Multiple AI agents, each with distinct roles and capabilities, working together to solve tasks no single model could handle alone.
ORCHESTRATOR
Plans & Delegates
ROUTER
Dispatches Tasks
SPECIALISTS
Execute & Return
CRITIC
Evaluates Output
TOOL AGENT
Calls APIs & DBs
— Agent Types
🧠
Orchestrator
The Planner
// Master coordinator
Receives the top-level goal, breaks it into subtasks, decides which agents to invoke and in what order. Holds the "big picture" and synthesizes all outputs into a final response.
Classifies incoming tasks and routes them to the most capable specialist. Uses intent detection, confidence scoring, or model capability profiles to choose the right agent for each job.
Fine-tuned or prompted for a specific domain — code generation, data analysis, copywriting, SQL, legal reasoning. Multiple specialists run in parallel. Each returns structured output back to the orchestrator.
Reviews outputs from specialist agents for accuracy, hallucinations, format compliance, and goal alignment. Can trigger re-runs, request revisions, or score confidence. Acts as the system's self-correction layer.
Bridges AI reasoning and real systems. Calls REST APIs, queries databases, runs code, reads files, sends webhooks. Returns raw data or side effects back to the requesting agent in the chain.
API callsCode executionDB queriesWebhook dispatch
💾
Memory Agent
The Rememberer
// Persistent context layer
Manages short-term working memory (conversation state) and long-term episodic/semantic memory (vector stores). Retrieves relevant context on demand and writes important outcomes back for future use.