ThinkRail
ThinkRail is a lightweight, adaptable agentic IDE built on the pi coding agent. It opens git repos as projects, creates isolated git-worktree workspaces, and pairs agent chats with Monaco tabs, terminals, diffs, and living specs.
Visit JetBrains/thinkrailOverview
ThinkRail is an agentic IDE from JetBrains InnovationHub built around the pi coding agent. It treats any local git repository as a project and creates isolated git-worktree workspaces so multiple agents can work in parallel without touching the main checkout. The workbench combines Monaco editor tabs, a files tree, live git diffs, terminals, and documents with concurrent pi chat sessions. Project intent lives in a connected graph of SPEC.md files that the agent can search and maintain through dedicated tools.
Key Features
- Isolated workspaces: any local git repository can become a project, and workspaces are cut as git worktrees with their own branch, files, terminals, and chats.
- Parallel agent work: several pi sessions can run per workspace with per-session models, live token and cost readouts, steering, queued follow-ups, subagents, and a shared editable plan.
- Real IDE workbench: Monaco editor tabs, a files tree, live git Changes diffs, real PTYs for terminals, documents, and anchored review comments scoped to the active workspace.
- Spec graph as ground truth: SPEC.md files with frontmatter form a connected graph beside the code, and the agent uses dedicated spec tools to search, navigate, and maintain it.
- Skills and workflows: portable Agent Skills are read in place, and bundled workflows guide project setup, design brainstorming, and shipping a pull request through the user's gh.
- MCP and provider flexibility: MCP servers can be connected with approval controls, and users sign in with JetBrains AI or connect any provider through pi auth.
Use Cases
- Running multiple coding agents in parallel on separate worktree branches while keeping the main checkout clean until a branch is merged.
- Developing spec-first, with the agent reading and maintaining a project intent graph so decisions survive a chat session.
- Reviewing agent-made changes through live diffs, terminals, and structured review comments before merging.
- Extending agent workflows with existing Agent Skills, MCP servers, and bundled project setup or pull-request workflows.
Getting Started
- Install the desktop app from the releases page, or use the platform installer to put the thinkrail CLI on the PATH.
- Run thinkrail, optionally with a git repository path, to open the browser UI or desktop app.
- Sign in with JetBrains AI or connect your own provider credentials through pi auth.
Deployment & Requirements
- macOS desktop builds support Apple Silicon only; there is no Intel desktop build.
- Windows desktop builds are x64 and distributed as a setup ZIP.
- Linux desktop builds for x64 and ARM64 need glibc 2.38 or newer with GTK 3, WebKitGTK 4.1, Ayatana AppIndicator 3, and librsvg 2, such as Ubuntu 24.04 or newer.
- The thinkrail CLI requires git on the PATH and a model provider, either JetBrains AI sign-in or your own provider credentials.
Before You Adopt
- License: Apache-2.0. Review its terms before using, modifying, or distributing the project.
- Linux desktop and CLI artifacts are unsigned.
- The macOS CLI is signed but not notarized.