Knowledge Graph

The knowledge graph is the core data structure in LoomBrain. Everything you capture becomes a node. Relationships between nodes are detected automatically and represented as links.

Nodes

A node is the fundamental unit of your graph. Each node contains:

Field	Description
Title	Canonical name of the content
Summary	2–4 sentence abstract
Body	Full markdown representation of the source
Key points	Bulleted list of extracted facts
Tags	PARA assignments (Projects, Areas, Resources)
Content type	article, tweet, video, repo, pdf, audio, note, image
Status	Current processing or review state (see below)

Node statuses

Status	Meaning
raw	Freshly created; processing may still be in progress
validated	You have confirmed the node is correct and useful
dismissed	Marked as not relevant; hidden from default views

Nodes start as raw. You can promote a node to validated from the dashboard, CLI, or MCP. Dismissing a node removes it from search results and graph views without deleting it.

Links

Links are bidirectional connections between related nodes. They are created automatically during the ingestion pipeline when the AI detects a meaningful relationship between a new node and existing ones.

Each link has:

• Source and target node IDs
• Confidence weight — how strongly the AI believed the relationship was meaningful
• Optional label — a short description of the relationship type

You can also create and delete links manually from the dashboard or via the API.

Search modes

LoomBrain supports three search modes. You can select the mode explicitly in the CLI (--mode) or dashboard filter panel.

Keyword

Uses SQLite FTS5 full-text search. Matches exact terms and their inflections. Fast and precise — best when you know specific words that appear in the content you are looking for. Respects all active filters (status, content type, PARA tag).

lb search "rate limiting exponential backoff" --mode keyword

Semantic

Uses cosine similarity on vector embeddings. Matches by meaning rather than exact wording. Useful when you remember the concept but not the specific vocabulary used in the source.

lb search "how to slow down retries" --mode semantic

Hybrid (default)

Combines keyword and semantic results using Reciprocal Rank Fusion (RRF). RRF merges two ranked lists into a single ranking without requiring score normalization. The result typically outperforms either mode alone — precise when the exact term is present, resilient when it is not.

lb search "rate limiting" --mode hybrid

Hybrid is the default for both the CLI and the dashboard search bar.

Graph visualization

The dashboard includes a graph view powered by Cytoscape.js with a force-directed layout. Nodes are rendered as circles, sized by connection count. Links are drawn as edges between them.

You can:

• Click a node to open its detail panel
• Drag nodes to rearrange the layout
• Filter the visible graph by PARA tag, content type, or node status
• Use the search bar to highlight matching nodes within the graph

The graph view is intended for exploration and pattern recognition — finding clusters of related knowledge you may not have noticed through list-based search.