Agent Blueprints

Overview

The Enterprise Kit ships with 5 pre-built agent blueprints designed to run on Azure AI Foundry Agent Service. Each blueprint deploys as a first-class Foundry Agent Application (ARM resource) with manifest-driven configuration, Foundry Toolbox integration, Foundry Guardrails for content safety, and Azure AI Search for RAG grounding. Blueprints are published to Agent 365 for cross-platform governance and enterprise discovery. Each includes YAML manifests, system prompts, toolbox configuration, and evaluation criteria.

Blueprint Catalog

Manifest Structure

Each agent is defined by a YAML manifest following the Agent Manifest Specification:

# agent-manifest.yaml
apiVersion: foundry.azure.com/v1
kind: AgentManifest
metadata:
  name: knowledge-agent
  version: "1.0.0"
  labels:
    class: retrieval
    tier: production
spec:
  model:
    deployment: gpt-4o
    temperature: 0.3
    max_tokens: 4096
  tools:
    - name: azure_ai_search
      type: azure_ai_search
      connection: ai-search-connection
    - name: file_search
      type: file_search
  instructions_file: prompts/system-prompt.md
  evaluation:
    pack: builtin
    threshold: 0.8
  governance:
    policy_pack: default
    certification_level: standard
    data_boundary: us-east

Customizing a Blueprint

1. Copy the blueprint directory: cp -r agent-templates/knowledge-agent my-agent
2. Edit agent-manifest.yaml with your model, tools, and policies
3. Customize prompts/system-prompt.md for your domain
4. Map toolbox config in toolbox-configs/
5. Run evaluation: python sdk-starters/python/eval_runner.py --manifest my-agent/agent-manifest.yaml
6. Deploy via CI/CD pipeline or SDK client
7. Publish to Agent 365: python trust-extensions/agent365_publisher.py --manifest my-agent/agent-manifest.yaml

Agent Classes

The class label determines the agent's runtime behavior, tool access patterns, and governance requirements:

Azure Services per Blueprint