Edge Types

Every relationship in the graph is typed. The type is not decoration — it changes how your agent should use the information.

The four types

Type	Meaning	Agent use
REQUIRES	Hard prerequisite — must exist first	Plan sequencing, gap detection
ENABLES	Unlocks a capability downstream	Optimization paths
RELATES_TO	Conceptual proximity, not dependency	Disambiguation, context
IMPLEMENTS	Concrete instantiation of an abstraction	Architecture mapping

---

REQUIRES

A REQUIRES edge means one concept cannot function without the other. Use it to sequence deployments, detect gaps, and plan what must be in place before you build.

get_prerequisites("Llm Nim", "nvidia-nim")

→ LLM NIM → Triton Inference Server Backend → NIM Weight Cache
  → TensorRT-LLM Engine → Model Weights → GPU Profile
  → NGC Container Registry → NVIDIA GPU → NGC API Key

Every node in that chain has a REQUIRES relationship to the one before it. If any link is missing, the deployment fails.

---

ENABLES

A ENABLES edge means the source concept makes something possible downstream — but that downstream concept is not required for the source to function.

query_ckg("Tensorrt Llm", "nvidia-tensorrt-triton", depth=2)

→ TensorRT-LLM enables:
    Triton Inference Server → NIM Microservice Runtime

TensorRT-LLM can run without NIM. But NIM performs significantly better with TensorRT-LLM underneath it. That's ENABLES — an optimization path, not a hard gate.

---

RELATES_TO

A RELATES_TO edge means two concepts work in the same space. One does not block the other — but they are conceptually adjacent, often co-deployed, or commonly confused.

Use RELATES_TO for context and disambiguation. Do not use it for sequencing.

Don't confuse proximity with dependency

A RELATES_TO edge between TensorRT-LLM and cuBLAS means they work together — not that one requires the other. Your agent should treat RELATES_TO as context, not a prerequisite.

---

IMPLEMENTS

A IMPLEMENTS edge connects an abstraction to its concrete instantiation.

NIM Microservice [SERVICE] --IMPLEMENTS--> Inference Service [ABSTRACTION]

Use IMPLEMENTS for architecture mapping — understanding what a concept concretely is, not what it depends on.

---

Silence is signal

If an edge isn't declared, the traversal returns nothing rather than hallucinating a path.

That silence is signal. A missing edge means the relationship has not been declared — either because it doesn't exist, or because it hasn't been reviewed yet. Check the confidence states below to calibrate.

---

Confidence states

State	Meaning	How to use
confidence: high	Reviewed, cross-referenced with NVIDIA docs	Trust for planning
confidence: null	Plausible, not yet audited	Scaffold — verify before production
confidence: low	Flagged as uncertain	Treat as a hint, not a fact

A REQUIRES with confidence: low is a hypothesis. A RELATES_TO with confidence: null is still useful context. The type and the confidence together tell the full story.

---

Corrections

Spotted a wrong edge? A RELATES_TO that should be REQUIRES? Edge corrections are the highest-value contribution to the graph.

Open an issue or PR at github.com/Yarmoluk/ckg-nvidia-ai — see Contributing for the review format.