What this map covers
A map of the concrete nodes I use when describing agent systems.
My current view: the architecture gets safer when orchestration and write access are separated on purpose.
Core concepts
- mcp
- tool-calling
- tool-schema
- agent workspaces
- workspace-state
- context-window
- scoped-agent
- process-orchestrator
- agent-trace
- agent-evaluation
- observability
- orchestrating-scoped-agents
- scoped-system-specialist-agents
Important claims
Field notes
See also
Open questions
- Which concept links are still missing from the current cluster?