JetBrains' Free ThinkRail IDE Shelves Each Workspace on Its Own When Its Pull Request Merges or After a Quiet Week

The README describes ThinkRail as “a rich interface around the pi coding agent.” You open any git repository as a project and cut workspaces from it, each “with its own branch and working directory.” Agents work in parallel, you “merge the good branch, delete the rest,” and the finished ones clean themselves up.

That shelf is the detail that fixes a real annoyance. Anyone who has run several agents on several branches knows the sidebar fills with half-dead experiments. Here, “Finished and dormant workspaces settle onto a per-project shelf on their own — when their pull request merges or closes, or after a quiet week — and come back the moment you work in them.”

Around that sits a real editor: Monaco tabs, a file tree with live git marks, terminals, and review comments you can anchor to a diff and send to a chat as context. The topbar shows where your branch stands against origin and its pull request, with “the next step one click away: publish, push, or open a PR.” Once the PR is open, it shows CI checks and lets you hand failing checks to the agent.

The part JetBrains leans on hardest is memory. “A connected graph of SPEC.md files lives beside the code,” and the agent reads and maintains it through its own spec_* tools “so intent survives the session.” It also reads the Agent Skills you already keep for other coding agents, connects MCP servers, and asks before calling any server tool that isn’t marked read-only.

You bring your own model. You can sign in with JetBrains AI or any provider pi supports, or point it at a local server such as Ollama, LM Studio or llama.cpp, and “ThinkRail has no accounts of its own.” It does send basic usage events and error reports tied to a random installation ID, and the README says those “never include prompts, code, files, credentials, or account identity.”

The repository’s description promises an IDE “that customises itself around the way you work.” Read the README closely and most of that is future tense. Today it learns your project “through living specs,” and “Next, we’re adding full extension support, so ThinkRail can create the tools it needs.” The v0.3.0 release is more practical: translated menus, GitHub review actions in the topbar, and Windows updates that actually install.

I like that this starts from the boring problem. Everyone is building agents that write more code, and fewer people are building the place where you sort through what they wrote. A shelf that empties itself when a PR merges is a small idea, and it’s the kind of small idea that decides whether you keep using a tool after the first week. The self-customizing pitch can wait for the extension work to land. The workspace handling is useful right now.