Maestro: Agent Orchestration Command Center
Cross-platform desktop app for orchestrating multiple AI coding agents in parallel, with automation via playbooks and Git worktrees.
Evidence: SECURITY.md explicitly states that processes run with user privileges and advises users to review agent actions, indicating some consideration of least privilege. User confirmation: README mentions Auto Run executes tasks automatically, but does not clearly state a user confirmation mechanism. Data flow transparency: README describes data flow (e.g., MCP tools, permissions, authentication) but lacks detailed data flow paths. Sensitive data handling: SECURITY.md mentions Sentry DSN is public, but does not explain how user sensitive data is handled. Dependency security: package.json lists dependencies but no vulnerability scanning or audit evidence. External effects: SECURITY.md mentions local web server and Cloudflare tunnel, but not default state. Rollback: No rollback mechanism mentioned. Source attribution: package.json and README provide author and repository info, but publisher is unverified. Deductions: Lack of explicit user confirmation, data flow transparency, sensitive data handling details, and rollback mechanism.
Evidence: README and package.json descriptions are consistent, e.g., feature list and version number, showing good self-consistency. Dependency availability: package.json lists many dependencies but no lock file or integrity checks, posing availability risk. Failure messages: SECURITY.md mentions response times but no specific error handling or user prompts. Deductions: Dependency locking and failure message implementation not evident in static files.
Evidence: README clearly defines target users (hackers, power users) and use cases (parallel multi-project), audience and scenarios are clear. Capability boundaries: README lists supported agents and features but not unsupported ones. Trigger precision: README mentions Auto Run and Playbooks but lacks detailed trigger conditions. Environment fit: README provides installation requirements and platform support but not all environment limitations. Deductions: Trigger precision and some environment limitations not detailed.
Evidence: README provides detailed feature list and documentation links, information architecture is clear. Install notes: README provides build-from-source steps but not detailed installation guide. Naming stability: package.json name and version are consistent but no naming conventions. Examples and FAQ: README provides video and doc links but no FAQ. Known limitations: README mentions limited support but not all limitations. License: LICENSE file is AGPL-3.0, compliant. Versioning changelog: No CHANGELOG file. Maintenance responsibility: SECURITY.md mentions volunteer maintenance but no clear responsibility assignment. Deductions: Missing CHANGELOG, FAQ, and detailed known limitations.
Evidence: README describes output filtering, JSONL output, etc., output usability is good. Marginal value: Maestro offers unique features like multi-agent orchestration and Auto Run, high marginal value. Cost benefit: README mentions cost tracking but no pricing or cost analysis. Deductions: Cost benefit lacks concrete data.
Evidence: README feature descriptions correspond to dependencies and scripts in package.json, but no test results or verification evidence. Cross-source corroboration: README and SECURITY.md provide some security info but not cross-validated with other sources. Fact-inference separation: README descriptions are mostly assertions, not clearly separating facts and inferences. Deductions: Lack of test evidence and explicit verification methods.
- Publisher identity is unverified; proceed with caution.
- Dependencies are not locked, posing supply chain risks.
- Auto Run may execute tasks without explicit user confirmation, posing security risks.
- No rollback mechanism; actions may be irreversible.
What does this agent do, and when should you use it?
Maestro is a cross-platform desktop application (macOS, Windows, Linux) that supports Claude Code, OpenAI Codex, OpenCode, Factory Droid, and Copilot-CLI (beta). It provides a Linear/Superhuman-level responsive interface for managing multiple agents in parallel. Key features include Git Worktrees for conflict-free parallel development, and a file-system-based Auto Run & Playbooks task runner that batch-processes markdown checklists through AI agents, each task in a fresh session. It supports Group Chat for coordinating multiple agents, mobile remote control via a built-in web server, a CLI for headless operation, message queueing, session discovery, Git integration, file explorer, output filtering, slash commands, and more. Maestro is a pass-through to your AI provider, meaning MCP tools, skills, and authentication configured in Claude Code, Codex, or OpenCode work identically in Maestro, but tasks are executed non-interactively.
Maestro acts as a non-interactive front-end to multiple command-line AI coding agents such as Claude Code, Codex, and OpenCode. It allows users to launch and manage multiple agent sessions, each with its own workspace and isolated context. It sends prompts to the underlying agent's CLI and receives responses, whether starting a new session or resuming a previous one. It automatically discovers and imports existing sessions from all supported providers. Git Worktrees allows creating sub-agents from the git branch menu, each operating in its own directory, and creating PRs with one click. Auto Run reads markdown checklists from the file system and batch-processes them through AI agents, with each task getting its own AI session. Group Chat uses a moderator AI to orchestrate discussions, routing questions to appropriate agents and synthesizing responses. Message queueing holds messages while the AI is busy and sends them automatically when ready. The CLI (maestro-cli) supports headless operation: it can list agents/groups, run playbooks from cron jobs or CI/CD pipelines, and output human-readable or JSONL for scripting. It includes a usage dashboard that tracks token usage and cost per session and globally, with CSV export. Document Graph builds a visual knowledge graph from markdown documents, mapping internal wiki-links and markdown links, with interactive nodes and edges.
- A developer juggling multiple projects simultaneously, using Claude Code, Codex, and OpenCode, wants a unified interface to manage and switch between agents.
- A developer needing to run multiple tasks in parallel on isolated branches to avoid conflicts, using Git Worktrees.
- A team wanting to automate repetitive workflows such as code reviews, testing, or documentation generation through Playbooks for unattended batch execution.
- A developer needing to monitor and control agents from their phone while away from the desk, using the mobile remote control feature.
- An advanced user integrating agent tasks into CI/CD pipelines or cron jobs using the CLI (maestro-cli) for headless operation.
- A team wanting to coordinate multiple agents in a single conversation for cross-project architecture discussions or Q&A, using Group Chat.
What are this agent's strengths and limitations?
- Supports multiple major AI coding agents (Claude Code, Codex, OpenCode) via a unified non-interactive interface.
- Git Worktrees enable true parallel development without branch conflicts.
- Auto Run and Playbooks support unattended execution of long-running tasks, improving automation efficiency.
- Group Chat enables coordination of multiple agents for complex discussions.
- Keyboard-first design with customizable shortcuts and achievement system enhances productivity for power users.
- Requires installation and authentication of at least one external AI coding agent, adding deployment dependencies.
- Acts as a pass-through, so it does not provide its own AI model; depends on external providers' availability and cost.
- Shortcuts differ between macOS and Windows/Linux, which may require relearning on different platforms.
- Copilot-CLI is marked as beta, so support may be less stable.
- AGPL-3.0 license may restrict commercial integration or modification.
How do you install or deploy this agent?
Download the latest release for your platform from the Releases page. Or build from source: clone the repository, run npm install and then npm run dev. You need at least one supported AI coding agent installed and authenticated: Claude Code, Codex, OpenCode, or Copilot-CLI (beta). Git is optional for git-aware features.
How do you use this agent?
Install and launch Maestro. Create a new agent with Cmd+N (Ctrl+N) and select which AI agent to use (e.g., Claude Code). Type your request in the prompt area, and Maestro will send it to the selected agent's CLI. Use Cmd+J (Ctrl+J) to toggle between AI and command terminals. Use Cmd+K (Ctrl+K) for quick actions, and Cmd+[ / Cmd+] to switch between agents. To use Auto Run, create a playbook file containing markdown checklists and run it from the Auto Run menu. To use Group Chat, start a group chat session from the interface, input a message and let the moderator AI coordinate.