Claudette Workspace Orchestrator
Run coding sessions in parallel, isolated Git worktrees and manage them from a desktop app.
- Source repo
- utensils/claudette
- Stars
- ★ 87
- Last updated
- 4d ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 70/100 · Some gaps
At a glance
- How it runs
- Works with
- Portable with changesClaude CodeOpenAI API (Partial support)
- Setup effort
- High · needs real infrastructure
- You'll need
- Typical use
- Developers handling several features or bugs at once can run a Claude Code session for each task in an isolated worktree.
- Not a fit if
- Users who only need a lightweight prompt interface without workspace management
- Teams unable to install Rust, Bun, and Tauri build dependencies
- Source review
- 70/100 · Some gaps
What does this agent do, and when should you use it?
Claudette is a cross-platform desktop application built with Tauri 2, a Rust backend, and a React/TypeScript frontend for managing and orchestrating Claude Code sessions. It creates isolated Git workspaces for parallel sessions and supports checkpoint rollback, workspace forking, plan approvals, and handling user questions. Its companion claudette CLI drives a running desktop app over a local IPC socket, with scripted operations reflected in the GUI. The app also provides remote workspace connections, MCP server monitoring, and GitHub and GitLab SCM plugins. Beyond Claude Code, it lists Ollama, local models, the OpenAI API, and Codex as agent backends; building and developing require Rust, Bun, Tauri CLI, and platform dependencies.
Users add repositories in the desktop app and create workspaces; Claudette starts Claude Code sessions in separate Git branches and worktrees and shows sessions, terminals, and code diffs. Users can send prompts, inspect transcripts and tool activity, stop sessions, and resolve pending questions or plan approvals in the GUI or with claudette chat answer, claudette chat approve-plan, and claudette chat deny-plan. The CLI can also create workspaces in batches and dispatch prompts; the app supports checkpoint rollback and forking from a checkpoint. A remote machine can run the embedded server or standalone claudette-server; clients connect over TLS WebSocket, with mDNS discovery on LAN. MCP configuration and SCM status are also managed from the app.
- Developers handling several features or bugs at once can run a Claude Code session for each task in an isolated worktree.
- Engineering leads breaking a plan into parallel coding tasks can use a YAML manifest and
claudette batch runto create workspaces and dispatch prompts. - Developers who need to review an agent's plan or provide missing input can resolve pending approvals and questions from the GUI or CLI.
- Developers working across machines can connect to remote workspaces and access remote repositories, agents, and terminals.
- Teams that need pull request and CI status can use the GitHub or GitLab SCM providers to view status and merge from the app.
How do you install or deploy this agent?
The README documents building and running from source, but gives no command for installing a prebuilt application. Prepare the Rust stable toolchain (edition 2024), Bun, Tauri CLI 2, and platform build dependencies; the default Linux voice feature also needs ALSA. Install frontend dependencies, then launch with the platform-specific development script:
cd src/ui && bun install && cd ../..
./scripts/dev.shOn Windows, use the PowerShell launcher:
cd src/ui; bun install; cd ../..
.\scripts\dev.ps1Build an optimized desktop binary with:
cargo tauri buildFor a headless remote host, install and start the standalone server:
cargo install --path src-server
claudette-serverHow do you use this agent?
Start the desktop app and add a repository to create workspaces and chat sessions in the GUI. CLI workspace and chat operations require the app to be running; after installing or building the claudette CLI, list repositories, create a workspace, and send a prompt:
claudette repo list
claudette workspace create <repo-id> my-task
claudette chat send <session-id> "Implement the requested change"For parallel dispatch from a manifest, validate it and then run it:
claudette batch validate plan.yaml
claudette batch run plan.yamlFor remote access, choose Share this machine in the desktop app to start its server, or run standalone claudette-server; connect from another app through the Nearby LAN entry or by pasting a claudette:// connection string and entering the pairing token. The Claude Code backend requires the official claude CLI installed and configured locally. The README also lists Ollama, the OpenAI API, and Codex backends.
What are this agent's strengths and limitations?
- Creates isolated Git worktrees for parallel Claude Code sessions, reducing direct workspace collisions between tasks.
- The desktop app and CLI share a local IPC command core, so scripted actions update the GUI workspace list, tray icons, and notifications.
- Provides checkpoint rollback and forking, with GUI and CLI paths for plan approval and answering questions.
- Connects remote workspaces over TLS WebSocket with first-connection certificate fingerprint pinning and supports mDNS discovery on LAN.
- Documents Claude Code, Ollama, OpenAI API, and Codex backends, plus GitHub and GitLab SCM plugins.
- The documented setup is a source build and requires Rust, Bun, Tauri CLI, and platform build dependencies.
- Most CLI workspace and chat commands require the desktop app to be running; the CLI is not a complete headless control client.
- The Claude Code backend depends on the official local
claudeCLI and its configured credentials; backend paths may have different features. - The default Linux voice feature requires ALSA, though the README gives a command to build without voice.
- The README does not state software or model-backend pricing and does not document prebuilt app installation.
How does this agent compare with similar options?
The README loosely compares Claudette's spirit to Conductor.build and says its feature set is more focused; it provides no detailed feature-by-feature comparison.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Claudette Workspace Orchestrator This agent | 70 · Some gaps | Desktop app | ★ 87 | 4d ago | TypeScript | Claude Code |
| Superset | 66 · Some gaps | Desktop appFreemium | ★ 16k | today | TypeScript | Codex · Claude Code |
| Agor | 68 · Some gaps | CLIFree + model costs | ★ 1.5k | 2d ago | TypeScript | Codex · Claude Code |
| Poracode | 67 · Some gaps | Desktop appFree + model costs | ★ 121 | today | TypeScript | ChatGPT · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The packet shows that the CLI calls the GUI over local IPC and the GUI owns database writes. The README says inherited subscription tokens are stripped from Claude CLI subprocesses, plugin secrets use a separate namespace, and remote connections use TLS with first-connection certificate pinning. Deductions: the CLI exposes raw RPC and arbitrary plugin invocation, the server defaults to binding 0.0.0.0, and the packet does not establish confirmation thresholds, fine-grained permissions, or recovery paths for all side-effecting operations.
The README, CLI comments, and source agree that the GUI must be running for IPC operations; the CLI propagates errors and exits nonzero. CI configuration lists Rust, frontend, and mobile checks. Deductions: the packet contains no execution results and does not show complete GUI or server failure-handling paths, so full marks are not supported.
The product targets Claude Code users and parallel workflows, with documentation for desktop, CLI, remote access, mobile, alternative backends, and multiple operating systems. Prerequisites and platform differences are detailed. Deductions: the sampled skills use fairly broad triggers, and the material centers on Claudette workspaces and agent orchestration without fully defining boundaries for other audiences.
The README provides project structure, prerequisites, installation and development commands, CLI examples, and contribution guidance; an MIT license is included. CLAUDE.md describes repository contribution conventions rather than an end-user Agent skill. Deductions: no changelog or clear maintenance owner/update commitment is included in the packet; Cargo metadata gives a version but does not supply change history.
CLI examples cover machine-readable output, workspace and session control, batch fan-out, and plan approval. The product connects GUI and CLI workflows, giving it clear practical value. Deductions: parallel agents, remote access, voice, and multiple backends add dependency and operational costs, which the packet does not quantify or bound by scale.
Some README claims are corroborated by the CLI source, server entry point, Cargo metadata, and CI configuration; the skills give concrete commands and prerequisites. Deductions: many implementation details appear only as documentation claims, and CI configuration shows intended checks but is not evidence of execution or correctness in this review.
- Raw RPC and arbitrary plugin invocation broaden the CLI's operation surface; check the target workspace and action before writes, prompt dispatch, or plugin calls.
- The standalone server defaults to listening on all interfaces; check network exposure before enabling remote access and handle pairing tokens and connection strings carefully.
- The test and CI commands in the packet describe configured checks; they do not show that checks were run or verified in this review.
Maintain this project? Add the FARS badge to your README
[](https://followagents.com/en/agents/claudette)<a href="https://followagents.com/en/agents/claudette"><img src="https://followagents.com/badge/claudette.svg" alt="FARS review"></a>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
Automated nightly build of a69a1fc496902084c4cc68c55f4038c1427dfcaf. Version: 0.29.0-dev.788.ga69a1fc Built: 2026-09-22T20:31:54Z [Changes since v0.28.0](https://github.com/utensils/claudette/compare/v0.28.0...a69a1fc496902084c4cc68c55f4038c1427dfcaf) > These builds are unstable pre-releases.View on GitHub ↗ -
Release notes excerpt
[0.28.0](https://github.com/utensils/claudette/compare/v0.27.1...v0.28.0) (2026-08-30) Features * chat: let pinned prompts open a new tab and pick a model ([#1021](https://github.com/utensils/claudette/issues/1021)) ([f6f6978](https://github.com/utensils/claudette/commit/f6f6978750f86efdaa74b6b73db5c1fcccc35159)) Bug Fixes * chat: resolve backgrounded Workflow runs from the Rust notification path ([#1023](https://github.com/utensils/claudette/issues/1023)) ([2c730bd](https://github.com/utensils/claudette/commit/2c730bd9a4ae1d9c3522d60381ca7594e7ebf726)) …View on GitHub ↗ -
Release notes excerpt
[0.27.1](https://github.com/utensils/claudette/compare/v0.27.0...v0.27.1) (2026-08-05) Bug Fixes * chat: clear the Workflow status pill when a run ends or is killed ([#1013](https://github.com/utensils/claudette/issues/1013)) ([8c22db7](https://github.com/utensils/claudette/commit/8c22db7c975b3a24d7fa4edd742a6c6d26c68e32)) * chat: render insight callouts that the model wrapped in a code fence ([#1016](https://github.com/utensils/claudette/issues/1016)) ([0db5c4e](https://github.com/utensils/claudette/commit/0db5c4e7fe32e0dbd00eb15c81efa01aa96dad46))View on GitHub ↗
FAQ
Does Claudette log into Anthropic for me or read my OAuth token?
claude CLI, which uses credentials already configured for that CLI. The README says Claudette does not read, copy, or forward Claude OAuth tokens.Can parallel agents affect my Claude subscription usage?
Can I use the CLI without the desktop app?
claudette completion and claudette version are the documented exceptions.Can I run remote access without a desktop GUI on the remote machine?
claudette-server with cargo install --path src-server and using it to provide remote access over TLS WebSocket.Which desktop systems are supported?
Thanks — we use this to improve the page.