Thinkrail: An IDE That Builds Itself Around You
JetBrains ships thinkrail, an Apache-2.0 TypeScript IDE built around the pi agent, with install scripts, an e2e suite, and 5,130 stars.
TL;DR: Thinkrail is a JetBrains project that wraps the pi coding agent in a real IDE and lets the agent reshape that IDE around your working style. It is Apache-2.0, written in TypeScript, and sits at 5,130 stars with 243 open issues.
What it is
The description is unusual for a JetBrains repository: “Vibe code with pi in a lightweight, real IDE that customises itself around the way you work.”
Two claims sit in there. The first is that it is a real IDE rather than an editor plugin. The second is that the IDE customises itself, which is the part that deserves unpacking, because that is a design commitment rather than a feature description.
The agent at the center is pi, and the repository structure confirms it: there is a .pi/ directory at the root, and pi is the agent project this blog covered back in September.
What the tree tells you
.pi/
apps/
e2e/
design-audit/
.changeset/
.husky/
architecture.md
central-integration.md
goal-and-requirements.md
install.sh
install.ps1
package.json
bun.lock
biome.json
AGENTS.md
Three of those files are worth reading for what they reveal about how the project is run.
install.sh and install.ps1 at the root mean the install path is a script, and shipping a separate PowerShell version means Windows is a first-class target rather than something a contributor added later.
goal-and-requirements.md and architecture.md are design documents kept in the repository itself, which is unusual and useful. A project that ships its own reasoning has already made most of the decisions you would otherwise have to reverse engineer.
design-audit/ and e2e/ are the directories that suggest this is treated as a product. A design audit folder is not what a script collection contains.
The toolchain is Bun, not npm: bun.lock is present, and there is no package-lock.json or yarn.lock.
Why a vendor publishing this matters
JetBrains is a first-party commercial vendor, and this is not a side project from an individual contributor. Their IDE products are the incumbent in this category, so a repo like this is either a prototype for a future product line or an experiment on the edges of the main one.
Either way, the signal is different from a hobby project. The open issue count supports the reading: 243 open issues on 5,130 stars is a lot of people filing structured feedback against a real interface.
| Metric | Value |
|---|---|
| Stars | 5,130 |
| Forks | 267 |
| Open issues | 243 |
| Primary language | TypeScript |
| License | Apache-2.0 |
| Created | 2026-07-08 |
| Last push | 2026-10-29 |
How to try it
git clone https://github.com/JetBrains/thinkrail.git
cd thinkrail
ls install.sh install.ps1
The install scripts are at the root, and the project site at thinkrail.ai is linked from the repository metadata. There is no npm install line here because the actual setup path is those scripts, and their flags are documented in the repository rather than guessable from the file names.
When to use it
If you want the loop of an agent that can reshape its own environment rather than one that is fenced into a fixed sandbox, this is the clearest public example of that idea built by a vendor with an interest in shipping it.
The design documents are independently useful. architecture.md and goal-and-requirements.md are the kind of files that tell you what a team decided and why, which is rare to find published at all.
When not to use it
Two hundred and forty-three open issues on a project that is three and a half months old means the rough edges are visible and accumulating quickly. The star count grew by 1,068 today, so interest is running far ahead of stability.
Also note the repo is TypeScript with Bun, not the JVM stack JetBrains is known for. If you expected IntelliJ or Fleet, this is a different thing entirely.
Source and Accuracy Notes
- Project page: https://thinkrail.ai/
- Source repository: JetBrains/thinkrail (verified via GitHub API
html_url) - License: Apache-2.0 (verified via GitHub API
license.spdx_id) - Stars at time of writing: 5130
Related Posts
ai-setup
Recall – Persistent Memory for Claude Code via MCP Hooks
Recall gives Claude Code a permanent memory store that survives session restarts and context compaction. Four hooks capture and restore context automatically — with cloud SaaS or self-hosted options.
2/28/2026
dev-tools
Automotive Skills Suite for AI Engineering
Evaluate Automotive Skills Suite for APQP, ASPICE, HARA, safety-plan, and DIA workflows with setup notes, governance risks, and SME review guidance.
5/28/2026
dev-tools
awesome-agentic-ai-zh Roadmap Guide
Explore awesome-agentic-ai-zh as a Chinese agentic AI learning roadmap, with setup notes, track selection, study workflow, and evaluation guidance.
5/28/2026