Entry Linking & Knowledge Graph

Questi contenuti non sono ancora disponibili nella tua lingua.

A vault full of isolated entries is a filing cabinet. A vault with links between entries is a knowledge graph. Soleri uses Zettelkasten principles: each entry is atomic (one idea), and entries are connected by typed, directional links that describe how ideas relate to each other.

Link types

Every link has a type that describes the relationship:

Type	Meaning	Example
extends	Target builds on or refines the source	”Rate Limiting Strategy” extends “API Security Basics”
supports	Target provides evidence or foundation for the source	”Load Test Results” supports “Caching Strategy”
contradicts	Target is an opposing approach or counterpoint	”Microservices” contradicts “Monolith-First”
sequences	Source must happen before target	”Schema Migration” sequences “Deploy New API Version”

contradicts links are especially valuable — they surface anti-patterns automatically when you search for a pattern, so you see both sides of a decision.

Creating links

After capturing two related entries, connect them:

You: “Link the ‘Optimistic Updates’ entry to ‘Error Recovery UX’ as extends”

Agent: Linked: Optimistic Updates → Error Recovery UX (extends). Source now has 3 links.

You can add an optional note to explain why the link exists:

You: “Link ‘Database Indexing’ to ‘Query Performance’ as supports, note: indexes are the primary lever for read performance”

Agent: Linked with note.

Links are directional — source points to target. But you can always see both outgoing links and incoming backlinks for any entry.

Viewing links

See all connections for an entry:

You: “Show links for the ‘API Security Basics’ entry”

Agent: Entry: API Security Basics

Outgoing:

extends → Rate Limiting Strategy

extends → Authentication Patterns

Incoming (backlinks):

JWT Best Practices supports → this entry

Total: 3 links

Traversing the graph

The real power of a knowledge graph is traversal — following links to discover connected knowledge you might not have searched for directly:

You: “Traverse from ‘API Security Basics’ with depth 2”

Agent: Starting from: API Security Basics Depth: 2 hops

Connected entries:

Rate Limiting Strategy (1 hop, extends)

Authentication Patterns (1 hop, extends)

JWT Best Practices (1 hop, backlink)

Token Rotation (2 hops, via JWT Best Practices)

Redis Rate Limiter (2 hops, via Rate Limiting Strategy)

5 entries found within 2 hops

Traversal depth goes from 1 to 5 hops. Two hops is the default and usually the sweet spot — deep enough to find related context, shallow enough to stay relevant.

Suggesting links

As your vault grows, you’ll miss connections. The agent can find entries that are semantically similar and suggest links:

You: “Suggest links for the ‘Database Migration Safety’ entry”

Agent: 5 link candidates for “Database Migration Safety”:

Rollback Strategy (score: 0.89)

Zero-Downtime Deploys (score: 0.76)

Schema Versioning (score: 0.72)

Feature Flags for Data Changes (score: 0.65)

Backup Before Destructive Ops (score: 0.58)

Suggestions use FTS5 text similarity — they’re fast and don’t require an LLM call. Review the list and create links for the ones that make sense.

Finding orphans

Orphan entries have zero links — they’re disconnected from the graph. Periodic orphan checks keep your vault well-connected:

You: “Find orphan entries”

Agent: 12 entries with no links:

CSS Grid vs Flexbox

Monitoring Alert Fatigue

Git Bisect Workflow …

For each orphan, either link it to related entries or decide it’s standalone knowledge that doesn’t need connections.

Bulk linking

Backfill links

If you have a vault with many entries but few links, backfill creates links automatically using text similarity:

You: “Backfill links for orphan entries”

Agent: Backfill complete: 34 links created across 12 entries. Threshold: 0.7, max 3 links per entry.

You can control the quality threshold (higher = more selective) and max links per entry. Use dryRun: true to preview what would be created before committing.

Relink vault

For a deeper re-linking pass, relink_vault uses an LLM to evaluate entry pairs and create links with reasoning notes:

You: “Relink the vault”

Agent: Relink complete:

150 entries processed

87 candidates evaluated

42 links created (with notes)

45 pairs skipped (not meaningfully related)

Manual links preserved: 23

This is a heavier operation — it calls the LLM for each batch of candidates. Use it when you want high-quality links with explanations, not just similarity scores. Run with dryRun: true first to see the scope.

Graph statistics

Get an overview of your knowledge graph health:

You: “Show link stats”

Agent:

156 entries, 203 links, 8 orphans

By type: extends (89), supports (67), contradicts (23), sequences (24)

Links with notes: 42 (21%)

Most connected: “API Security Basics” (12 links)

A healthy vault has few orphans, a mix of link types, and contradicts links that surface trade-offs.

Building a Knowledge Base — what to capture before you start linking
Vault Branching — experiment with link reorganization on a branch before merging
Content Ingestion — ingested entries benefit from linking to existing knowledge
Sync & Export — knowledge packs export entries together with their links
Capabilities — full list of linking operations
API Reference — parameter details for link_entries, traverse, suggest_links, get_orphans, backfill_links, relink_vault

Previous: Vault Branching — experiment with knowledge changes safely. Next: Content Ingestion — feed articles and documents into your vault.