The 13 Agents of a Swarm

Approves, blocks, or halts

Enforces system policies, risk limits, and safety rules; can approve, block, or halt agent actions before execution.

Watches the whole system

Oversees the agent system; monitors performance, cost, safety, loops, and failures.

Turns goals into plans

Converts goals into executable plans, selects strategies, defines workflows, and adapts when conditions change.

Routes work to the right place

Selects the best agents, models, tools, and workflows for each task while optimizing cost, speed, and quality.

Orchestrates execution

Orchestrates execution across specialized agents, manages dependencies, tracks progress, and ensures objectives are completed.

Remembers and retrieves

Manages short-term, long-term, and compressed memory; controls storage and retrieval.

Critiques and improves

Critiques outputs and improves them through feedback loops.

Does the domain work

Domain-specific agents operating under managers.

Calls tools and APIs

Executes explicit functions through tools and APIs.

Triggers external functions

Emits structured outputs that trigger predefined external functions.

Connects to outside services

Uses sandboxed extensions to access external services.

Packages and hands off

Packages outputs into polished deliverables for humans or clients.

Models outcomes first

Models possible outcomes and tests strategies before execution.