Click any box in the map to the left.
This map shows how a raw language model becomes a capable assistant. Read it top-to-bottom:
- The model predicts text — nothing more, on its own.
- Context window is its short-term memory (what it can "see" now).
- Tools let it act; MCP is the universal way to plug in tools & data.
- Skills teach it a procedure; subagents let it delegate.
- Memory persists facts across sessions; RAG grounds answers in real data.
- The agentic loop ties it together into something that does things.
The connecting lines show how each layer depends on the one above.
