An orchestration framework for AI agents to pass tasks through explicit queues. Build pipelines in YAML, run locally with SQLite, and power them with Multi-LLM.

_{📖 한국어}

aqm

Build AI agent teams in YAML. No code. No API keys. Just pipelines.

An orchestration framework where multiple AI agents pass tasks through explicit queues — or discuss in real-time sessions until consensus. Define once, run anywhere, share with anyone.

  [user] ──input──► [planner] ──► [reviewer] ──approve──► [design_session] ──► [implementer]
                        ▲              │                    ┌──┬──┬──┐
                        └── reject ────┘                    ▼  ▼  ▼  ▼  round-robin
                        └── ask user ──►[user]             [arch][sec][fe]  until consensus

Why aqm?

A single AI agent writes code and reviews it with the same bias. It can't catch its own blind spots.

aqm gives you a team — each agent has a dedicated role, a separate prompt, and optionally a different LLM. A quality gate rejects bad output automatically. A session lets agents debate before deciding.

# One YAML file. That's the entire pipeline.
agents:
  - id: developer
    runtime: gemini
    system_prompt: "Implement: {{ input }}"
    handoffs: [{ to: reviewer }]

  - id: reviewer
    runtime: gemini                    # Different LLM catches different bugs
    system_prompt: "Review for security: {{ input }}"
    gate:
      type: llm
      prompt: "Is this production-ready?"
      max_retries: 3                   # Auto-reject → retry up to 3 times
    handoffs:
      - { to: deployer, condition: on_approve }
      - { to: developer, condition: on_reject }

  - id: deployer
    runtime: gemini
    context_strategy: none             # 85% token savings — no context needed
    system_prompt: "Deploy: {{ input }}"

pip install aqm && aqm init && aqm run "Add JWT authentication"

What makes aqm different