Jean Desktop AI Development Assistant
A unified desktop environment for managing multiple projects, Git worktrees, and AI chat sessions across multiple CLI assistants.
- Source repo
- coollabsio/jean
- Stars
- ★ 1.3k
- Last updated
- today
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 47/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code · OpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer juggling multiple projects who wants to manage Git worktrees from a single interface without manual steps.
- Not a fit if
- Users who want zero-config setup without installing each CLI tool
- Teams needing reliable Windows support (not fully tested)
- Source review
- 47/100 · Major gaps
What does this agent do, and when should you use it?
Jean is a Tauri-based desktop app that provides a cohesive local interface for AI-assisted development. It integrates multiple CLI AI tools including Claude CLI, Codex CLI, Cursor CLI, OpenCode, PI, Command Code, Grok, and Kimi, and offers multi-project management, Git worktree automation, session management, chat, Magic Commands (e.g., code review, auto commit messages), GitHub and Linear integrations, a multi-dock terminal, and a diff viewer. Everything runs locally with your own CLI installations, ensuring no vendor lock-in. Jean also supports Web Access via HTTP/WebSocket, allowing browser-based or headless server access.
Jean manages multiple projects and their Git worktrees (create, archive, restore, delete) through a native desktop interface. It creates and resumes sessions, supports plan/build/yolo execution modes with approval flows, and provides chat UIs for various CLI assistants (Claude CLI, Codex CLI, Cursor CLI, etc.), including model selection, MCP support, multi-agent collaboration, and attachments. Magic Commands automate investigating issues, code review, commit message generation, PR content creation, merge conflict resolution, and more. The GitHub dashboard lets you browse issues, PRs, security alerts, and investigate Dependabot. Linear integration loads context and investigates issues. Developer tools include a multi-dock terminal, command palette, Git operations, and diff viewer. Web Access exposes the full UI over HTTP/WebSocket with token auth, usable in browsers or as a headless server (jean-server).
- A developer juggling multiple projects who wants to manage Git worktrees from a single interface without manual steps.
- A team relying on Claude CLI or Codex CLI for AI-assisted coding, needing centralized session management and execution modes.
- A developer automating code review, commit message generation, or merge conflict resolution via Magic Commands.
- A maintainer of several repositories who uses the GitHub dashboard to triage issues, PRs, and security alerts.
- A remote or headless user who accesses Jean's Web Access from a browser to manage AI sessions and terminals.
How do you install or deploy this agent?
Download the latest version from GitHub Releases or jean.build. On macOS, use Homebrew: brew tap coollabsio/jean && brew install --cask jean. To build from source, install Node.js, Rust, and Git; on Windows, also install the Visual Studio C++ workload. For the headless server (Linux), use the install script or run jean-server manually.
How do you use this agent?
Launch Jean, add or link projects, then select an AI CLI (must be pre-installed). Create worktrees or sessions to start AI-assisted development. Use Magic Commands or chat to interact with AI. To enable Web Access, go to Settings → Web Access, enable the HTTP server, set port and bind address, then open the token URL in a browser. For the headless server, use jean-server --host 127.0.0.1 --port 3456 and configure token auth.
What are this agent's strengths and limitations?
- Supports multiple leading CLI AI tools (Claude, Codex, Cursor, etc.) with no vendor lock-in.
- Built-in Git worktree automation and multi-project support significantly boost productivity.
- Magic Commands automate commit messages, PR content, code review, etc., reducing repetitive work.
- Web Access enables remote and headless usage, offering flexible deployment.
- Windows support is not fully tested, so issues may arise.
- Requires manual setup and configuration of each CLI AI tool, adding integration overhead.
- As a desktop app, it may consume significant resources and requires Node.js and Rust for building.
How does this agent compare with similar options?
Compared to terminal-based AI assistants like GitHub Copilot CLI or Aider, Jean offers a graphical interface and richer multi-project workflow management.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Jean Desktop AI Development Assistant This agent | 47 · Major gaps | Desktop appFree + model costs | ★ 1.3k | today | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
| Diri | 59 · Major gaps | Desktop appFree + model costs | ★ 453 | today | Rust | Codex · Claude Code |
| wmux Agent Workspace | 74 · Some gaps | Desktop app | ★ 431 | today | TypeScript | Codex · Claude Code |
| Cezar — AI Coding Agent Orchestrator | 63 · Some gaps | CLIFree | ★ 722 | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows the project provides Web Access with token auth and recommends secure practices like Tailscale, indicating some consideration for permissions and sensitive data. However, no explicit least-privilege design, user confirmation mechanisms, or detailed data flow transparency were found. Dependency security has CI but no dependency audit evidence. External effects: Web Access may expose services, but docs provide security advice. Rollback: session recovery and archiving exist, but no system-level rollback. Source attribution: clear author and license, but publisher unverified. Thus scores are low due to insufficient evidence or partial satisfaction.
Project has CI (preflight.yml) with typecheck, lint, tests, indicating some assurance of consistency and dependency availability. But no specific failure message examples or error handling docs. Self-consistency: README and package.json versions match (0.1.72), but no detailed failure handling. Dependency availability: lock file (bun.lock) and CI cache exist, but no guarantee. So scores are moderate-low.
Project targets developers, supports multiple CLIs and platforms, scenarios clear. Capability boundaries described in README but not detailed. Trigger precision: command palette and shortcuts exist, but no detailed trigger conditions. Environment fit: supports macOS, Windows, Linux, but Windows not fully tested, Linux community tested. So scores moderate.
Information architecture clear with README, CONTRIBUTING, SECURITY. Install notes detailed including Homebrew, source build, headless server. Naming stability: version exists but no naming conventions. Examples and FAQ: screenshots and e2e tests, but no FAQ. Known limitations: README mentions Windows not fully tested, but not systematic. License Apache-2.0 clear. Versioning changelog not provided, but version number exists. Maintenance responsibility: core maintainer info. So scores moderate.
Output usability: UI and Web Access, but no output format specs. Marginal value: feature-rich, but no comparison with other tools. Cost-benefit: open-source free, but no performance or resource data. So scores moderate-low.
Feature claims in README lack implementation details or test evidence. Cross-source verification: e2e tests and CI exist, but no independent verification. Fact-inference separation: docs do not clearly distinguish facts from inferences. So scores low.
- Web Access may expose services; keep token auth enabled and avoid binding 0.0.0.0.
- Windows platform not fully tested; use with caution in production.
- Publisher identity unverified; review code before installation.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Features - Added checkboxes to the context mention popover. You can now investigate several issues and PRs together in one new session, or in a separate new session for each item. - The context link badge now opens the linked session or worktree. - pstack now installs as a native Claude and Codex plugin. Fixes - Jean now asks you to confirm before it closes a session that has unsent draft text. - Review sessions now show as unread only after the review finishes. - Web access now uploads large attachments (up to 50 MB) over HTTP. Before, they could close the WebSocket connection. …View on GitHub ↗ -
Release notes excerpt
Features - Added labels, authors, and opened times to worktree issue and PR lists, with clickable label filters. The new worktree modal opened fullscreen, and draft PRs got a muted icon. Fixes - Fixed plan approval with "Clear context" or a new worktree deleting the original planning session. The session was archived instead, so its history stayed restorable (#795). - Fixed MCP send_chat_message and get_session_status failing with "Unknown sessionId" right after create_session or for reopened sessions (#794, closes #793). …View on GitHub ↗ -
Release notes excerpt
Features - Added experimental Agents, disabled by default, with a coordinator, custom workers, persistent memory, schedules, task tracking, notifications, and approval controls across local and remote servers. - Added unified CLI setup with parallel installation, guided sign-in, SSH host discovery, connection checks, and remote installation progress. - Added SSH identity key selection for server connections. - Added Commit in Session and optional closing references for linked GitHub issues. - Added video file previews and New Worktree actions in worktree menus. …View on GitHub ↗