Pulse supervision
Scheduled checks dispatch workers, merge ready PRs, close loops, and surface stuck work — automatically.
AI that does the whole job.
One conversation in, shipped work out. Maestro plans the work, dispatches parallel workers, protects your secrets, manages branches and PRs, watches CI, and learns from what went wrong last time — across code, infrastructure, business, and content.
npm install -g maestro-devops
A normal AI harness is great for one job and weak at the work around it. Maestro owns the surrounding discipline — context, routing, safety, git hygiene, verification, memory, and follow-through.
Scheduled checks dispatch workers, merge ready PRs, close loops, and surface stuck work — automatically.
Code, automation, product, business, marketing, legal, content, research, SEO, health, and reports each get their own guidance.
Risky operations can be reviewed by a second provider to reduce shared failure modes.
Credentials, scans, confirmations, and audit logs are part of the workflow — never pasted into chat.
Hosting, Git platforms, DNS, security, monitoring, deployment, payments, and comms via repeatable helpers.
Large goals split into milestones with validation, budget tracking, and automatic advancement.
SQLite-backed memory recalls past lessons on demand instead of re-deriving them every session.
Progressive disclosure and model tiering ship more useful work per dollar and per token.
Maestro runs a continuous loop. Every pass leaves the system better than it found it.
Break the goal into milestones and worker-ready tasks.
Spin up parallel workers in isolated git worktrees.
Tests, lint, CI, and review gates before anything is "done".
Branch, PR, and merge with task IDs and a clean audit trail.
Capture lessons to memory and feed them back into planning.
The same task, handled with real discipline instead of copy-paste.
| Capability | Plain AI chat | Single-agent CLI | Maestro |
|---|---|---|---|
| Context loading | Manual copy-paste | Whole repo in prompt | On-demand, progressive |
| Model spend | One model for all | One model for all | Tiered: cheap routine, strong judgement |
| Parallel work | No | Rare | Worktrees + parallel workers |
| Secrets handling | Pasted into chat | Ad hoc | Encrypted vault, never in chat |
| Git hygiene | Manual | Manual | Auto worktrees, PRs, task IDs |
| Verification | You check | Sometimes | Tests / lint / CI gates |
| Memory across sessions | None | Limited | SQLite memory, recalled on demand |
| Supervision | You watch | You watch | Pulse loop dispatches & unsticks |
Two levers do most of the work: progressive disclosure loads only the guidance a task needs, and model tiering routes routine work to fast, cheap models while reserving strong models for judgement.
You need exactly two things: a terminal and one model provider key. Maestro runs on macOS and Linux (use WSL on Windows).
npm install -g maestro-devops
maestro updateAlso available via bun and Homebrew. The installed command is maestro.
maestro model-accounts-pool add openai
# or: maestro model-accounts-pool add anthropicThat's the only hard prerequisite — secrets, GitHub auth, and services set up later, only when needed.
# in OpenCode
/onboardingThe wizard ends by running a small real task on your repo — so your first session ships something.
maestro status | Check what's installed |
maestro init | Enable Maestro features in any git repo |
maestro update | Update the framework and deployed agents |
maestro security | Full security assessment (posture, secrets, supply chain) |
maestro secret set NAME | Store a secret safely (encrypted, AI-safe) |
/pulse | Run the autonomous supervisor loop |
/skills recommend "TASK" | Find the right capability for a task |
Security, teamwork, token efficiency, and quality control — built in.