetBrains IDEs but a separate, lightweight app built around one coding agent, pi: pi handles models, skills, context compaction, cost and sessions, while ThinkRail supplies the editor, terminals, diffs and workspace management. It runs on macOS, Windows and Linux, as a desktop app or through a thinkrail CLI that opens the same UI in a browser.
Why it is trending now. The project was introduced in August, but it took off in late October: a Hacker News post on 25 October reached 242 points and 174 comments, and the repository went from roughly 1,800 stars on 26 October (per Oflight) to about 7,100 now, with around 5,700 gained in a week on GitHub trending.
How it works. Each task gets its own git worktree and branch, so several agents can work in parallel without mixing changes. You get Monaco editor tabs, a file tree with change marks, terminals, review comments you can send to the agent, several concurrent pi chats per workspace with a per-chat model choice and live token/cost readout, and a graph of SPEC.md files the agent reads and keeps up to date. It reads existing Agent Skills and uses MCP servers from pi's config, asking for approval before calls that are not read-only. Models come from a JetBrains AI sign-in, any provider pi supports, or an OpenAI-compatible server such as Ollama, vLLM or LM Studio.
What 0.3.0 adds (29 October). Shipping a branch from the top bar: publish, push or sync (it shows the git pull --rebase it will run and "never force-pushes"), open a PR or GitLab MR with an agent-written description, watch checks and merge (PR actions need the gh CLI). A first-run wizard detects accounts already signed in to Claude Code, Codex, Copilot, Kimi CLI or Grok Build, and the model list now shows only models your account can use. Japanese, Spanish and Simplified Chinese are built in (agent replies and tool output are not translated). Also new: a visualize tool for HTML mockups at desktop/tablet/phone widths, subagent costs counted in the chat total, and an Intel Mac build.
Still early. Versions are pre-1.0 (README: stable 0.1.2, nightly 0.2.0), it works only with pi, some macOS builds may be blocked by Gatekeeper and the macOS CLI is not notarized. It sends basic usage events and error reports tied to a random installation ID; this can be turned off. ThinkRail is free, but model usage is billed by your provider. Reviewers suggest trying it alongside your current tools rather than for production work.
