Skip to content
← Back

Maestro

Open Source

An orchestration layer for AI coding agents that implements research-backed principles rather than unlimited agent scaling. Point your IDE at the repo and augment your existing agent config — no dependencies, no SDK, no configuration.

GitHub →

Most multi-agent frameworks add agents to make things faster. The research says the opposite: adding agents usually makes things worse. Maestro implements the architecture that peer-reviewed research actually points to, a thin routing layer that only activates multi-agent coordination when the task demands it. No dependencies. No config. No SDK.

Architecture

MAESTRO ARCHITECTURE Your Task DECISION GATE single or multi-agent? Single Agent most tasks Planner Specialist A Specialist B Specialist C Cross-Talk Routing Staff Engineer adversarial review

The Decision Gate is the key. Most tasks stay single-agent with zero overhead. Multi-agent coordination only activates when the task genuinely benefits from parallel execution or adversarial review. This bias is intentional: the research shows coordination overhead makes simple tasks worse, not better.

Portable Core, Thin Adapters

Maestro separates portable orchestration doctrine from runtime-specific adapters. The core logic lives in AGENTS.md and works across any agent runtime. Each runtime gets a thin wrapper that imports the shared doctrine and adds only what’s specific to that environment.

PORTABLE CORE + RUNTIME ADAPTERS AGENTS.md Portable Orchestration Doctrine @import @import @import @import CLAUDE.md subagents, teams, hooks GEMINI.md execution, precedence AGENTS.md Codex (direct use) .cursorrules full embed (no imports) Claude Code Gemini CLI Codex Cursor New runtime = one thin adapter file that imports AGENTS.md

Claude Code: Subagents vs Agent Teams

Claude Code offers subagents and agent teams. Maestro’s Claude adapter automatically routes to the right one:

  • Subagents (default): narrow, independent tasks where only the result matters
  • Agent teams: long-running parallel workstreams where peer-to-peer coordination is materially useful

Agent teams are experimental and Claude Code-only. Maestro’s portable core uses the general concept of “specialists” which each runtime maps to its own execution model.

Why Not CrewAI / LangGraph / AutoGen?

Maestro is not a framework. It’s an orchestration layer for AI coding agents that already exist. You don’t write agent code; you copy a couple of files and your existing agent gains multi-agent capabilities.

If you need a standalone multi-agent application with custom tools, APIs, and deployment pipelines, use a framework. If you want your AI coding agent to handle complex tasks better without changing your workflow, use Maestro.

Research Foundation

The architecture is grounded in 700+ sources across computer science, library science, safety engineering, and knowledge theory.

KEY RESEARCH FINDINGS Failure rate 41-87% of multi-agent systems fail MAST Root cause 79% from coordination, not capability MAST Optimal count 3 agents outperform 7 DyLAN Cost savings 53-68% token cost reduction DyLAN Reasoning loss 39-70% degradation under multi-agent DeepMind Handoff quality Structured: 3.9/4 vs unstructured: 2.1/4 MetaGPT Fault tolerance 0.3% degradation under failure GTD

Read the full analysis in Why Your Multi-Agent AI System Keeps Failing.