Skip to main content

Azure OpenAI Adapter

A Transform Custom node that sends the message reaching it to an Azure OpenAI model deployment and pushes the model's response downstream.

Part of the Linkiir Adapters package — see requesting the package.

What it does

When a message arrives, the node combines your system prompt with the message content, calls your Azure OpenAI deployment, and pushes the response onward. The endpoint URL you configure decides which Azure API shape is used, so pointing at a different deployment style needs no other change.

upstream node → Azure OpenAI Adapter → your next node

Azure OpenAI deployment
Decide what content may leave your environment

This node sends message content to Azure. Before you turn Live Mode on, confirm that sending this data to your Azure resource is permitted under your organisation's agreements and your patient-data policy, and constrain what reaches the node upstream. See Security.

Before you start

From the Azure portal or Azure AI Foundry:

  • Deploy a model and note its deployment name.
  • Copy the deployment's endpoint URL, including the API version query parameter.
  • Choose an authentication mode:
ModeWhat you need
API KeyThe key from the deployment page
Entra IDAn app registration granted the Azure OpenAI user role on the resource, plus its client ID, tenant ID, and client secret

API Key is quicker to set up. Entra ID avoids a long-lived key and is the better fit for production.

Set it up

  1. Open the Azure OpenAI node in the Workflow Builder and click Edit.

  2. Set Authentication Mode, which decides which credential fields appear:

    ModeFill in
    API KeyAPI Key
    Entra IDAzure Client ID, Azure Tenant ID, Azure Client Secret
  3. Paste the deployment endpoint into Model URI, and set Model Name to the deployment name.

  4. Write the System Prompt: the standing instructions for the model, covering tone, output format, and anything it must not do.

  5. Set Live Mode off, then Save. Send one message through and read the log to confirm the request is shaped as you expect.

  6. Turn Live Mode on and send another message.

Configuration reference

FieldTypeDefaultPurpose
Authentication ModelistAPI KeyChooses between a static API key and Entra ID
Model URIstring(empty)Full deployment endpoint URL, including the API version query parameter
API Keypassword(empty)Azure OpenAI key. Shown only in API Key mode
Azure Client IDstring(empty)Application (client) ID. Shown only in Entra ID mode
Azure Tenant IDstring(empty)Directory (tenant) ID. Shown only in Entra ID mode
Azure Client Secretpassword(empty)Client secret. Shown only in Entra ID mode
System Promptstring(a healthcare assistant prompt)Standing instructions: tone, format, constraints
Model Namestring(empty)Deployment identifier. Required for deployments that use the responses endpoint
Max Tokensnumber4096Upper limit on generated output. Raise it if responses are cut off
Temperaturenumber(empty)Sampling temperature, 0.01.0. Empty uses the model default
TopPnumber(empty)Nucleus sampling, 0.01.0. Empty uses the model default
Request Timeoutnumber30Seconds to wait for a response. Raise it for long prompts
Live ModebooltrueWhen off, requests are simulated and nothing leaves the installation
Which fields are visible

The credential fields depend on Authentication Mode: switch the mode and the other set appears. This is the conditional-field behavior described in Interfaces and Core Nodes.

Verify it worked

  • With Live Mode off, a message passes through and the log shows the request that would have been sent, with no call leaving the installation.
  • With Live Mode on, the downstream node receives the model's response.
  • Responses that end mid-sentence mean Max Tokens is too low for your prompt.

If it didn't work

Symptom in the logCauseFix
CONFIG_ERROR naming a fieldA field required by the selected Authentication Mode is emptyFill the field named
HTTP_401 in API Key modeThe key is wrong or has been rotatedCopy the key again from the deployment page
HTTP_401 in Entra ID modeThe app registration lacks the Azure OpenAI role on the resourceGrant the role, then restart the node
HTTP_404Model URI points at a deployment that does not exist, or omits the API versionRe-copy the endpoint URL from the deployment page
HTTP_429The deployment's rate or token quota is exhaustedSlow the upstream feed, or raise the quota in Azure
TIMEOUTThe prompt takes longer than Request TimeoutRaise the timeout, or shorten the input
Response is truncatedMax Tokens is too lowRaise Max Tokens
failed to decrypt fieldThe project was imported to a different installationRe-enter API Key or Azure Client Secret here
Nothing is sentLive Mode is offTurn Live Mode on

Next