A ThinkRail-branded desktop-and-mobile client for the pi coding agent. ThinkRail is a thin host that runs pi in-process and bridges it to a rich, mobile-first UI — pi owns models, skills, compaction, cost, and session state; the app owns the workspace, the editor, and the wire.

Website: thinkrail.ai — a landing page that is the IDE, its blog, and the vibecoder-focused experience (see apps/website).

V1 is a Worktree IDE: open a git repo as a project, spin up workspaces as git worktrees (each its own branch and cwd), and work across a tabbed Monaco editor, git Changes view, terminals, a read-only spec-graph viewer, and multiple concurrent pi chat sessions — all scoped to the active worktree.

ThinkRail ships in two additive forms: an unsigned native desktop installer and the self-contained thinkrail CLI, which opens the same app in your browser. Both embed the same in-process agent host and are published with SHA256SUMS on the releases page.

Download the matching thinkrail-desktop-* asset: a DMG for macOS Apple Silicon, a setup ZIP for Windows x64, or a setup tarball for Linux x64/ARM64. These initial desktop builds are unsigned, so the operating system may ask you to confirm the first launch.

Linux desktop builds require Ubuntu 24.04 or another glibc 2.38+ distribution with GTK 3, WebKitGTK 4.1, Ayatana AppIndicator 3, and librsvg 2. On Ubuntu 24.04:

sudo apt install libgtk-3-0 libwebkit2gtk-4.1-0 libayatana-appindicator3-1 librsvg2-2

The CLI installer downloads the right binary, verifies its SHA-256 checksum, and puts thinkrail on your PATH.

macOS / Linux (also Windows under Git Bash):

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

Windows — the same command works from cmd and PowerShell:

powershell -c "irm https://raw.githubusercontent.com/JetBrains/thinkrail/main/install.ps1 | iex"

Nightly builds and pinned versions:

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/JetBrains/thinkrail/main/install.sh | bash -s -- --channel nightly
curl -fsSL https://raw.githubusercontent.com/JetBrains/thinkrail/main/install.sh | bash -s -- --version 0.2.0

# Windows — options are env vars (THINKRAIL_CHANNEL, THINKRAIL_VERSION, THINKRAIL_PREFIX, THINKRAIL_NO_MODIFY_PATH)
$env:THINKRAIL_CHANNEL='nightly'; irm https://raw.githubusercontent.com/JetBrains/thinkrail/main/install.ps1 | iex   # PowerShell
set "THINKRAIL_VERSION=0.2.0" && powershell -c "irm https://raw.githubusercontent.com/JetBrains/thinkrail/main/install.ps1 | iex"   # cmd

Then run thinkrail (add a git repo path to open it as a project: thinkrail ~/code/my-repo). To update later, run thinkrail update on any platform — it re-runs the installer for your channel (on Windows it replaces the running thinkrail.exe in place). To remove it, run thinkrail uninstall: it takes out the executable, the PATH entry the installer added, and the install metadata, and asks whether to delete your ~/.thinkrail app state (kept by default — pass --remove-data to delete it, -y to skip the questions). thinkrail --help lists the flags; thinkrail --version prints the build.

Prebuilt platforms: macOS (Apple Silicon), Linux arm64 + x64, Windows x64 (.exe). Intel macOS isn't prebuilt — use Apple Silicon or build from source.

Prefer a manual CLI install? Download a binary + SHA256SUMS from the releases page, verify the checksum, chmod +x, and move it onto your PATH.

Runtime prerequisites: git on PATH, and an authenticated pi provider (the agent runs against your real provider credentials). App state lives under ~/.thinkrail.

• Bun ≥ 1.3 (the package manager and runtime)
• Node.js ≥ 22.19 (required by the in-process pi engine)
• An authenticated pi provider (the agent runs against your real provider credentials)

git clone <repo-url>
cd thinkrail
bun install
bun run dev

bun run dev boots the host and the web client together. Press Ctrl+C to stop.

To run the V1 launchers:

bun run --filter @thinkrail/cli dev  # browser launcher
bun run build:binary                 # standalone CLI artifact
bun run desktop:dev                  # package and open the Electrobun app
bun run desktop:build                # package without opening it

On-disk app state (projects, workspaces, worktrees) lives under ~/.thinkrail.

• Engine host — packages/server (+ packages/shared), launched by apps/cli or apps/desktop. createServer() is a Bun.serve HTTP+WS host with an AgentSessionManager (one in-process pi AgentSession per tab).
• The wire — packages/contracts: the typed, versioned protocol (types-only).
• UI client — apps/web: mobile-first React 19 + Zustand + Tailwind v4, ships independently and dials a host over the wire.

The engine is pi only, run in-process via @earendil-works/pi-coding-agent. apps/web depends on packages/contracts only — never on the server — which is what makes the UI shippable on its own.

See goal-and-requirements.md and architecture.md for the canonical product and design specs.

apps/
  cli/        V1 entrypoint: boot host + open browser
  web/        mobile-first UI client
  desktop/    Electrobun local-host launcher + native packaging
  website/    public landing + blog + vibecoding site (Cloudflare Pages)
packages/
  server/     createServer(): Bun.serve + AgentSessionManager
  contracts/  the wire (types-only)
  shared/     server-side helpers (shellEnv, freePort)
  spec-graph/ portable pi extension: spec_* tools + skill

Fast gates (also the husky pre-commit hook):

bun run lint        # biome
bun run typecheck   # tsc across all packages
bun run test        # unit tests (root tooling + each package)

End-to-end tests drive the real web UI against isolated hosts. The no-agent gate builds once and uses a machine-adaptive number of independent shards (half the available CPUs, capped at eight):

bunx playwright install chromium                    # one-time
bun run e2e                                         # complete no-agent gate
bun run e2e -- e2e/changes.spec.ts                  # focused iteration
bun run e2e -- --last-failed                        # repair loop
bun run e2e:serial                                  # one-host debugging fallback
bun run e2e -- --shards=12                          # explicit 1–16 override
bun run e2e:binary                                  # packaged CLI host (build first)
bun run e2e:desktop                                 # packaged desktop host (build first)
bun run e2e:full                                    # everything; needs pi auth
bun run e2e:agent                                   # only @agent; remains serial

ThinkRail is developed spec-first: hierarchical, interconnected specs live in the repo alongside the code — top-level specs at the root (goal-and-requirements.md, architecture.md) and a co-located SPEC.md for every module. When you change a boundary, contract, or decision, update the corresponding spec in the same change. See AGENTS.md for the spec workflow.

ThinkRail sends anonymous usage analytics to PostHog (EU cloud; on by default; a notice is printed the first time anything is sent). The data answers product questions — how many installs are active, on which versions/platforms, which models and providers get used, and which features matter — and nothing more.

This applies to every way of running ThinkRail, including a build you compiled yourself and a run straight from a source checkout — each is reported as what it is (see channel and build below) rather than kept silent. Automated runs never send: anything under CI, bun test, and the e2e suites are all muted, so test traffic can't be mistaken for a person.

The only stable identifier is a random per-install id (a uuid4) minted on your machine and stored in ~/.thinkrail/installation.json; it never leaves the host except as the anonymous distinct_id on events. Events additionally carry only low-cardinality, non-personal metadata: app version, release channel (stable/nightly/dev), how the code was built (desktop, binary for a compiled CLI, or source for a repo checkout), OS (macos/linux/windows), architecture (x64/arm64), and — on chat/login events — the model/provider name only if it is a pi built-in (anything user-configured is reported as custom). Message activity is counted as a bare send event carrying only how it was sent (prompt/steer/follow_up) — never the message itself. Events are sent personless (no person profiles are ever built) and with GeoIP lookup disabled.

Never collected: file paths or names, prompts, code, chat transcripts, API keys, token counts, hostnames, usernames, or IP-derived fields.

Contributors: running bun run dev or the CLI from a checkout reports too, tagged channel=dev / build=source. Your test and CI runs don't (they're muted, as above), and if you'd rather not report at all, either switch it off in-app or export THINKRAIL_NO_ANALYTICS=1 in your shell.

Turn it off any time:

• In-app: Settings → Privacy → toggle off (saved on the host, synced to every client).
• Per run: thinkrail --no-analytics (or THINKRAIL_NO_ANALYTICS=1) — mutes that run without touching the saved setting.

Turning analytics off stops all sending immediately; the install id is kept (never rotated) and simply goes unused until you turn it back on.

Contributions are welcome — see CONTRIBUTING.md. This project and community are governed by the Code of Conduct.

Licensed under the Apache License 2.0.