feat(Azure OpenAI Chat Model Node): Add the deployment dropdown (#39643)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

packages/@n8n/nodes-langchain/nodes/llms/LmChatAzureOpenAi/LmChatAzureOpenAi.node.ts

@@ -41,6 +41,7 @@ export class LmChatAzureOpenAi implements INodeType {
 		icon: 'file:azure.svg',
 		group: ['transform'],
 		version: [1, 1.1],
+		defaultVersion: 1.1,
 		description: 'For advanced usage with an AI chain',
 		defaults: {
 			name: 'Azure AI Foundry Chat Model',
@@ -96,7 +97,10 @@ export class LmChatAzureOpenAi implements INodeType {
 				'authentication',
 				itemIndex,
 			) as AuthenticationType;
-			const modelName = this.getNodeParameter('model', itemIndex) as string;
+			// Version 1 stores a plain string; `extractValue` returns it unchanged.
+			const modelName = this.getNodeParameter('model', itemIndex, '', {
+				extractValue: true,
+			}) as string;
 			const allOptions = this.getNodeParameter('options', itemIndex, {}) as AzureOpenAIOptions;
 			// Held back from the spread below: both clients take it as `modelKwargs`, and spreading the
 			// raw JSON string would put an `extraBody` field on the constructor.

packages/@n8n/nodes-langchain/nodes/llms/LmChatAzureOpenAi/__tests__/LmChatAzureOpenAi.node.test.ts

@@ -1,7 +1,7 @@
 import { AzureChatOpenAI, ChatOpenAI } from '@langchain/openai';
 import { getProxyAgent, makeN8nLlmFailedAttemptHandler } from '@n8n/ai-utilities';
 import { createMockExecuteFunction } from 'n8n-nodes-base/test/nodes/Helpers';
-import type { INode, ISupplyDataFunctions } from 'n8n-workflow';
+import { NodeHelpers, type INode, type ISupplyDataFunctions } from 'n8n-workflow';
 
 import { LmChatAzureOpenAi } from '../LmChatAzureOpenAi.node';
 
@@ -392,4 +392,43 @@ describe('LmChatAzureOpenAi', () => {
 			await expect(new LmChatAzureOpenAi().supplyData.call(ctx, 0)).rejects.toThrow(message);
 		});
 	});
+
+	describe('model parameter', () => {
+		const { description } = new LmChatAzureOpenAi();
+		// `@n8n/ai-utilities` is mocked, so the connection-hint property is undefined here.
+		const modelPropertyAt = (typeVersion: number) =>
+			description.properties.filter(
+				(p) =>
+					p?.name === 'model' && NodeHelpers.displayParameter({}, p, { typeVersion }, description),
+			);
+
+		it('keeps the plain text field on version 1', () => {
+			const shown = modelPropertyAt(1);
+			expect(shown).toHaveLength(1);
+			expect(shown[0].type).toBe('string');
+		});
+
+		it('shows a deployment list backed by searchModels on version 1.1', () => {
+			const shown = modelPropertyAt(1.1);
+			expect(shown).toHaveLength(1);
+			expect(shown[0].type).toBe('resourceLocator');
+			expect(shown[0].modes?.map((m) => m.name)).toEqual(['list', 'id']);
+			// The list needs a Foundry credential, and new credentials default to classic.
+			expect(shown[0].default).toEqual({ mode: 'id', value: '' });
+			expect(shown[0].modes?.[0].typeOptions?.searchListMethod).toBe('searchModels');
+			expect(description.defaultVersion).toBe(1.1);
+		});
+
+		it('reads the deployment name with extractValue, so both version shapes resolve', async () => {
+			const ctx = setupMockContext('azureOpenAiApi', apiKeyCredential);
+
+			await new LmChatAzureOpenAi().supplyData.call(ctx, 0);
+
+			expect(ctx.getNodeParameter).toHaveBeenCalledWith('model', 0, '', { extractValue: true });
+			expect(vi.mocked(AzureChatOpenAI).mock.calls[0][0]).toMatchObject({
+				model: 'gpt-4o',
+				azureOpenAIApiDeploymentName: 'gpt-4o',
+			});
+		});
+	});
 });

packages/@n8n/nodes-langchain/nodes/llms/LmChatAzureOpenAi/methods/searchModels.test.ts

@@ -49,7 +49,7 @@ describe('LmChatAzureOpenAi -> searchModels', () => {
 		vi.clearAllMocks();
 	});
 
-	it('skips discovery for a classic api-key credential (no data-plane deployments API)', async () => {
+	it('rejects a classic api-key credential with a clear error (no data-plane deployments API)', async () => {
 		ctx.getNodeParameter = vi
 			.fn()
 			.mockReturnValueOnce(AuthenticationType.ApiKey)
@@ -59,24 +59,62 @@ describe('LmChatAzureOpenAi -> searchModels', () => {
 			resourceName: 'my-resource',
 		});
 
-		const result = await searchModels.call(ctx);
+		await expect(searchModels.call(ctx)).rejects.toThrow(
+			'Only an Azure AI Foundry credential can list deployments',
+		);
 
 		expect(listAzureOpenAiModelsSpy).not.toHaveBeenCalled();
-		expect(result).toEqual({ results: [] });
 	});
 
-	it('skips discovery for a classic Entra ID credential', async () => {
+	it('rejects a classic Entra ID credential with a clear error', async () => {
 		ctx.getNodeParameter = vi
 			.fn()
 			.mockReturnValueOnce(AuthenticationType.EntraOAuth2)
 			.mockReturnValueOnce('');
 		ctx.getCredentials = vi.fn().mockResolvedValue({ resourceName: 'my-resource' });
 
-		const result = await searchModels.call(ctx);
+		await expect(searchModels.call(ctx)).rejects.toThrow(
+			'Only an Azure AI Foundry credential can list deployments',
+		);
 
 		expect(listAzureOpenAiModelsSpy).not.toHaveBeenCalled();
 		expect(n8nOAuth2TokenCredentialSpy).not.toHaveBeenCalled();
-		expect(result).toEqual({ results: [] });
+	});
+
+	it('asks for the project before it lists deployments', async () => {
+		ctx.getNodeParameter = vi
+			.fn()
+			.mockReturnValueOnce(AuthenticationType.ApiKey)
+			.mockReturnValueOnce('');
+		ctx.getCredentials = vi.fn().mockResolvedValue({
+			apiKey: 'secret-key',
+			endpointType: 'foundry',
+			foundryEndpoint: 'https://my-resource.services.ai.azure.com/openai/v1',
+		});
+
+		await expect(searchModels.call(ctx)).rejects.toThrow(
+			'Enter the Project to list its deployments',
+		);
+
+		expect(listAzureOpenAiModelsSpy).not.toHaveBeenCalled();
+	});
+
+	it('asks for the project before it requests an Entra token', async () => {
+		ctx.getNodeParameter = vi
+			.fn()
+			.mockReturnValueOnce(AuthenticationType.EntraOAuth2)
+			.mockReturnValueOnce('');
+		ctx.getCredentials = vi.fn().mockResolvedValue({
+			endpointType: 'foundry',
+			foundryEndpoint: 'https://my-resource.services.ai.azure.com/openai/v1',
+		});
+
+		await expect(searchModels.call(ctx)).rejects.toThrow(
+			'Enter the Project to list its deployments',
+		);
+
+		expect(getTokenSpy).not.toHaveBeenCalled();
+		expect(listAzureOpenAiModelsSpy).not.toHaveBeenCalled();
 	});
 
 	it('derives the base URL from the Foundry endpoint origin for an api-key credential', async () => {

packages/@n8n/nodes-langchain/nodes/llms/LmChatAzureOpenAi/methods/searchModels.ts

@@ -23,6 +23,22 @@ function resolveFoundryBaseURL(node: INode, foundryEndpoint?: string): string {
 	return new URL(requireFoundryEndpoint(node, foundryEndpoint)).origin;
 }
 
+// The dropdown shows only the message, so it also says how to fix the problem.
+function classicCredentialError(node: INode): NodeOperationError {
+	return new NodeOperationError(
+		node,
+		"Only an Azure AI Foundry credential can list deployments. Select By ID and enter the deployment name, or set the credential's Endpoint Type to Azure AI Foundry.",
+	);
+}
+
+// Without this, the list call reports `Project "" was not found`.
+function missingProjectError(node: INode): NodeOperationError {
+	return new NodeOperationError(
+		node,
+		'Enter the Project to list its deployments. Or select By ID and enter the deployment name.',
+	);
+}
+
 export async function searchModels(
 	this: ILoadOptionsFunctions,
 	filter?: string,
@@ -42,18 +58,21 @@ export async function searchModels(
 			'azureOpenAiApi',
 		);
 		if (credential.endpointType !== 'foundry') {
-			return { results: [] };
+			throw classicCredentialError(this.getNode());
 		}
 		baseURL = resolveFoundryBaseURL(this.getNode(), credential.foundryEndpoint);
+		if (!project) throw missingProjectError(this.getNode());
 		headers = { 'api-key': credential.apiKey };
 	} else {
 		const credential = await this.getCredentials<AzureEntraCognitiveServicesOAuth2ApiCredential>(
 			'azureEntraCognitiveServicesOAuth2Api',
 		);
 		if (credential.endpointType !== 'foundry') {
-			return { results: [] };
+			throw classicCredentialError(this.getNode());
 		}
 		baseURL = resolveFoundryBaseURL(this.getNode(), credential.foundryEndpoint);
+		// Checked before the token request, so an empty Project does not cost a call to Entra.
+		if (!project) throw missingProjectError(this.getNode());
 		// Mints a token for the Foundry audience, which this call needs.
 		const token = await new N8nOAuth2TokenCredential(
 			this.getNode(),

packages/@n8n/nodes-langchain/nodes/llms/LmChatAzureOpenAi/properties.ts

@@ -43,6 +43,11 @@ export const properties: INodeProperties[] = [
 		description: 'The name of the model(deployment) to use (e.g., gpt-4, gpt-35-turbo)',
 		required: true,
 		default: '',
+		displayOptions: {
+			show: {
+				'@version': [1],
+			},
+		},
 	},
 	{
 		displayName: 'Project',
@@ -52,6 +57,42 @@ export const properties: INodeProperties[] = [
 			'The Azure AI Foundry project that owns the deployment. Required for an Azure AI Foundry resource; leave empty for a classic Azure OpenAI resource.',
 		default: '',
 	},
+	{
+		displayName: 'Model (Deployment)',
+		name: 'model',
+		type: 'resourceLocator',
+		default: { mode: 'id', value: '' },
+		required: true,
+		builderHint: {
+			propertyHint:
+				'Set the project parameter before you list deployments. The list needs an Azure AI Foundry credential. With a classic credential, use the id mode and enter the deployment name.',
+		},
+		modes: [
+			{
+				displayName: 'From List',
+				name: 'list',
+				type: 'list',
+				placeholder: 'Select a deployment...',
+				typeOptions: {
+					searchListMethod: 'searchModels',
+					searchable: true,
+				},
+			},
+			{
+				displayName: 'By ID',
+				name: 'id',
+				type: 'string',
+				placeholder: 'e.g. gpt-4o',
+			},
+		],
+		description:
+			'The deployment to use. Choose from the list (Azure AI Foundry credentials only), or enter the deployment name.',
+		displayOptions: {
+			show: {
+				'@version': [{ _cnd: { gte: 1.1 } }],
+			},
+		},
+	},
 	{
 		displayName: 'Use Responses API',
 		name: 'responsesApiEnabled',

packages/core/src/execution-engine/node-execution-context/utils/__tests__/extract-value.test.ts

@@ -1,8 +1,59 @@
-import type { INode, INodeType } from 'n8n-workflow';
+import type { INode, INodeProperties, INodeType } from 'n8n-workflow';
 
 import { extractValue } from '../extract-value';
 
+const node: INode = {
+	id: 'node-id',
+	name: 'Test Node',
+	type: 'test.node',
+	typeVersion: 1,
+	position: [0, 0],
+	parameters: {},
+};
+
+const nodeTypeWith = (property: INodeProperties): INodeType => ({
+	description: {
+		displayName: 'Test Node',
+		name: 'testNode',
+		group: ['transform'],
+		version: 1,
+		description: 'Test node',
+		defaults: { name: 'Test Node' },
+		inputs: [],
+		outputs: [],
+		properties: [property],
+	},
+});
+
 describe('extractValue', () => {
+	it('returns a plain value unchanged when the property has no extractValue', () => {
+		const nodeType = nodeTypeWith({
+			displayName: 'Field',
+			name: 'field',
+			type: 'string',
+			default: '',
+		});
+
+		expect(extractValue('some-value', 'field', node, nodeType)).toBe('some-value');
+	});
+
+	it('returns the inner value of a resource locator whose mode has no extractValue', () => {
+		const nodeType = nodeTypeWith({
+			displayName: 'Field',
+			name: 'field',
+			type: 'resourceLocator',
+			default: { mode: 'list', value: '' },
+			modes: [
+				{ displayName: 'List', name: 'list', type: 'list' },
+				{ displayName: 'ID', name: 'id', type: 'string' },
+			],
+		});
+
+		expect(
+			extractValue({ __rl: true, mode: 'list', value: 'some-value' }, 'field', node, nodeType),
+		).toBe('some-value');
+	});
+
 	it('extracts resource locator values with regex metadata', () => {
 		const node: INode = {
 			id: 'node-id',