ArtificialIntelligence.How

Checked August 2026

Knowledge graph construction microscope

Build and interrogate a knowledge graph with Graphify

A useful knowledge graph is not a spray of circles. It is a typed, evidence-bearing model of entities and relationships. This laboratory walks an inspectable fixture through detection, extraction, provenance, graph building, community analysis, and auditable traversal.

Filter EXTRACTED, INFERRED, and AMBIGUOUS relations; select any node; then query, path, and explain how instructions, evidence, tools, policy, transport, and memory connect. Every count comes from the visible fixture, and an unsupported path remains empty.

A precision physical knowledge graph instrument with distinct node communities, typed material edges, bridge nodes, and source cards
Knowledge graph microscope · Typed paths with evidence and provenance
Manipulate
Build stage · provenance · path question · output
Inspect
Node type · relation · community · source location
Honest boundary
Graph structure ≠ semantic truth
Graph construction microscopeTyped relations · evidence · deterministic traversal
Current build stageQuery + explain

Traverse relationships, inspect paths, and explain every hop from its evidence.

Inspectable teaching graph10 revealed nodes · 11 revealed typed edges · 3 build-time communitiesThe stage control progressively reveals the exact fixture artifact. Provenance filters never silently recompute its build-time community assignments.
Context
Execution
State
EXTRACTED · parser or exact source factINFERRED · evidence-backed interpretationAMBIGUOUS · unresolved alternative
explain resultWhat path can turn a request into durable state?6 traversed relations
→
→
→
→
→
→
01
FEEDS

request → compiler

EXTRACTEDruntime/context.ts:42
02
ASSEMBLES_FOR

compiler → model

EXTRACTEDruntime/agent.ts:88
03
PROPOSES_CALL

model → schema

EXTRACTEDruntime/agent.ts:104
04
VALIDATED_BY

schema → policy

EXTRACTEDruntime/policy.ts:31
05
AUTHORIZES

policy → handler

EXTRACTEDruntime/policy.ts:48
06
MAY_WRITE

handler → memory

INFERREDruntime/memory.ts:63
Bridge / “god” nodesContext compiler · degree 4

Degree is counted from the currently visible edges. High degree can reveal a bridge—or noisy configuration—so rank is a review lead, never proof of importance.

Community detection3 build-time fixture communities

The full fixture stores a precomputed partition. Query-time provenance filters preserve those assignments instead of pretending to rerun Leiden in the browser.

Graph vs vector retrievalPaths answer relationships; vectors answer conceptual similarity.

A hybrid system can use both, but a similarity match must not be mislabeled as an explicit graph edge.

Graphify-style workflowExtract, type, build, detect communities, then query, path, and explain.

Portable graph artifacts keep source and provenance attached so an answer can be audited.

Honest limitationParsers find structural facts; semantic claims need separate evidence and review.

God-node rankings can surface configuration noise. Dangling endpoints and identity collisions belong in graph health, not under the rug.

Audit every revealed relation

requestFEEDScompilerEXTRACTEDruntime/context.ts:42

requestACTIVATESskillINFERREDskills/research/SKILL.md:1

skillGUIDEScompilerEXTRACTEDruntime/context.ts:51

evidenceRETRIEVED_INTOcompilerEXTRACTEDruntime/retrieval.ts:70

compilerASSEMBLES_FORmodelEXTRACTEDruntime/agent.ts:88

modelPROPOSES_CALLschemaEXTRACTEDruntime/agent.ts:104

schemaVALIDATED_BYpolicyEXTRACTEDruntime/policy.ts:31

policyAUTHORIZEShandlerEXTRACTEDruntime/policy.ts:48

handlerTRANSPORTED_BYserverEXTRACTEDmcp/client.ts:55

serverRETURNS_TOmemoryEXTRACTEDruntime/memory.ts:63

handlerMAY_WRITEmemoryINFERREDruntime/memory.ts:63

Primary research

Inspect the mechanism at its source.

Project architecture and terminology were checked against official documentation on August 10, 2026. Research papers are linked to their original publication records.

Build, query, path, and explain workflowGraphify documentationOriginal source ↗Current implementation and schemaGraphify open-source repositoryOriginal source ↗Typed graphs and evidence modelIntroducing GraphifyOriginal source ↗Graph-based retrieval researchFrom Local to Global: GraphRAGOriginal source ↗

Keep experimenting

Change the layer, keep the system visible.

AI Lab indexSee every interactive laboratoryBrowse the suite →Transformer microscopeGo beneath the agent runtime into the modelInspect the matrices →