Skip to main content
PAS
Available now

Structured facts and relationships your assistant will not forget

Store and query entities, relations, and attributes. Power precise recall for planning and context-aware assistance.

See KG demoSchemas & queries

What Knowledge Graph does

Knowledge Graph stores structured facts as triples (subject-predicate-object) and enables precise queries over entities and relationships. Perfect for remembering people, places, dates, dependencies, and other structured information that powers context-aware planning.

Core capabilities

  • Register entities and relationships
  • Query by entity, relation, or attribute
  • Join with vector recall for hybrid search
  • Delete or update facts

Use cases

  • Remember people, roles, and preferences
  • Track project dependencies and ownership
  • Store system facts for change-impact analysis
Local-first by default: All facts stored on-premises. Optional ingest approvals for sensitive data.

Who benefits from Knowledge Graph

Individuals

Remember people, places, and dates with perfect accuracy

Example: "Boss = Alice; email=alice@example.com; prefers mornings" — used in scheduling and drafting.

Teams & Managers

Track project facts, dependencies, and ownership

Example: "Project X depends on Service Y; owned by Team Z" — used for impact analysis and planning.

Developers & IT

SPARQL-like or JSON query schemas for structured data

Example: Query "All services owned by Team X" with predictable JSON responses.

Security & Compliance

On-prem graph with controlled ingest

Control: All facts stored locally. Optional ingest approvals for sensitive relationships.

How it works

1

Register facts as triples

Use kg.register_item to store (subject, predicate, object) triples. Example: (Alice, manages, Bob).

2

Query by entity or relationship

Call kg.query with filters (entity, relation, attribute) to retrieve matching facts.

3

Join with vector recall

Combine structured facts from KG with semantic search from vector memory for hybrid recall.

4

Update or delete facts

Use kg.delete to remove outdated facts or update relationships.

Storage: Facts stored in a local graph database. No network calls for queries.

Example workflows

Remember people and preferences

Runs entirely offline
Input:

"Remember: Alice is my boss, email alice@example.com, prefers mornings"

Steps:
  1. kg.register_item (Alice, role, boss)
  2. kg.register_item (Alice, email, alice@example.com)
  3. kg.register_item (Alice, prefers, mornings)
Output:

Later: "Draft an email to my boss" → retrieves Alice's email and preferences

Track project dependencies

Runs entirely offline
Input:

"Project X depends on Service Y, owned by Team Z"

Steps:
  1. kg.register_item (Project X, depends_on, Service Y)
  2. kg.register_item (Service Y, owned_by, Team Z)
Output:

Query: "What depends on Service Y?" → returns Project X and impact analysis

System facts for change-impact analysis

Runs entirely offline
Input:

Store system architecture facts

Steps:
  1. kg.register_item (API Gateway, calls, Auth Service)
  2. kg.register_item (Auth Service, uses, Database X)
  3. kg.query (what calls Auth Service?)
Output:

Impact report: "Changing Auth Service affects API Gateway"

Technical details

Key tools

  • kg.register_item
  • kg.query
  • kg.delete
  • kg.list_entities
View tool schemas

Configuration

  • STORAGE_BACKEND — local graph DB (default: SQLite)
  • NAMESPACES — separate graphs for different contexts
  • MAX_TRIPLES — capacity limit (default: 1M)

Performance notes

  • Query: 10-50ms for 10K triples
  • Register: 5-20ms per triple
  • Scales to 1M+ triples on typical hardware

Observability

  • Triple count and graph size
  • Query latency and result counts
  • Register/delete throughput

Security posture

Local graph database

All facts stored on-premises. No network calls for queries.

Optional ingest approvals

Require approval for registering sensitive facts (e.g., PII, credentials).

Audit logs

All register, query, and delete operations logged locally.

Namespaces for isolation

Separate graphs for different contexts. Prevents cross-contamination.

Roadmap & status

Available

Current features

  • Register, query, and delete triples
  • Entity and relationship queries
  • Namespaces for multi-tenant use
Planned

Coming soon

  • Schema validation for triples
  • Sync adapters for distributed graphs
  • Graph visualization UI
View full roadmap

Frequently asked questions

Ready to build structured knowledge?

Get started with Knowledge Graph in minutes. All facts stored locally.

See KG demoView examplesTalk to us