exploring-knowledge-graph

Exploring the Knowledge Graph

Forgetful stores knowledge as an interconnected graph: memories link to other memories, entities link to memories, and entities relate to each other. Deep exploration reveals context that simple queries miss.

Triggers

Trigger Phrase	Operation
what do you know about X	Full knowledge graph traversal
how do I explore the knowledge graph	Graph exploration workflow
how are these concepts connected	Entity relationship traversal
give me comprehensive context on X	Deep multi-phase exploration
map out related knowledge for X	Entity discovery and memory linking

---

When to Explore

Explore the knowledge graph when:

• Starting complex work that spans multiple topics
• User asks "what do you know about X"
• Planning requires understanding existing decisions/patterns
• Investigating how concepts connect across projects
• Need comprehensive context, not just top search results

Exploration Phases

Track visited IDs to prevent cycles. Execute phases sequentially.

Phase 1: Semantic Entry Point

execute_forgetful_tool("query_memory", {
  "query": "<topic>",
  "query_context": "Exploring knowledge graph for comprehensive context",
  "k": 5,
  "include_links": true,
  "max_links_per_primary": 5
})

Collect: primary_memories + linked_memories (1-hop connections).

Phase 2: Expand Memory Details

For key memories, get full details:

execute_forgetful_tool("get_memory", {"memory_id": <id>})

Extract: document_ids, code_artifact_ids, project_ids, additional linked_memory_ids.

Phase 3: Entity Discovery

Find entities in discovered projects:

execute_forgetful_tool("list_entities", {
  "project_ids": [<discovered project ids>]
})

Phase 4: Entity Relationships

For relevant entities, map relationship graph:

execute_forgetful_tool("get_entity_relationships", {
  "entity_id": <id>,
  "direction": "both"
})

Relationship types: works_for, owns, manages, collaborates_with, etc.

Phase 5: Entity-Linked Memories

For each entity, find all linked memories:

execute_forgetful_tool("get_entity_memories", {
  "entity_id": <id>
})

Returns {"memory_ids": [...], "count": N}. Fetch any new memories not already visited.

Presenting Results

Group findings by type:

Memories: Primary (direct matches) → Linked (1-hop) → Entity-linked (via entities)

Entities: Name, type, relationship count, linked memory count

Artifacts: Documents and code snippets found via memory links

Graph Summary: Total nodes, key themes, suggested follow-up queries

Depth Control

• Shallow (phases 1-2): Quick context, ~5-15 memories
• Medium (phases 1-4): Include entities and relationships
• Deep (all phases): Full graph traversal, comprehensive context

Match depth to task complexity. Start shallow, go deeper if context insufficient.

When to Use

Use this skill when:

• Starting complex work spanning multiple topics
• User asks "what do you know about X"
• Planning requires understanding existing decisions and patterns
• Investigating how concepts connect across projects

Use using-forgetful-memory instead when:

• Creating or querying individual memories
• Simple semantic search is sufficient

Use curating-memories instead when:

• Updating, obsoleting, or linking specific memories
• Cleaning up duplicate or stale content

---

Anti-Patterns

Avoid	Why	Instead
Running all 5 phases for simple queries	Wastes tokens on unnecessary traversal	Start shallow (phases 1-2), go deeper only if needed
Not tracking visited IDs	Causes infinite cycles in graph traversal	Maintain a visited set, skip already-seen nodes
Expanding every linked memory	Exponential blowup on dense graphs	Focus on high-importance memories (7+)
Skipping entity phases when entities exist	Misses cross-project connections	Check for entities in Phase 3 before skipping
Presenting raw results without grouping	Overwhelming and unstructured	Group by type: memories, entities, artifacts

---

Verification

After graph exploration:

• Entry point query returned relevant results
• No cycles encountered (visited IDs tracked)
• Depth matched task complexity (shallow/medium/deep)
• Results grouped by type for readability
• Follow-up queries identified if context was insufficient

---

Efficiency Tips

• Check truncated flag from query_memory (8000 token budget)
• Skip Phase 3-5 if no entities exist in discovered projects
• Use project_ids filter to scope exploration
• Stop expanding when hitting diminishing returns