Azure PromptFlow with Meridian

Wire PromptFlow to Meridian via a Custom OpenAI connection. All traffic flows through your LOCK gateway — no direct egress to Azure OpenAI.

1. Create the connection

In PromptFlow, add a Custom connection with type OpenAI. Point api_base at your Meridian endpoint and supply your LOCK API key.

connection:
  name: meridian_lock
  type: Custom
  configs:
    api_type: open_ai
    api_base: https://LOCK.YOUR-TENANT.meridian.security/v1
    api_key: sk-lock-...

2. Reference in your flow

Use the connection name in your LLM tool node. PromptFlow resolves api_base and api_key from the connection config automatically.

- name: llm_node
  type: llm
  source:
    type: code
    path: llm_node.jinja2
  inputs:
    deployment_name: gpt-4o
    connection: meridian_lock
  provider: OpenAI

3. Verify the route

Run a test flow. In Meridian logs you should see the request originating from PromptFlow with the model name preserved. LOCK enforces policy, rate limits, and logging before the call ever reaches Azure.

# Meridian audit log
timestamp: 2026-05-26T14:02:11Z
source:    promptflow
model:     gpt-4o
tokens:    412 prompt / 89 completion
policy:    lock-prod (passed)

Troubleshooting

  • 401 Unauthorized — confirm the api_key value matches your LOCK tenant key exactly.
  • Connection refused — verify PromptFlow can reach your Meridian api_base URL (check VNet / firewall rules).
  • Model not found — ensure the deployment_name in your flow node matches a model allow-listed in your LOCK policy.