Sign in Subscribe
news

JetBrains Thinkrail

Sebastien Dubois

Sebastien Dubois

— 4 min read

Canonical version: JetBrains Thinkrail.

Thinkrail is an open-source agentic IDE from JetBrains, built around Pi (the coding agent by Mario Zechner). It's a JetBrains incubator project, licensed under Apache 2.0. The README pitches it as "the agentic IDE that gets better every time you use it", and the tagline is "Vibe code with pi in a lightweight, real IDE that customises itself around the way you work."

Why does it matter? Because JetBrains didn't write its own agent this time. Thinkrail runs Pi in-process and leaves models, skills, compaction, costs and session state to Pi. Thinkrail itself only handles the workspace, the editor and the UI, and the README says it doesn't assemble prompts behind your back. It even reads your existing Pi configuration (~/.pi/agent). So if you already use Pi in a terminal, Thinkrail is a GUI on top of the same setup, not a new tool to configure.

What you get

  • One workspace per task. A project is a git repository, and each workspace is a Git Worktree with its own branch and directory, so several agents can work in parallel. Finished or dormant workspaces go to a per-project shelf
  • Several Pi sessions per workspace, each with its own model and a live token and cost readout. You can steer a running turn, queue follow-ups and delegate to subagents. A shared plan can be edited by you and the agent, and each completed step can be committed and reviewed on its own
  • Spec-first development. A graph of SPEC.md files sits next to the code (one per module), and the agent reads and updates it through spec_* tools. When a change touches a boundary or a contract, the spec is updated in the same change
  • A review flow. You leave anchored comments on files and diffs, then send them to a chat as structured context. Proposed UI changes come with before/after mockups that you comment on like a diff
  • Git and PR integration. Branch status against origin, publish/push, open a PR, then follow CI checks and review decisions, merge, reopen or delete remote branches
  • A real (but small) IDE: Monaco editor tabs, a file tree with git change marks, a "Changed" view, terminals and documents, all scoped to the active workspace
  • Skills and MCP. Portable AI Agent Skills are read in place. MCP servers come from Pi's mcp.json or the settings; calls to servers that aren't marked read-only need a confirmation in the chat, and servers defined by a repository stay off until you approve them (see AI Agent Permissions)
  • Extra agent tools: web research, multiple-choice questions to the user, and inline diagrams and previews. The question and visualization tools are also published as standalone Pi packages (@thinkrail/pi-ask-user-question, @thinkrail/pi-visualize)

There are no Thinkrail accounts. You sign in with JetBrains AI, or connect any provider through Pi's own auth. Full extension support is announced as the next step.

Tech stack and install

Bun runs a Bun.serve HTTP/WebSocket host with one in-process Pi session per chat (Pi itself needs Node.js 22.19 or later). The UI uses React 19, Zustand and Tailwind v4, the editor is Monaco, and the desktop app is built with ElectroBun. Client and host talk through a typed, versioned protocol.

The desktop app runs on macOS 14+, Windows x64 and Linux x64/ARM64 (glibc 2.38+, so Ubuntu 24.04 or newer). The thinkrail CLI installs with:

curl -fsSL https://raw.githubusercontent.com/JetBrains/thinkrail/main/install.sh | bash
thinkrail ~/code/my-repo

Two things to know. Usage analytics (PostHog, tied to a random installation ID) are on by default; the README says they never include prompts, code, files or credentials, and you can turn them off with --no-analytics or THINKRAIL_NO_ANALYTICS=1. And the macOS CLI is signed but not notarized, while the Linux artifacts are unsigned.

Reception on Hacker News

The repo reached the Hacker News front page around 2026-10-26 (242 points, 174 comments). Most of the thread was about JetBrains, not about Thinkrail itself.

  • The top thread attacked the README as AI-written. exizt88: "I refuse to take a repo with an AI-generated readme seriously" (Pangram rated it 100% AI). hoppp replied "I am fine with AI readme as long as it has what I need", and davidbanham argued that an unchecked README shows how much the author cared
  • Many questioned JetBrains' AI strategy. dr_faustus: "Their AI strategy is all over the place". saghm recalled that Fleet was shut down, and xentripetal claimed Air was Fleet "just renamed" with more agent workflows. I couldn't confirm that last point. throwaw12 went further: "It feels like JetBrains is dying, and they're trying very hard to survive"; MBCook found the thread too negative
  • Several said they now use Claude Code, Codex or Zed instead of a JetBrains IDE, while NBJack said JetBrains' navigation and code intelligence still have no equal
  • Deukhoofd said the worktree support in JetBrains IDEs is poor, which is ironic for a product built on worktrees
  • Others questioned the choice of Monaco and Bun, and asked about local models and OpenAI-compatible APIs (Pi already covers those)

My take

What I find interesting here is the choice of Pi. JetBrains already has Junie and Air, and still built this one on someone else's open-source agent. To me, that says the agent loop is becoming a commodity and the IDE vendors now compete on everything around it, from workspaces and git integration to code review and specs. Spec-first development is the part I'd test first. A SPEC.md per module, maintained by the agent in the same change as the code, is a good answer to context rot... if the agent really keeps it up to date. But JetBrains' track record with new products (Fleet, now Air and Thinkrail side by side) is a fair reason to wait before you bet a workflow on it.

References

About Sébastien

Ready to get to the next level?

Found this valuable? Share it with someone who needs it.

Join 6,000+ readers. Get practical systems for knowledge & AI. Free.

Subscribe ✨

Free: Knowledge System Checklist

A clear roadmap to building your own knowledge system. Subscribe and get it straight to your inbox.

6,000+ readers. No spam. Unsubscribe anytime.

Subscribe