What Jean Does
Jean is a Tauri-built desktop app from the coolLabs team — the same group behind Coolify — that gives you a single opinionated interface for working with Claude CLI, Codex CLI, Cursor CLI, and OpenCode across multiple projects and git worktrees. It detects whichever CLIs you already have authenticated on your machine, surfaces them as selectable backends, and layers session management, GitHub and Linear integrations, and a Plan/Build/Yolo execution model on top. The whole thing is Apache 2.0 licensed and runs locally without proxying your code or prompts through coolLabs infrastructure.
Unified Multi-Agent Workflow
The reason to install Jean is that it removes the constant tab-switching tax of running parallel coding agents. Each project gets its own workspace, each worktree is a first-class entity, and within a worktree you open multiple sessions — one for Claude on the main branch, another for Codex reviewing the same diff, a third for Cursor working on a side branch — without having to juggle separate terminals. Plan-mode reviews work especially well here: you let one agent draft a plan, switch tabs, and have the other agent critique it before any code is written.
Setup friction is close to zero. If your Claude, Codex, Cursor, or OpenCode CLI is already installed and authenticated, Jean recognizes it automatically and adds it to the model picker. There is no separate API key management for the CLI backends, no second account to create. The desktop window then exposes the same models you would see at the terminal — Opus 4.5, 4.6, the 1M context variant, Sonnet 4.6, Haiku, plus Codex tiers — with per-mode thinking and effort overrides for teams that want fine control over reasoning budgets.
Plan, Build, and Yolo Modes
The execution model is one of Jean's strongest opinions. Plan mode forces the agent to draft an approach and pause for approval before touching files; Build mode executes step-by-step with diff confirmations; Yolo mode lets the agent run unattended for ambient tasks. Each mode has its own per-prompt model and effort override, so you can pair a high-effort Opus plan with a fast Sonnet build, or run a cheap Haiku Yolo loop on a sandbox worktree. For code review specifically, Plan mode plus the secondary-tab Codex agent is the workflow that converted the most skeptics on our side.
Magic Commands sit alongside the modes for repeatable scaffolding work. Investigating an issue, generating PR descriptions, drafting commit messages, resolving merge conflicts, and writing release notes are all single-command operations with their own model and backend overrides. The diff viewer — both unified and side-by-side — and the in-app browser preview close the loop for code review without sending you out to GitHub or VS Code mid-task.
GitHub and Linear Integrations
GitHub integration goes deeper than most desktop wrappers attempt. The dashboard pulls Issues, Pull Requests, Security Alerts and Advisories, and full Workflow Runs into the app, and you can investigate Dependabot findings or check out a PR directly as a worktree. Watching CI/CD actions from inside the same window where the agent is editing code closes a real productivity gap, and the auto-archive on PR merge keeps the worktree list tidy without manual cleanup.
Linear integration is API-key based, scoped per project with team configuration. The flow exists for issue investigation and context loading into a session — useful for teams that drive work from Linear tickets — but it requires you to generate and paste an API key into each project that needs it. That is not unusual for native apps, but compared to GitHub where the flow uses the standard CLI auth chain, the Linear path is a notch more friction.
Open Source, Local-First Trust Model
Apache 2.0 licensing and the local-first architecture are the trust foundation. The README is explicit that Jean has no vendor lock-in, that everything runs on your machine, and that your code never leaves your hardware. There is no enterprise edition behind a paywall and no premium tier withholding features. The coolLabs team funds the project through GitHub Sponsors, Open Collective, and managed cloud services for their other open-source products — a track record that, given Coolify's standing in the self-hosted infrastructure community, raises the trust ceiling for Jean as well.
The remote access story rounds this out. Jean can run headless and expose a web UI over HTTP with token-based auth, which makes it usable from a phone over Tailscale or a Cloudflare Tunnel. Bind it to a specific Tailscale IP and you have an agent workstation accessible from any device on your tailnet without exposing it to the public internet. For developers who already run Tailscale across their machines, this is a meaningful upgrade over desktop-only assistants.
The Bottom Line
Jean is the cleanest realization we have seen of "one window for every coding agent you actually use." If your day already runs across Claude CLI and Codex CLI, with the occasional worktree experiment in Cursor or OpenCode, the unification is genuinely valuable — and the Plan/Build/Yolo discipline plus Magic Commands push you toward better practices rather than papering over them. Apache 2.0 licensing, local-first architecture, and the coolLabs operational track record make this a low-risk install for teams who want a serious AI dev environment without committing to a closed platform. The macOS-first testing posture and the newness of the project are real but manageable caveats; for daily-driver multi-agent work in 2026, Jean earns its spot.
