About AI Workflows
- Last Updated: November 20, 2026
- 4 minute read
- Semaphore
- Version 5.13
- Documentation
AI Workflows in Semaphore Knowledge Model Management (KMM) use artificial intelligence (AI) to help you create knowledge models and enrich existing content.
Starting with Semaphore 5.13.0, AI Workflows provide two capabilities:
- Model from Spec: Create a knowledge model from a specification provided through a prompt and optional supporting documents.
- Enrichment: Extend an existing knowledge model, concept scheme, concept, or concept subtree with AI-generated content.
AI Workflows can reduce the manual effort involved in building and maintaining knowledge models. However, AI-generated content should always be reviewed for accuracy, relevance, and consistency with your modeling requirements.
Choose an AI Workflow
The workflow you choose depends on whether you want to create a new model or expand existing content.
| Model from Spec | Enrichment | |
|---|---|---|
| Purpose | Create a new knowledge model | Extend an existing model |
| Starting point | Prompt and optional documents | Existing model content, prompt, and optional documents |
| Scope | New knowledge model | Entire model, concept scheme, concept, or concept subtree |
| Typical use | Generate an initial model from requirements | Add or improve concepts, relationships, labels, and metadata |
| Review options | Create content in the master model or within a task | Generated changes are applied automatically to the selected model or task |
Model from Spec
Model from Spec helps you create a knowledge model from a description of the information you want to represent.
You can provide a prompt describing the model's purpose, subject area, and expected structure. You can also upload supporting documents to provide additional context.
Choose how to create the model
Model from Spec supports two approaches:
- Create a new model: Generate content directly in the new model's master version.
- Create a new model with a task: Generate content within a task so that you can review the results before incorporating them into the master model.
You can also select Structure only to generate the model structure without generating the concept taxonomy.
When to use Model from Spec
Use Model from Spec when you want to:
- Create an initial knowledge model from requirements or specifications.
- Generate a starting taxonomy for a new subject area.
- Use existing documents as context for model generation.
- Establish an initial model structure that you can review and refine.
For step-by-step instructions, see Creating a Model from a Specification.
Enrichment
Enrichment helps you extend or improve content in an existing knowledge model.
You can provide a prompt describing the changes you want and optionally upload supporting documents to guide the generation process.
Choose what to enrich
You can run Enrichment against:
- An entire knowledge model: Generate content within the selected model.
- A concept scheme: Enrich content within the selected scheme.
- A selected concept: Enrich a specific concept without including its descendants.
- A selected concept and its descendants: Enrich a concept subtree.
The available enrichment options depend on the selected scope.
Choose what to generate
The Enrichment Agent can generate concepts, relationships, labels, and metadata values, such as definitions. You can select multiple generation options in a single enrichment operation, depending on the selected scope.
For example, you can ask the Enrichment Agent to create concepts for renewable energy sources, add relationships between them, and generate definitions.
When to use Enrichment
Use Enrichment when you want to:
- Expand an existing taxonomy with new concepts.
- Add concepts within a particular concept scheme.
- Improve the labels or definitions of existing concepts.
- Generate relationships between concepts.
- Develop a specific part of a larger knowledge model.
Unlike Model from Spec, which creates a new model, Enrichment works with content that already exists.
For step-by-step instructions, see Enriching an Existing Model.
Before you begin
To use AI Workflows:
- An administrator must configure the AI Model Builder and a supported Large Language Model (LLM) provider in Semaphore Studio.
- You must have the appropriate permissions to create or modify knowledge models.
For configuration instructions, see Configure the AI Model Builder.
Review AI-generated content
AI-generated content may contain inaccuracies, duplicate or unsuitable concepts, or relationships that do not reflect the intended meaning of your model.
Review generated content before using it in production.
Pay particular attention to:
- Concepts: Check that generated concepts are relevant, correctly named, and not duplicates.
- Relationships: Verify that relationships reflect the intended meaning and do not conflict with the existing model structure.
- Labels: Review preferred and alternative labels for accuracy and consistency.
- Metadata: Check generated definitions and other values for correctness and completeness.
Review changes before committing them
Model from Spec can generate content within a task, allowing you to review the results before committing them to the master model.
Important: Enrichment applies generated changes automatically during execution. It does not ask you to approve individual changes.
When possible, run Enrichment within a task so that you can inspect the results before committing them to the master model.
If an enrichment operation reports an error, inspect the model before running it again. Some changes may already have been applied.
AI Workflows and the MCP Server
AI Workflows provide built-in AI-assisted functionality through the KMM user interface.
The Semaphore MCP Server provides tools that external MCP clients and custom AI agents can use to perform supported KMM operations.
Although the built-in AI Workflows agents use MCP internally, the MCP Server does not expose Model from Spec or Enrichment as directly callable tools for external clients.
For information about integrating external AI clients with Semaphore, see Using the Semaphore MCP Server.