neuron-explorer
Qwen3.5‑4B 전용 조향형 뉴런 탐색기, 모든 MLP 뉴런을 밀어 결과를 보관·탐색
Jean: 노트북·서버에서 격리된 작업 공간에 AI 에이전트를 병렬로 실행
Run AI agents in parallel in isolated workspaces - on your laptop or a server. Connect from a browser or the native app. Works with Claude, Codex, OpenCode, Cursor, Pi, Command Code, Grok, and Kimi Co
개발 도구TypeScript · RustGitHub @coollabsio온라인운영자 미확인
써 보셨어요? 한 번만 눌러 주세요. 로그인은 필요 없어요.
A dev environment for AI agents.
저장소 README에서 가져옴
README 전체를 불러오는 중입니다. 아래는 저장된 발췌입니다.
A desktop AI assistant for managing multiple projects, worktrees, and chat sessions with Claude CLI, Codex CLI, Cursor CLI, OpenCode, PI, Command Code, Grok, and Kimi Code.
Jean is an opinionated native desktop app built with Tauri that gives you a powerful interface for working with Claude CLI, Codex CLI, Cursor CLI, OpenCode, PI, Command Code, Grok, and Kimi Code across multiple projects. It has strong opinions about how AI-assisted development should work - managing git worktrees, chat sessions, terminals, GitHub and Linear integrations in one cohesive workflow.README 전문 보기 ↗A dev environment for AI agents. Run Claude, Codex, Cursor, OpenCode, Pi, Command Code, Grok, Kimi Code, Antigravity, and Muse in parallel - each in its own git worktree - on your laptop or your own server.
Tauri v2 · React 19 · Rust · TypeScript · Tailwind CSS v4 · shadcn/ui v4 · Zustand v5 · TanStack Query · Pierre Diffs · xterm.js
Jean is an opinionated AI agent workspace. It has strong opinions about how AI-assisted development should work: every task gets its own git worktree, agents run in parallel, and git work, reviews, terminals, and GitHub, Linear, and Sentry integrations live in one cohesive workflow.
Run it as a native desktop app, as a headless jean-server on Linux, or both: the desktop app shows local and remote projects together in one sidebar, grouped by server.
No vendor lock-in. Jean uses your own CLI installations and subscriptions, and everything runs on machines you control.
For more information, take a look at jean.build.
/goal tracking, session recap/digest, saved contexts with AI summarization, archiving with retention settings, recovery, auto-naming, canvas viewsCmd/Ctrl+Shift+B), usage and token trackingjean-server instances to the desktop app (or install one over SSH); their projects appear next to your local ones and every action runs on the server that owns the projectDownload the latest version from the GitHub Releases page or visit jean.build.
brew tap coollabsio/jean
brew install --cask jean
Prerequisites:
kernel32.lib and other system libraries required by Rust)See CONTRIBUTING.md for full development setup and guidelines.
Every Jean instance can run Web Access: an embedded HTTP + WebSocket server that serves the same UI in a browser. That applies to both:
| Mode | How you get Web Access |
|---|---|
| Native desktop (macOS / Windows / Linux) | Settings → Web Access - enable the HTTP server, set port/bind address, copy the token URL |
Headless server (jean-server) | Always on - the process is the Web Access endpoint |
Use it to open Jean from another machine on your LAN, a phone/tablet browser, or a remote host. Token authentication is on by default; keep it enabled for any non-localhost bind.
Web Access binds a normal TCP port (default 3456). For access beyond the local machine, prefer a private mesh VPN rather than exposing the port to the public internet:
jean-server --host 127.0.0.1 plus tailscale serve --bg 3456 so the
browser gets real HTTPS (clipboard and other secure-context APIs work).
Binding the Tailscale IP directly (--host tailscale) is HTTP-only and
browsers will treat it as an insecure origin.127.0.0.1Keep token auth enabled, use a long random token
(openssl rand -base64 32), and avoid binding 0.0.0.0 on untrusted networks
unless you also terminate TLS and restrict who can reach the port.
127.0.0.1 for local only, LAN IP,
0.0.0.0, or your Tailscale IP)?token=...) in a browser, or share it with
devices that can reach that hostThe desktop app can work with any number of Jean servers at the same time:
jean-server on that machine for you, orThere is no switching: sessions, terminals, file previews, and git actions run on the server that owns the project. Uncheck Include in combined dashboard to hide a server without removing it.
jean-server)Run Jean as a standalone Linux server with browser Web Access - no desktop window, GTK, or WebView required. Linux amd64 and arm64 only (glibc + OpenSSL 3; Ubuntu 22.04+ / Debian 12+ recommended).
Interactive install (prompts for bind interface + port when a TTY is available):
curl -fsSL https://raw.githubusercontent.com/coollabsio/jean/main/scripts/install-jean-server.sh | sudo bash
Non-interactive (defaults to 127.0.0.1:3456, or pass --host / --port):
curl -fsSL https://raw.githubusercontent.com/coollabsio/jean/main/scripts/install-jean-server.sh | sudo bash -s -- -y
Or from a clone:
sudo ./scripts/install-jean-server.sh --host 127.0.0.1 --port 3456 -y
The installer downloads the latest release, installs the binary, writes an env
file (host/port/token), and registers a systemd service. Re-run to upgrade;
existing tokens are preserved unless you pass --token.
Common options:
# Public bind with an explicit token
sudo ./scripts/install-jean-server.sh \
--host 0.0.0.0 \
--port 3456 \
--token "$(openssl rand -base64 32)" \
-y
# Tailscale-only bind (auto-detect Tailscale IPv4) - recommended for remote use
sudo ./scripts/install-jean-server.sh --host tailscale -y
# Current user only (user systemd unit)
./scripts/install-jean-server.sh --user-install --host 127.0.0.1 -y
jean-server --host 127.0.0.1 --port 3456
# or with an explicit token:
jean-server --host 127.0.0.1 --port 3456 --token "$JEAN_TOKEN"
--host accepts localhost, presets (tailscale, lan, 0.0.0.0, …), or
any IP/hostname to bind only that interface. Docker images are also published
as ghcr.io/coollabsio/jean-server.
See docs/headless-server.md for systemd details, reverse proxies, updates, and security notes.
See CONTRIBUTING.md for development setup and guidelines.
| Andras Bacsai |
|---|
Learn more about our approach: Philosophy