Data Model

Context Chain stores everything in a Memgraph graph database. Here's the complete data model.

Node Types

CodeEntity

Represents a code element in your repository. Created by Joern CPG analysis.

Property	Type	Description
name	string	Entity name (function name, file path, etc.)
type	enum	service, file, function, api_endpoint
scope	string[]	Module/service tags for filtering
repo	string	Which repository this belongs to

DecisionContext

A design decision extracted from an AI coding session, code analysis, or manual input.

Property	Type	Description
id	string	Unique identifier
content	string	Full natural language description
summary	string	Short summary (for progressive disclosure)
keywords	string[]	Extracted concept terms (Slot 2)
scope	string[]	Module tags
embedding	float[]	Semantic vector (Slot 5)
created_at	datetime	When created
updated_at	datetime	Last modification
owner	string	Who produced this decision
session_id	string	Source AI chat session
commit_hash	string	Code version at time of decision
source	enum	ai_chat, meeting, manual
confidence	enum	owner_confirmed, ai_inferred, auto_generated, refined
staleness	enum	active, stale, archived

AggregatedSummary

Module-level summaries generated by the refinement pipeline. Groups multiple fine-grained decisions into a higher-level overview.

INFO

Not yet implemented. Planned for the refinement pipeline.

Edge Types

Code Structure Edges

Generated by Joern CPG analysis and LLM inference.

Edge	From → To	Meaning
CONTAINS	CodeEntity → CodeEntity	Hierarchy: service → file → function
CALLS	CodeEntity → CodeEntity	Function calls another function
DEPENDS_ON_API	CodeEntity → CodeEntity	Cross-service API dependency (LLM-inferred)

Decision Anchor Edges

Connect decisions to the code they describe.

Edge	From → To	Meaning
ANCHORED_TO	DecisionContext → CodeEntity	Precise anchor (function-level)
APPROXIMATE_TO	DecisionContext → CodeEntity	Fuzzy anchor (file/directory-level, with confidence property)

Decision Relationship Edges

Connect decisions to each other, forming a decision graph.

Edge	From → To	Meaning
CAUSED_BY	Decision → Decision	A was caused by B
LEADS_TO	Decision → Decision	A led to B
CONFLICTS_WITH	Decision → Decision	Tension / trade-off between A and B
SUPERSEDES	Decision → Decision	B replaces A
DEPENDS_ON	Decision → Decision	A assumes B holds
CO_DECIDED	Decision → Decision	Made together in the same session
SUMMARIZES	AggregatedSummary → DecisionContext	Summary covers these decisions

Internal Tracking Edges

Used by the pipeline, not exposed to consumers.

Edge	From → To	Meaning
PENDING_COMPARISON	Decision → Decision	Not yet checked for relationships

TIP

PENDING_COMPARISON edges use an inverted tracking pattern: instead of recording "already compared, no relationship" (which pollutes the graph), the system creates PENDING edges for pairs that haven't been compared yet. As comparisons run, PENDING edges are deleted regardless of result — only meaningful relationship edges remain.

Graph Visualization

Open Memgraph Lab at http://localhost:3000 to browse the graph interactively.

Useful Cypher queries:

// All decisions anchored to a specific file
MATCH (d:DecisionContext)-[:ANCHORED_TO]->(c:CodeEntity {type: 'file', name: 'auth.ts'})
RETURN d, c

// Decision causal chain
MATCH path = (d1:DecisionContext)-[:CAUSED_BY|LEADS_TO*1..3]-(d2:DecisionContext)
WHERE d1.id = 'some-decision-id'
RETURN path

// Functions with no decisions (coverage gaps)
MATCH (c:CodeEntity {type: 'function'})
WHERE NOT (c)<-[:ANCHORED_TO]-(:DecisionContext)
RETURN c.name, c.repo
ORDER BY c.repo