Knowledge Graphs

TeamLoop builds a temporal knowledge graph from your enterprise tools. Unlike flat document stores, knowledge graphs capture relationships, enabling rich queries and synthesis.

What is a Knowledge Graph?

A knowledge graph consists of:

• Entities - Things you know about (decisions, people, systems)
• Relationships - How entities connect to each other
• Embeddings - Vector representations for semantic search
• Temporal metadata - When knowledge was valid

Entity Types

TeamLoop supports six core entity types:

DECISION

Architectural choices, policy changes, technology selections.

Fields:

• name - The decision title
• description - What was decided
• rationale - Why this choice was made
• alternatives - Other options considered
• status - active, superseded, proposed
• event_date - When the decision was made

DOCUMENT

PRDs, ADRs, specs, meeting notes, wiki pages.

Fields:

• name - Document title
• description - Summary or excerpt
• source_url - Link to original
• event_date - Last modified date

CHANGE

Code commits, configuration changes, status updates.

Fields:

• name - Change description
• description - Details of the change
• change_type - commit, field_update, status_change
• event_date - When the change occurred

PERSON

Team members, authors, decision makers.

Fields:

• name - Person’s name
• description - Role or title

PROJECT

Initiatives, features, workstreams.

Fields:

• name - Project name
• description - Project overview
• status - active, completed, archived

COMPONENT

Systems, services, modules, infrastructure.

Fields:

• name - Component name
• description - What it does

Relationship Types

Entities connect through typed relationships:

Type	Description	Example
DECIDED_BY	Who made a decision	Decision → Person
AUTHORED_BY	Who wrote a document	Document → Person
PART_OF	Component hierarchy	Change → Document
RELATES_TO	General association	Decision → Component
SUPERSEDES	Decision lineage	New Decision → Old Decision
DEPENDS_ON	Dependency chain	Component → Component
IMPLEMENTS	Decision realization	Component → Decision

Supersession

The SUPERSEDES relationship is special - it tracks decision evolution. Query this chain with teamloop_timeline using entity_id.

Building Your Graph

Automatic Extraction

When you query with teamloop_query, results include extraction instructions. Claude will suggest entities to create.

Manual Extraction

Use teamloop_save_knowledge to explicitly create entities and relationships. Embeddings for all entities in a save operation are generated in a single batch call for efficiency.

Semantic Search

All entities are embedded for semantic search:

• Queries match by meaning, not just keywords
• “database” finds “PostgreSQL”, “MySQL”, “data storage”
• Similarity scores rank results

Temporal Metadata

Every entity tracks time:

Field	Description
event_date	When the event occurred (from source)
created_at	When added to TeamLoop
updated_at	Last modification
valid_from	Start of validity window
valid_to	End of validity (for superseded)

This enables:

• Point-in-time queries (as_of)
• Evolution tracking (from_date to to_date)
• Comparison (date_a vs date_b)

Best Practices

1. Always Include event_date

Critical for temporal queries!

2. Use Source Attribution

Include source_url, source_integration, and source_external_id for traceability.

3. Create Relationships

Isolated entities are less useful. Connect them with typed relationships.

4. Use Appropriate Types

• DECISION for choices, not documents
• DOCUMENT for artifacts, not decisions within them
• CHANGE for modifications, not new entities

Next Steps

• Subgraphs - Organize entities into collections
• Synthesis - Generate documents from knowledge
• MCP Tools Reference - All available tools