Knowledge Graph

Why a graph?

Flat notes lose relationships. A decision you made in January connects to an insight from March — but in a notes app, they're just two files. In a graph, that connection is a first-class citizen you can traverse, query, and visualize.

Structure

Your knowledge graph has two primitives:

Thoughts (nodes)

A thought is any piece of knowledge: a decision, pattern, insight, question, or observation.

[DECISION] Use Knowledge Graph Over Flat Notes (2026-03-07):
Chose a graph-based knowledge structure (nodes + edges) over
traditional flat notes. Trade-off: slightly more complex capture,
but dramatically better retrieval.

Each thought has:

• UUID — Unique identifier
• Content — The text (with [TAG] prefix)
• Embedding — 768-dimensional vector for semantic search
• Created/Updated — Timestamps
• Status — Active, archived, or completed

Edges (connections)

Edges connect thoughts with typed relationships:

[DECISION] Use SurrealDB  --implements-->  [PATTERN] Graph-native storage
[INSIGHT] Privacy wedge   --leads_to-->    [STRATEGY] Privacy-first platform
[TASK] Add rate limiting  --blocks-->      [TASK] Launch beta

How it grows

1. Manual capture — You tell your AI to save a thought
2. Hebbian auto-association — When a new thought is added, the server finds similar existing thoughts and creates relates_to edges automatically (cosine similarity > 0.5)
3. AI-suggested links — Your AI can suggest connections based on conversation context
4. Memory consolidation — Background processes identify orphan nodes, detect patterns, and strengthen frequently-traversed paths

Querying

Two search modes:

• Keyword (arx_search) — Full-text search, fast and exact
• Semantic (arx_similar) — Embedding similarity, finds conceptually related content even with different wording

Graph traversal (arx_traverse) walks edges from any starting node, revealing the neighborhood of connected ideas.