Contents

ThinkRail is a lightweight IDE built around pi. It separates tasks with Git worktrees and takes you from chat to editing, review, and PRs using project specifications and skills.

Key Points#

  • Create Git worktrees for each project to run multiple tasks in parallel on separate branches.
  • Keep the Monaco editor, file tree, terminal, and diff review alongside chat in one workspace.
  • Use SPEC.md and Agent Skills to reference project intent and procedures across sessions.
  • Reuse pi settings and credentials to connect to JetBrains AI, supported providers, or your own model server.
  • Desktop and CLI versions are available for macOS, Windows, and Linux.

Separate Tasks with Worktrees#

In ThinkRail, you open a repository as a project and create task workspaces as Git worktrees. Each has its own branch and working directory, so you can run multiple agents without directly changing your usual checkout. You then merge the branches you want to keep and delete those you no longer need.

You can also add existing worktrees or work directly in the project folder. Inactive workspaces are tucked away in a per-project list and can be restored when you resume. You can also check branch status and PRs, push changes, and create PRs from the interface.

From Chat to Editing and Review#

Each chat uses a pi session, with model, token count, and cost information available. You can add instructions without waiting for a response or queue your next request. Plans can be edited together with the agent, and each completed item can be committed and reviewed.

The workspace can display the Monaco editor, file tree, changed-file list, terminal, and documents. You can comment on files or diffs and pass those comments to chat, reducing the need to explain where changes are needed. You can also open a worktree in your usual IDE.

Carry Specifications and Skills Across Sessions#

A specification graph connecting the project's SPEC.md files provides design information for the agent. Dedicated tools let it read, search, and update specifications; the IDE's specification tree view is read-only. This keeps project intent close to the code across sessions.

Agent Skills used with other coding agents can also be loaded as they are. ThinkRail includes workflows for project setup, design brainstorming, PR creation, and more. Extensions and MCP servers can also be configured, but repository-defined MCP servers are disabled by default. Calls to servers not designated as read-only require confirmation in chat.

What to Check Before Getting Started#

The desktop version supports macOS 14 or later, Windows x64, and Linux x64/ARM64. The Linux version has dependencies including glibc 2.38 or later. The CLI requires Git and a model provider; you configure the provider and project on first launch. You can also start by creating an empty Git repository.

ThinkRail loads existing ~/.pi/agent settings and uses authentication for JetBrains AI or pi-compatible providers. You can also connect your own OpenAI-compatible server, and registered models appear in the pi CLI as well. ThinkRail has no separate account system.

Analytics include basic events such as app launches and chat creation. The documentation states that prompts, code, files, and credentials are excluded from analytics events. Sharing additional usage data can be toggled in settings. Before getting started, check your target OS, model connection method, and organization's privacy requirements.

Next Steps#

Check the ThinkRail repository on GitHub for terms of use and the latest releases, then choose the installer for your OS. To try the CLI, start with the official installation instructions.

Source: https://github.com/JetBrains/thinkrail

Contents