Knowledge base

The knowledge base expands each definition from the background article into its own entry, with deeper explanation, patterns, trade-offs and links. It is being written now. Planned entries:

Concepts

Layer	Entries
Core stack	model · agent · harness · agent loop · graph engineering · orchestrator · control plane · data plane
Tools and protocols	tool · tool calling · MCP · A2A
Context	context engineering · memory · compaction · sub-agent
Safety and quality	sandbox · guardrail · human-in-the-loop · eval · LLM-as-judge
Infrastructure	model gateway · durable execution · observability

Patterns

supervisor / worker · handoff · pipeline · fan-out / fan-in · evaluator / optimiser · reconciler

Tools and landscape

One entry per open-source project: what it is, which layer it belongs to, its licence and when to use it. A landscape page will map every project to the stack.

Contributing

Every entry starts from a template in the repository's templates/ folder and carries a small metadata header (id, layer, tags, related, updated), so the KB can be searched and linked by tools as well as people. See CONTRIBUTING.md in the repository.