Deep Dive: Key Concepts¶
This section explains how the system works internally: the architecture of each component, its schemas, and the flow of data between services. Read it when you want to understand what happens behind the SDK calls described in Fundamentals and Advanced.
-
In-memory key-value context with NATS broadcasting.
-
Model registration, discovery and inference architecture.
-
Multi-stage planning, DAG optimisation and replanning.
-
Delegation pipeline and state synchronisation.
-
Verification schema, APIs and WebSocket updates.
-
DAG execution engine and node executors.
-
Two-phase behavior planning lifecycle and schema.
-
Messaging backbone, instance routing and load balancing.
Concepts at a glance¶
| Concept | Explains | Used by |
|---|---|---|
| Agent Context Cache | In-memory key-value context with NATS broadcasting | Agentic Memory, Context Management |
| Agent LLM Interaction | Model registration, discovery and inference architecture | Agent × AI Model Interface |
| Agent Planning System | Multi-stage planning, DAG optimisation and replanning | Behavior Controller, Prebuilt Operators |
| Agent Task Delegation | Delegation pipeline and state synchronisation | Task Delegation, Agentic Patterns |
| Agent Verification System | Verification schema, APIs and WebSocket updates | Verification System, Human-in-the-Loop |
| Agent Workflow Execution | DAG execution engine and node executors | Agentic Workflows, DSL Manager |
| Behavior Controller | Two-phase behavior planning lifecycle and schema | Behavior Controller, Advanced Behaviors |
| Communication System | Messaging backbone, instance routing and load balancing | Agents Communication, Scaling |
All architecture diagrams are collected in All Architectures.