Kyb/DocsBeta
Start free

AI Assistant

The Kyb Analyst is an AI-powered conversational assistant that helps you build your architecture graph through natural language. Describe your systems, and the AI will classify entities, propose relationships, and help you fill gaps.

Opening the Chat

In the Explorer, click the chat icon in the right panel to open the Kyb Analyst. The chat is context-aware: it knows about the entities and relationships already in your graph and the current view you are looking at.

What the AI Can Do

Create entities

Describe a system, service, or process and the AI will classify it into the right ArchiMate type, assign a confidence score, and propose it for your confirmation.

Create relationships

Tell the AI how systems connect ("the payment service calls the billing API") and it will propose the appropriate ArchiMate relationship.

Search the graph

Ask about existing entities ("what systems does the Order Service depend on?") and the AI will query your graph and return results.

Analyse gaps

Ask the AI to find orphan entities, missing descriptions, incomplete ArchiMate patterns, or entities without relationships.

Inspect views

Ask about the current view ("what's in this view?") or list all available views.

How Classification Works

When you describe a system, the AI maps your description to the Kyb ontology: a taxonomy of ArchiMate types and Kyb specializations. The classification process considers:

  • The name and description you provide.
  • Context from existing entities and relationships in the graph.
  • ArchiMate validity rules to ensure the type and its relationships are semantically correct.

Each classification carries a confidence score. Higher scores mean the AI is more certain about the type assignment. You can always override the classification during review.

Confirming Proposals

The AI never writes to your graph without your permission. When it proposes a new entity or relationship, you will see a confirmation prompt. You can:

  • Confirm -- Accept the proposal and add it to the graph.
  • Edit and confirm -- Modify the name, type, or other properties before adding.
  • Skip -- Reject the proposal. It will not be added.

Tips for Better Results

  • Be specific about what each system does, not just its name. "Our payment processing service that handles Stripe webhooks and manages subscription lifecycle" gives much better classification than just "PaymentService".
  • Describe relationships explicitly: "The API Gateway routes requests to the Order Service, which reads from the Orders Database."
  • Use gap analysis after initial population to find missing connections and descriptions.
  • You can have a conversation: follow up, correct, and refine. The AI maintains context within the chat session.

Token Usage

The AI Assistant uses tokens (a measure of AI processing). The free tier includes a monthly token budget. You can see your current usage in the token meter displayed near the chat input. For unlimited usage, see API Keys & Usage to set up your own API key.

Next: Views & Dashboards

Learn how to organise your graph into views in Views & Dashboards.