Concepts Copy

The Memory Graph

A high-performance, deterministic knowledge graph linking files, code symbols, team decisions, and operational policies with sub-300ms recall.

BetterContext abandons flat vector stores in favor of an active, multi-relational knowledge graph. Nodes represent grounded facts and code constructs, while edges capture functional, architectural, and causal dependencies.

Node & Entity Ontology

The graph is populated with four distinct node classes:

Code AST Nodes

Functions, classes, schemas, interfaces, and module exports parsed directly from source files.

Document & Specification Nodes

Markdown documentation, ADRs (Architectural Decision Records), OpenAPI specs, and runbooks.

Operational Event Nodes

Git commits, release tags, resolved incident post-mortems, and pull request review discussions.

Policy & Invariant Nodes

Negative constraints (e.g. "Never store raw credit card numbers") and regulatory compliance rules.

Confidence Half-Life Decay

Information in engineering organizations changes rapidly. Code is refactored, APIs are deprecated, and team ownership shifts. Static vector databases treat a document written 3 years ago with the same authority as one committed 10 minutes ago.

BetterContext applies a continuous exponential decay model to edge weights:

Decay Formula: w(t) = w_0 · 2^(-Δt / t_half)
• Active nodes reinforced by recent commits, agent citations, or human verification maintain high confidence (0.95+).
• Orphaned or unreferenced documentation decays in relevance, preventing agents from citing stale or misleading patterns.

Traversing the Graph via Python SDK

from bettercontext import BetterContextClient

client = BetterContextClient()

# Inspect neighbors and dependencies of a specific service
node = client.graph.get_node(id="auth-middleware-v2")
print(f"Node Type: {node.type}")
print(f"Confidence Score: {node.confidence}")

# List inbound and outbound dependencies
for edge in node.edges:
    print(f"-> {edge.relation} -> {edge.target_node.title} (Weight: {edge.weight})")

Next Steps

Was this page helpful?

ON THIS PAGE

Overview
Node Ontology
Confidence Decay
Graph Traversal
Next Steps