- Does my code leave my machine?
- No. Maestro is a Python CLI that runs entirely on your own machine. The only outbound calls are to api.anthropic.com using your own API key (the same call the claude CLI makes by hand), to whichever source-control and CI hosts you have configured (api.github.com by default, or your own Bitbucket, Jenkins or Jira endpoints) for operations you request, and to licence.maestrodevs.com for a periodic tier check. Your source code is never sent to a Maestro server, because there is no Maestro server in the request path.
- Can I use Maestro with a Claude Pro or Max subscription?
- Maestro authenticates with a standard Anthropic API key, not a subscription login. In April 2026 Anthropic updated its policy to restrict the use of Claude Pro and Max subscription OAuth tokens in third-party agent frameworks. Maestro is unaffected by that change because it never uses subscription OAuth tokens. You supply an Anthropic API key and pay Anthropic directly for token usage.
- How is this different from just running multiple Claude Code terminals?
- Opening several Claude Code terminals gives you parallel sessions with no coordination: they share no plan, no dependency order, and no quality gates, and they will happily edit the same files at once. Maestro assigns each session a specialised role (Planner, Builder, Tester, Documenter, Security, Releaser), runs each in its own Git worktree to prevent write conflicts, and enforces dependency ordering and acceptance-criteria gates between roles. You describe the work as structured orders and Maestro routes, sequences, and validates them.
- Does Maestro work in air-gapped environments?
- Yes. Set dataResidency.mode to air-gapped in workspace.json and Maestro refuses every outbound HTTP request except localhost. Licence validation has a 14-day offline grace period, so an air-gapped estate keeps working between revalidations. Claude inference still needs to reach an Anthropic endpoint, so fully offline use requires a local or on-premises model proxy.
- Do I need an Anthropic API key?
- Yes. Maestro is bring-your-own-key: you sign up with Anthropic directly and supply your own API key. Maestro's licence covers the orchestration layer only. You pay Anthropic for model token usage. Your billing relationship and your key stay entirely under your control.
- What exactly does Maestro send over the network?
- Three kinds of destination, and only these. Crew prompts go to api.anthropic.com under your key. Source-control, CI and issue-tracking calls go to the providers you configured, and only when a crew you dispatched requests them: that is api.github.com out of the box, or your own Bitbucket, Jenkins or Jira hosts if you point Maestro at those instead. A licence check goes to licence.maestrodevs.com. Workspace state (orders, plans, reports, the audit log) stays in the .mso/ directory on local disk and is never uploaded, and restrictive mode refuses any destination outside your allowlist.
- Which operating systems does Maestro run on?
- Maestro runs on Windows, macOS, and Linux. It requires Python 3.9 or newer and the Claude Code CLI. On Windows the MAESTRO_ITER_MANAGED environment variable handles a subprocess-signalling difference; the dispatcher sets it automatically.
- How many agents can run at once?
- Up to five crews run concurrently on Pro and above, across unlimited registered projects; the Community tier is capped at one registered project and two crews across four roles (INV/PLN/BLD/TST). Each crew operates in an isolated Git worktree, so multiple sessions can work the same repository in parallel without colliding.
- Is Maestro open source?
- The orchestration engine and CLI ship as a proprietary pip wheel; the source is not published. The framework still runs locally, and its data-flow behaviour is documented and verifiable: you can inspect every network destination and confirm that workspace state stays on disk.
- How does Maestro support security and governance?
- Every file write passes a pretool secret scan that blocks AWS keys, GitHub PATs, and other credentials before they reach a commit. A posttool scanner checks code against OWASP and CWE patterns, an MCP registry classifies every server before a crew may use it, and a hash-chained audit log records privileged actions. These controls are covered in detail on the financial-services use-case page.
- What is a voyage?
- A voyage is Maestro's term for a sprint or work package. One voyage maps to one Git branch and one pull request, and contains a set of orders to complete. The name comes from the default nautical persona; the mechanics are identical whichever persona you choose.
- What is Maestro Hub, and do I need Docker to run it?
- Maestro Hub is the web dashboard over your fleet: live voyages, orders, crews, queue depth, token usage and the human-review queue, across every registered project rather than one terminal at a time. You can also create a voyage, file an order against it, edit an order that has not started, approve or decline a plan, retry a stranded order, and pause or resume the fleet. No Docker and no Node are required: the dashboard ships inside the pip package and mso hub start serves it from your own machine on a local port. Docker remains an option for multi-container or server deployments. The Hub is a Team and Enterprise feature.
- Does the Hub do everything the CLI does?
- No, and its documentation publishes the full boundary rather than leaving you to discover a missing button. Merging a pull request, tagging, publishing, workspace migration and licence activation stay command-line and human. Each deferred action is listed alongside the CLI command that does the job today, so you can tell a deliberate omission from a defect.
- Do I have to use GitHub?
- No. Source control, CI and issue tracking resolve through a provider seam, configured per workspace with mso providers set. Maestro ships GitHub and Bitbucket for source control, GitHub Actions and Jenkins for CI, and GitHub Issues and Jira for issue tracking. The same checks and the same Releaser pass run unchanged whichever you pick, and a capability you have not configured is reported as a named skip rather than being silently assumed to be GitHub.
- Can I stop a run without losing work?
- Yes. mso pause drains the fleet: no new work is claimed, the crews already running finish their turn, and then the dispatcher exits cleanly. Nothing is killed, salvaged or retried, and mso resume carries on from where it stopped with no cleanup step. That is distinct from stopping the dispatcher outright, which is still available when you need work to stop immediately.
- What is the resident LEAD sub-agent?
- On Pro, Team, and Enterprise, a per-project LEAD sub-agent runs as a local background process that survives your terminal closing and keeps watching the fleet nights and weekends. It calls a deterministic health check, interprets what needs judgement, and performs a deliberately narrow set of reversible, idempotent housekeeping actions on its own: archiving a voyage whose pull request has merged, and sweeping the artefact records. That list is two items on purpose, and every action it takes is logged with the finding that triggered it. Anything riskier it detects and proposes for a human to run, including pruning a stale worktree. It never auto-merges a pull request, changes scope, resolves a security finding, or touches production code unattended. Those stay yours to approve. Community manages voyages and orders manually, without the resident agent.