Dorothy Agent Workbench
Run, coordinate, and monitor multiple coding agents in one desktop workspace.
- Source repo
- Charlie85270/Dorothy
- Stars
- ★ 351
- Last updated
- 1d ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 49/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesClaude CodeCodex (Partial support)
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Developers working across several repositories can run a separate terminal agent for each project and follow its live output.
- Not a fit if
- Teams using only Codex, Gemini, or Grok without Claude Code
- Users who need a fully working setup without configuration
- Source review
- 49/100 · Major gaps
What does this agent do, and when should you use it?
Dorothy is an Electron desktop app with a Next.js and React interface for managing coding agents across projects. It starts an isolated terminal session per agent through node-pty, while a Super Agent can create, delegate to, and monitor other agents using MCP tools. Users can organize work with a Kanban board, scheduled tasks, and GitHub or JIRA automations, and control agents remotely through Telegram or Slack. The README presents Claude Code CLI as the primary execution path, while also listing Codex, Gemini, Grok, and local agents; its Google Workspace MCP integration can be registered with configured providers. Dorothy is aimed at local desktop deployment; browser development mode lacks agent management and terminal features.
Users create agents in the Electron app, set a project path, skills, and model, then start a task; the Agent Manager runs a separate CLI process for each agent through node-pty and streams terminal output to the interface. The Super Agent uses mcp-orchestrator tools to create, start, stop, and inspect agents and wait for completion. Kanban tasks can be matched to agents by required skills or trigger agent creation; scheduled tasks run on cron, while automations poll GitHub or JIRA, spawn temporary agents for items, and deliver results to Telegram, Slack, or comments. The app also provides a persistent Vault, Twitter/X tools via SocialData, and Google Workspace MCP integration.
- Developers working across several repositories can run a separate terminal agent for each project and follow its live output.
- Engineering teams triaging pull requests or issues can poll GitHub or JIRA and have temporary agents process new items and return results.
- Teams with recurring project maintenance can schedule Claude Code CLI tasks on a cron expression.
- Users who need remote task control can check status and delegate work through Telegram or Slack.
- Teams that want task assignment and progress in a visual workflow can use the Kanban board with skill requirements and automatic assignment.
How do you install or deploy this agent?
The README documents building the Electron desktop app from source. Prerequisites are Node.js 18+, npm or yarn, and Claude Code CLI; GitHub automations also require GitHub CLI. Clone the repository, enter the app directory, install dependencies, rebuild Electron native modules, then start development mode or build the production app:
git clone https://github.com/Charlie85270/Dorothy.git
cd Dorothy/app/dorothy
npm install
npx @electron/rebuild
npm run electron:devTo build the production app, run:
npm run electron:buildYou can also download a release from GitHub Releases. The README does not specify release package version requirements, signing, or desktop installation steps for Windows or Linux.
How do you use this agent?
For a first run, install and configure the CLI tools and credentials needed for the integrations you plan to use. The documented Claude Code CLI installation command is:
npm install -g @anthropic-ai/claude-codeEnter Telegram, Slack, or SocialData credentials in Dorothy settings as needed. Google Workspace requires the gcloud CLI and gws CLI, followed by OAuth setup and login; GitHub automations use gh CLI. Then add a project and create an agent in the interface, enter a task, and start it. You can also create a Kanban task, schedule a cron task, or use Telegram commands such as /status and /ask <message> to check status and delegate. The README does not provide a step-by-step first-agent walkthrough or a complete list of required credentials.
What are this agent's strengths and limitations?
- Runs each agent in its own PTY session, with parallel execution, live output, and interactive input.
- The Super Agent manages agents through MCP tools and can create temporary agents for one-off work.
- Combines Kanban auto-assignment, cron scheduling, and GitHub/JIRA polling automations.
- Offers Telegram and Slack remote control plus a cross-agent SQLite full-text-search Vault.
- The README presents Claude Code CLI as the main execution engine; it does not detail the capabilities and limits of each other provider.
- Agent management and terminal features require the Electron app; browser mode does not provide them.
- Setup involves Node.js and Claude Code CLI, while GitHub automations require gh; Google Workspace also needs gcloud, gws, and OAuth.
- The documented autonomous mode includes
--dangerously-skip-permissions, so project permissions need careful handling.
How does this agent compare with similar options?
The README contrasts Dorothy with running one AI CLI tool at a time in one terminal: Dorothy manages multiple isolated terminal sessions and adds scheduling, Kanban, and remote control.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Dorothy Agent Workbench This agent | 49 · Major gaps | Desktop appFree + model costs | ★ 351 | 1d ago | TypeScript | Claude Code |
| Manifold | 60 · Some gaps | Self-hosted serviceFree | ★ 502 | 1d ago | Go | OpenAI API · Claude API |
| SuperAgentX Workflow Framework | 37 · Major gaps | Library / SDKFree + model costs | ★ 202 | 4mo ago | Python | OpenAI API |
| Kronova Asset Intelligence Platform | 30 · Major gaps | Web appFreemium | ★ 252 | 1mo ago | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
The README identifies agent management, automation polling, MCP tools, and data or external-effect paths through Telegram, Slack, GitHub, Jira, and Google Workspace, making data flows partly visible. Deductions: agent creation defaults skipPermissions to true and the README advertises --dangerously-skip-permissions; automation examples can post comments and transition Jira issues, with no stated per-action confirmation, least-privilege controls, credential protections, recovery process, or dependency security measures. The README and license also show different copyright names, limiting attribution.
The README describes agent lifecycle states, automation deduplication, and logs; package.json defines a test script and CI runs npm ci and npm test. Deductions: the README claims five MCP servers and 40+ tools, while the build scripts also include mcp-x and mcp-world; the README requires Node 18+, while CI uses only Node 24. Failure handling and dependency availability are lightly documented, and static files cannot establish runtime results.
The product targets developers using Claude Code, Codex, Gemini, Grok, and local agents, with parallel orchestration, Kanban tasks, scheduled work, and remote management. Automation sources, template variables, schedules, and the browser development limitation are described with useful specificity. Deductions: access boundaries for agents, appropriate limits for autonomous mode, and environment-specific constraints are not fully defined.
The README provides a table of contents, architecture, installation and configuration sections, feature tables, and usage examples; an MIT license is included. Deductions: installation guidance is mainly for macOS, with no complete platform installation guide; there is no FAQ, known-limitations list, or clear maintenance responsibility; package.json has a version but the supplied material has no changelog; the documented MCP server count differs from the build list, weakening naming stability evidence.
Parallel agents, MCP orchestration, automations, task management, and remote control could consolidate several workflows, and the README gives concrete tools and processes. Deductions: evidence for costs, resource use, and benefits of autonomous execution is limited, while permission defaults and integration setup add operational burden.
The README provides concrete descriptions of features, parameters, workflows, and installation, with partial corroboration from package.json, the license, and CI configuration. Deductions: many capability claims rely primarily on the README, with no runtime evidence; the documented server count and Node requirement differ from other supplied files; the material does not consistently distinguish implemented features, planned features, and unverified outcomes.
- Agent creation defaults to skipPermissions, and automations can publish comments or change Jira status; verify permissions and automation settings before enabling them.
- The README's MCP server count and Node version requirement conflict with other supplied files. This assessment is based on static material only; the project was not run.
FAQ
Is the software itself free?
Can I use it without Claude Code?
Can I manage agents in a browser?
Which sources do automations support?
Can agents run unattended?
--dangerously-skip-permissions, so assess permissions for the project environment.